Fixes iOS hand landmarker connections

PiperOrigin-RevId: 565442497

.github/ISSUE_TEMPLATE/00-task-issue-template.yaml

@@ -40,18 +40,16 @@ body:
       label: Programming Language and version (e.g. C++, Python, Java)
     validations:
       required: true
-  - type: textarea
+  - type: input
     id: current_model
     attributes:
       label: Describe the actual behavior
-      render: shell
     validations:
       required: true
-  - type: textarea
+  - type: input
     id: expected_model
     attributes:
       label: Describe the expected behaviour
-      render: shell
     validations:
       required: true
   - type: textarea

.github/ISSUE_TEMPLATE/11-model-maker-issue-template.yaml

@@ -41,18 +41,16 @@ body:
       label: Task name (e.g. Image classification, Gesture recognition etc.)
     validations:
       required: true
-  - type: textarea
+  - type: input
     id: current_model
     attributes:
       label: Describe the actual behavior
-      render: shell
     validations:
       required: true
-  - type: textarea
+  - type: input
     id: expected_model
     attributes:
       label: Describe the expected behaviour
-      render: shell
     validations:
       required: true
   - type: textarea

.github/ISSUE_TEMPLATE/12-studio-issue-template.yaml

@@ -31,18 +31,16 @@ body:
       label: URL that shows the problem
     validations:
       required: false
-  - type: textarea
+  - type: input
     id: current_model
     attributes:
       label: Describe the actual behavior
-      render: shell
     validations:
       required: false
-  - type: textarea
+  - type: input
     id: expected_model
     attributes:
       label: Describe the expected behaviour
-      render: shell
     validations:
       required: false
   - type: textarea

.github/ISSUE_TEMPLATE/14-feature-request-issue-template.yaml

@@ -28,37 +28,33 @@ body:
         - 'No'
     validations:
       required: false
-  - type: textarea
+  - type: input
     id: behaviour
     attributes:
       label: Describe the feature and the current behaviour/state
-      render: shell
     validations:
       required: true
-  - type: textarea
+  - type: input
     id: api_change
     attributes:
       label: Will this change the current API? How?
-      render: shell
     validations:
       required: false
-  - type: textarea
+  - type: input
     id: benifit
     attributes:
       label: Who will benefit with this feature?
     validations:
        required: false
-  - type: textarea
+  - type: input
     id: use_case
     attributes:
       label: Please specify the use cases for this feature
-      render: shell
     validations:
        required: true
-  - type: textarea
+  - type: input
     id: info_other
     attributes:
       label: Any Other info
-      render: shell
     validations:
       required: false

.github/ISSUE_TEMPLATE/15-build-install-issue-template.yaml

@@ -87,14 +87,13 @@ body:
       placeholder:
     validations:
       required: false
-  - type: textarea
+  - type: input
     id: what-happened
     attributes:
       label: Describe the problem
       description: Provide the exact sequence of commands / steps that you executed before running into the [problem](https://google.github.io/mediapipe/getting_started/getting_started.html)
       placeholder: Tell us what you see!
       value: "A bug happened!"
-      render: shell
     validations:
       required: true
   - type: textarea

.github/ISSUE_TEMPLATE/16-bug-issue-template.yaml

@@ -80,18 +80,16 @@ body:
       label: Xcode & Tulsi version (if issue is related to building for iOS)
     validations:
       required: false
-  - type: textarea
+  - type: input
     id: current_model
     attributes:
       label: Describe the actual behavior
-      render: shell
     validations:
       required: true
-  - type: textarea
+  - type: input
     id: expected_model
     attributes:
       label: Describe the expected behaviour
-      render: shell
     validations:
       required: true
   - type: textarea

.github/ISSUE_TEMPLATE/18-solution-legacy-issue-template.yaml

@@ -48,18 +48,16 @@ body:
       placeholder: e.g. C++, Python, Java
     validations:
       required: false
-  - type: textarea
+  - type: input
     id: current_model
     attributes:
       label: Describe the actual behavior
-      render: shell
     validations:
       required: false
-  - type: textarea
+  - type: input
     id: expected_model
     attributes:
       label: Describe the expected behaviour
-      render: shell
     validations:
       required: false
   - type: textarea

WORKSPACE

@@ -73,12 +73,9 @@ http_archive(
 http_archive(
     name = "zlib",
     build_file = "@//third_party:zlib.BUILD",
-    sha256 = "c3e5e9fdd5004dcb542feda5ee4f0ff0744628baf8ed2dd5d66f8ca1197cb1a1",
+    sha256 = "b3a24de97a8fdbc835b9833169501030b8977031bcb54b3b3ac13740f846ab30",
-    strip_prefix = "zlib-1.2.11",
+    strip_prefix = "zlib-1.2.13",
-    urls = [
+    url = "http://zlib.net/fossils/zlib-1.2.13.tar.gz",
-        "http://mirror.bazel.build/zlib.net/fossils/zlib-1.2.11.tar.gz",
-        "http://zlib.net/fossils/zlib-1.2.11.tar.gz",  # 2017-01-15
-    ],
     patches = [
         "@//third_party:zlib.diff",
     ],
@@ -157,22 +154,22 @@ http_archive(
 # 2020-08-21
 http_archive(
     name = "com_github_glog_glog",
-    strip_prefix = "glog-0a2e5931bd5ff22fd3bf8999eb8ce776f159cda6",
+    strip_prefix = "glog-3a0d4d22c5ae0b9a2216988411cfa6bf860cc372",
-    sha256 = "58c9b3b6aaa4dd8b836c0fd8f65d0f941441fb95e27212c5eeb9979cfd3592ab",
+    sha256 = "170d08f80210b82d95563f4723a15095eff1aad1863000e8eeb569c96a98fefb",
     urls = [
-        "https://github.com/google/glog/archive/0a2e5931bd5ff22fd3bf8999eb8ce776f159cda6.zip",
+        "https://github.com/google/glog/archive/3a0d4d22c5ae0b9a2216988411cfa6bf860cc372.zip",
     ],
 )
 http_archive(
     name = "com_github_glog_glog_no_gflags",
-    strip_prefix = "glog-0a2e5931bd5ff22fd3bf8999eb8ce776f159cda6",
+    strip_prefix = "glog-3a0d4d22c5ae0b9a2216988411cfa6bf860cc372",
-    sha256 = "58c9b3b6aaa4dd8b836c0fd8f65d0f941441fb95e27212c5eeb9979cfd3592ab",
+    sha256 = "170d08f80210b82d95563f4723a15095eff1aad1863000e8eeb569c96a98fefb",
     build_file = "@//third_party:glog_no_gflags.BUILD",
     urls = [
-        "https://github.com/google/glog/archive/0a2e5931bd5ff22fd3bf8999eb8ce776f159cda6.zip",
+        "https://github.com/google/glog/archive/3a0d4d22c5ae0b9a2216988411cfa6bf860cc372.zip",
     ],
     patches = [
-        "@//third_party:com_github_glog_glog_9779e5ea6ef59562b030248947f787d1256132ae.diff",
+        "@//third_party:com_github_glog_glog.diff",
     ],
     patch_args = [
         "-p1",
@@ -485,10 +482,10 @@ http_archive(
 )
 
 # TensorFlow repo should always go after the other external dependencies.
-# TF on 2023-06-13.
+# TF on 2023-07-26.
-_TENSORFLOW_GIT_COMMIT = "491681a5620e41bf079a582ac39c585cc86878b9"
+_TENSORFLOW_GIT_COMMIT = "e92261fd4cec0b726692081c4d2966b75abf31dd"
 # curl -L https://github.com/tensorflow/tensorflow/archive/<TENSORFLOW_GIT_COMMIT>.tar.gz | shasum -a 256
-_TENSORFLOW_SHA256 = "9f76389af7a2835e68413322c1eaabfadc912f02a76d71dc16be507f9ca3d3ac"
+_TENSORFLOW_SHA256 = "478a229bd4ec70a5b568ac23b5ea013d9fca46a47d6c43e30365a0412b9febf4"
 http_archive(
     name = "org_tensorflow",
     urls = [
@@ -496,6 +493,7 @@ http_archive(
     ],
     patches = [
         "@//third_party:org_tensorflow_compatibility_fixes.diff",
+        "@//third_party:org_tensorflow_system_python.diff",
         # Diff is generated with a script, don't update it manually.
         "@//third_party:org_tensorflow_custom_ops.diff",
     ],

docs/getting_started/hello_world_cpp.md

@@ -50,7 +50,7 @@ as the primary developer documentation site for MediaPipe as of April 3, 2023.*
 3.  The [`hello world`] example uses a simple MediaPipe graph in the
     `PrintHelloWorld()` function, defined in a [`CalculatorGraphConfig`] proto.
 
-    ```C++
+    ```c++
     absl::Status PrintHelloWorld() {
       // Configures a simple graph, which concatenates 2 PassThroughCalculators.
       CalculatorGraphConfig config = ParseTextProtoOrDie<CalculatorGraphConfig>(R"(
@@ -126,7 +126,7 @@ as the primary developer documentation site for MediaPipe as of April 3, 2023.*
     ```c++
     mediapipe::Packet packet;
     while (poller.Next(&packet)) {
-      LOG(INFO) << packet.Get<string>();
+      ABSL_LOG(INFO) << packet.Get<string>();
     }
     ```
 

docs/getting_started/hello_world_ios.md

@@ -138,7 +138,7 @@ Create a `BUILD` file in the `$APPLICATION_PATH` and add the following build
 rules:
 
 ```
-MIN_IOS_VERSION = "11.0"
+MIN_IOS_VERSION = "12.0"
 
 load(
     "@build_bazel_rules_apple//apple:ios.bzl",

mediapipe/BUILD

@@ -14,81 +14,155 @@
 
 licenses(["notice"])  # Apache 2.0
 
-# Note: yes, these need to use "//external:android/crosstool", not
+load("@mediapipe//mediapipe:platforms.bzl", "config_setting_and_platform")
-# @androidndk//:default_crosstool.
 
+# Generic Android
 config_setting(
     name = "android",
-    values = {"crosstool_top": "//external:android/crosstool"},
+    constraint_values = [
+        "@platforms//os:android",
+    ],
     visibility = ["//visibility:public"],
 )
 
-config_setting(
+# Android x86 32-bit.
+config_setting_and_platform(
     name = "android_x86",
-    values = {
+    constraint_values = [
-        "crosstool_top": "//external:android/crosstool",
+        "@platforms//os:android",
-        "cpu": "x86",
+        "@platforms//cpu:x86_32",
-    },
+    ],
     visibility = ["//visibility:public"],
 )
 
-config_setting(
+# Android x86 64-bit.
+config_setting_and_platform(
     name = "android_x86_64",
-    values = {
+    constraint_values = [
-        "crosstool_top": "//external:android/crosstool",
+        "@platforms//os:android",
-        "cpu": "x86_64",
+        "@platforms//cpu:x86_64",
-    },
+    ],
     visibility = ["//visibility:public"],
 )
 
-config_setting(
+# Android ARMv7.
-    name = "android_armeabi",
+config_setting_and_platform(
-    values = {
-        "crosstool_top": "//external:android/crosstool",
-        "cpu": "armeabi",
-    },
-    visibility = ["//visibility:public"],
-)
-
-config_setting(
     name = "android_arm",
-    values = {
+    constraint_values = [
-        "crosstool_top": "//external:android/crosstool",
+        "@platforms//os:android",
-        "cpu": "armeabi-v7a",
+        "@platforms//cpu:armv7",
-    },
+    ],
     visibility = ["//visibility:public"],
 )
 
-config_setting(
+# Android ARM64.
+config_setting_and_platform(
     name = "android_arm64",
-    values = {
+    constraint_values = [
-        "crosstool_top": "//external:android/crosstool",
+        "@platforms//os:android",
-        "cpu": "arm64-v8a",
+        "@platforms//cpu:arm64",
-    },
+    ],
     visibility = ["//visibility:public"],
 )
 
-# Note: this cannot just match "apple_platform_type": "macos" because that option
+# Generic MacOS.
-# defaults to "macos" even when building on Linux!
+config_setting(
-alias(
     name = "macos",
-    actual = select({
+    constraint_values = [
-        ":macos_i386": ":macos_i386",
+        "@platforms//os:macos",
-        ":macos_x86_64": ":macos_x86_64",
+    ],
-        ":macos_arm64": ":macos_arm64",
-        "//conditions:default": ":macos_i386",  # Arbitrarily chosen from above.
-    }),
     visibility = ["//visibility:public"],
 )
 
-# Note: this also matches on crosstool_top so that it does not produce ambiguous
+# MacOS x86 64-bit.
-# selectors when used together with "android".
+config_setting_and_platform(
+    name = "macos_x86_64",
+    constraint_values = [
+        "@platforms//os:macos",
+        "@platforms//cpu:x86_64",
+    ],
+    visibility = ["//visibility:public"],
+)
+
+# MacOS ARM64.
+config_setting_and_platform(
+    name = "macos_arm64",
+    constraint_values = [
+        "@platforms//os:macos",
+        "@platforms//cpu:arm64",
+    ],
+    visibility = ["//visibility:public"],
+)
+
+# Generic iOS.
 config_setting(
     name = "ios",
-    values = {
+    constraint_values = [
-        "crosstool_top": "@bazel_tools//tools/cpp:toolchain",
+        "@platforms//os:ios",
-        "apple_platform_type": "ios",
+    ],
-    },
+    visibility = ["//visibility:public"],
+)
+
+# iOS device ARM32.
+config_setting_and_platform(
+    name = "ios_armv7",
+    constraint_values = [
+        "@platforms//os:ios",
+        "@platforms//cpu:arm",
+    ],
+    visibility = ["//visibility:public"],
+)
+
+# iOS device ARM64.
+config_setting_and_platform(
+    name = "ios_arm64",
+    constraint_values = [
+        "@platforms//os:ios",
+        "@platforms//cpu:arm64",
+    ],
+    visibility = ["//visibility:public"],
+)
+
+# iOS device ARM64E.
+config_setting_and_platform(
+    name = "ios_arm64e",
+    constraint_values = [
+        "@platforms//os:ios",
+        "@platforms//cpu:arm64e",
+    ],
+    visibility = ["//visibility:public"],
+)
+
+# iOS simulator x86 32-bit.
+config_setting_and_platform(
+    name = "ios_i386",
+    constraint_values = [
+        "@platforms//os:ios",
+        "@platforms//cpu:x86_32",
+        "@build_bazel_apple_support//constraints:simulator",
+    ],
+    visibility = ["//visibility:public"],
+)
+
+# iOS simulator x86 64-bit.
+config_setting_and_platform(
+    name = "ios_x86_64",
+    constraint_values = [
+        "@platforms//os:ios",
+        "@platforms//cpu:x86_64",
+        "@build_bazel_apple_support//constraints:simulator",
+    ],
+    visibility = ["//visibility:public"],
+)
+
+# iOS simulator ARM64.
+config_setting_and_platform(
+    name = "ios_sim_arm64",
+    constraint_values = [
+        "@platforms//os:ios",
+        "@platforms//cpu:arm64",
+        "@build_bazel_apple_support//constraints:simulator",
+    ],
     visibility = ["//visibility:public"],
 )
 
@@ -102,52 +176,24 @@ alias(
     visibility = ["//visibility:public"],
 )
 
-config_setting(
+# Windows 64-bit.
-    name = "macos_i386",
+config_setting_and_platform(
-    values = {
-        "apple_platform_type": "macos",
-        "cpu": "darwin",
-    },
-    visibility = ["//visibility:public"],
-)
-
-config_setting(
-    name = "macos_x86_64",
-    values = {
-        "apple_platform_type": "macos",
-        "cpu": "darwin_x86_64",
-    },
-    visibility = ["//visibility:public"],
-)
-
-config_setting(
-    name = "macos_arm64",
-    values = {
-        "apple_platform_type": "macos",
-        "cpu": "darwin_arm64",
-    },
-    visibility = ["//visibility:public"],
-)
-
-[
-    config_setting(
-        name = arch,
-        values = {"cpu": arch},
-        visibility = ["//visibility:public"],
-    )
-    for arch in [
-        "ios_i386",
-        "ios_x86_64",
-        "ios_armv7",
-        "ios_arm64",
-        "ios_arm64e",
-        "ios_sim_arm64",
-    ]
-]
-
-config_setting(
     name = "windows",
-    values = {"cpu": "x64_windows"},
+    constraint_values = [
+        "@platforms//os:windows",
+        "@platforms//cpu:x86_64",
+    ],
+    visibility = ["//visibility:public"],
+)
+
+# Linux 64-bit.
+config_setting_and_platform(
+    name = "linux",
+    constraint_values = [
+        "@platforms//os:linux",
+        "@platforms//cpu:x86_64",
+    ],
+    visibility = ["//visibility:public"],
 )
 
 exports_files(

mediapipe/calculators/audio/BUILD

@@ -12,6 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+# Placeholder: load py_proto_library
 load("//mediapipe/framework/port:build_config.bzl", "mediapipe_cc_proto_library")
 
 licenses(["notice"])
@@ -145,6 +146,7 @@ cc_library(
         "//mediapipe/framework/port:logging",
         "//mediapipe/framework/port:status",
         "//mediapipe/util:time_series_util",
+        "@com_google_absl//absl/log:absl_check",
         "@com_google_absl//absl/strings",
         "@com_google_audio_tools//audio/dsp/mfcc",
         "@eigen_archive//:eigen3",
@@ -163,8 +165,9 @@ cc_library(
         "//mediapipe/framework/formats:matrix",
         "//mediapipe/framework/formats:time_series_header_cc_proto",
         "//mediapipe/framework/port:integral_types",
-        "//mediapipe/framework/port:logging",
         "//mediapipe/util:time_series_util",
+        "@com_google_absl//absl/log:absl_check",
+        "@com_google_absl//absl/log:absl_log",
         "@com_google_absl//absl/strings",
         "@com_google_audio_tools//audio/dsp:resampler",
         "@com_google_audio_tools//audio/dsp:resampler_q",
@@ -185,6 +188,7 @@ cc_library(
         "//mediapipe/framework/port:core_proto",
         "//mediapipe/framework/port:status",
         "//mediapipe/util:time_series_util",
+        "@com_google_absl//absl/log:absl_check",
     ],
     alwayslink = 1,
 )
@@ -224,6 +228,7 @@ cc_library(
         "//mediapipe/framework/formats:time_series_header_cc_proto",
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/util:time_series_util",
+        "@com_google_absl//absl/log:absl_check",
         "@com_google_audio_tools//audio/dsp:window_functions",
         "@eigen_archive//:eigen3",
     ],
@@ -294,6 +299,7 @@ cc_test(
         "//mediapipe/framework/port:integral_types",
         "//mediapipe/framework/port:status",
         "//mediapipe/util:time_series_test_util",
+        "@com_google_absl//absl/log:absl_log",
         "@com_google_audio_tools//audio/dsp:number_util",
         "@eigen_archive//:eigen3",
     ],
@@ -327,6 +333,7 @@ cc_binary(
         "//mediapipe/framework:packet",
         "//mediapipe/framework/formats:matrix",
         "//mediapipe/framework/formats:time_series_header_cc_proto",
+        "@com_google_absl//absl/log:absl_check",
         "@com_google_benchmark//:benchmark",
     ],
 )
@@ -345,6 +352,7 @@ cc_test(
         "//mediapipe/framework/port:integral_types",
         "//mediapipe/framework/port:status",
         "//mediapipe/util:time_series_test_util",
+        "@com_google_absl//absl/log:absl_log",
         "@com_google_audio_tools//audio/dsp:window_functions",
         "@eigen_archive//:eigen3",
     ],

mediapipe/calculators/audio/mfcc_mel_calculators.cc

@@ -23,6 +23,7 @@
 #include <vector>
 
 #include "Eigen/Core"
+#include "absl/log/absl_check.h"
 #include "absl/strings/str_cat.h"
 #include "absl/strings/string_view.h"
 #include "absl/strings/substitute.h"
@@ -138,7 +139,7 @@ absl::Status FramewiseTransformCalculatorBase::Process(CalculatorContext* cc) {
     TransformFrame(input_frame, &output_frame);
 
     // Copy output from vector<float> to Eigen::Vector.
-    CHECK_EQ(output_frame.size(), num_output_channels_);
+    ABSL_CHECK_EQ(output_frame.size(), num_output_channels_);
     Eigen::Map<const Eigen::MatrixXd> output_frame_map(&output_frame[0],
                                                        output_frame.size(), 1);
     output->col(frame) = output_frame_map.cast<float>();

mediapipe/calculators/audio/rational_factor_resample_calculator.cc

@@ -16,6 +16,8 @@
 
 #include "mediapipe/calculators/audio/rational_factor_resample_calculator.h"
 
+#include "absl/log/absl_check.h"
+#include "absl/log/absl_log.h"
 #include "audio/dsp/resampler_q.h"
 
 using audio_dsp::Resampler;
@@ -45,9 +47,9 @@ void CopyVectorToChannel(const std::vector<float>& vec, Matrix* matrix,
   if (matrix->cols() == 0) {
     matrix->resize(matrix->rows(), vec.size());
   } else {
-    CHECK_EQ(vec.size(), matrix->cols());
+    ABSL_CHECK_EQ(vec.size(), matrix->cols());
   }
-  CHECK_LT(channel, matrix->rows());
+  ABSL_CHECK_LT(channel, matrix->rows());
   matrix->row(channel) =
       Eigen::Map<const Eigen::ArrayXf>(vec.data(), vec.size());
 }
@@ -77,7 +79,7 @@ absl::Status RationalFactorResampleCalculator::Open(CalculatorContext* cc) {
       r = ResamplerFromOptions(source_sample_rate_, target_sample_rate_,
                                resample_options);
       if (!r) {
-        LOG(ERROR) << "Failed to initialize resampler.";
+        ABSL_LOG(ERROR) << "Failed to initialize resampler.";
         return absl::UnknownError("Failed to initialize resampler.");
       }
     }

mediapipe/calculators/audio/rational_factor_resample_calculator.h

@@ -27,7 +27,6 @@
 #include "mediapipe/framework/formats/matrix.h"
 #include "mediapipe/framework/formats/time_series_header.pb.h"
 #include "mediapipe/framework/port/integral_types.h"
-#include "mediapipe/framework/port/logging.h"
 #include "mediapipe/util/time_series_util.h"
 
 namespace mediapipe {

mediapipe/calculators/audio/spectrogram_calculator_test.cc

@@ -22,6 +22,7 @@
 #include <vector>
 
 #include "Eigen/Core"
+#include "absl/log/absl_log.h"
 #include "audio/dsp/number_util.h"
 #include "mediapipe/calculators/audio/spectrogram_calculator.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
@@ -882,11 +883,11 @@ void BM_ProcessDC(benchmark::State& state) {
 
   const CalculatorRunner::StreamContents& output = runner.Outputs().Index(0);
   const Matrix& output_matrix = output.packets[0].Get<Matrix>();
-  LOG(INFO) << "Output matrix=" << output_matrix.rows() << "x"
+  ABSL_LOG(INFO) << "Output matrix=" << output_matrix.rows() << "x"
-            << output_matrix.cols();
+                 << output_matrix.cols();
-  LOG(INFO) << "First values=" << output_matrix(0, 0) << ", "
+  ABSL_LOG(INFO) << "First values=" << output_matrix(0, 0) << ", "
-            << output_matrix(1, 0) << ", " << output_matrix(2, 0) << ", "
+                 << output_matrix(1, 0) << ", " << output_matrix(2, 0) << ", "
-            << output_matrix(3, 0);
+                 << output_matrix(3, 0);
 }
 
 BENCHMARK(BM_ProcessDC);

mediapipe/calculators/audio/stabilized_log_calculator.cc

@@ -18,6 +18,7 @@
 #include <memory>
 #include <string>
 
+#include "absl/log/absl_check.h"
 #include "mediapipe/calculators/audio/stabilized_log_calculator.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/formats/matrix.h"
@@ -59,7 +60,7 @@ class StabilizedLogCalculator : public CalculatorBase {
     output_scale_ = stabilized_log_calculator_options.output_scale();
     check_nonnegativity_ =
         stabilized_log_calculator_options.check_nonnegativity();
-    CHECK_GE(stabilizer_, 0.0)
+    ABSL_CHECK_GE(stabilizer_, 0.0)
         << "stabilizer must be >= 0.0, received a value of " << stabilizer_;
 
     // If the input packets have a header, propagate the header to the output.

mediapipe/calculators/audio/time_series_framer_calculator.cc

@@ -18,6 +18,7 @@
 #include <vector>
 
 #include "Eigen/Core"
+#include "absl/log/absl_check.h"
 #include "audio/dsp/window_functions.h"
 #include "mediapipe/calculators/audio/time_series_framer_calculator.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
@@ -104,7 +105,7 @@ class TimeSeriesFramerCalculator : public CalculatorBase {
     // All numbers are in input samples.
     const int64_t current_output_frame_start = static_cast<int64_t>(
         round(cumulative_output_frames_ * average_frame_step_samples_));
-    CHECK_EQ(current_output_frame_start, cumulative_completed_samples_);
+    ABSL_CHECK_EQ(current_output_frame_start, cumulative_completed_samples_);
     const int64_t next_output_frame_start = static_cast<int64_t>(
         round((cumulative_output_frames_ + 1) * average_frame_step_samples_));
     return next_output_frame_start - current_output_frame_start;

mediapipe/calculators/audio/time_series_framer_calculator_benchmark.cc

@@ -17,6 +17,7 @@
 #include <random>
 #include <vector>
 
+#include "absl/log/absl_check.h"
 #include "benchmark/benchmark.h"
 #include "mediapipe/calculators/audio/time_series_framer_calculator.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
@@ -70,7 +71,7 @@ void BM_TimeSeriesFramerCalculator(benchmark::State& state) {
     }
     // Initialize graph.
     mediapipe::CalculatorGraph graph;
-    CHECK_OK(graph.Initialize(config));
+    ABSL_CHECK_OK(graph.Initialize(config));
     // Prepare input header.
     auto header = std::make_unique<mediapipe::TimeSeriesHeader>();
     header->set_sample_rate(kSampleRate);
@@ -78,13 +79,13 @@ void BM_TimeSeriesFramerCalculator(benchmark::State& state) {
 
     state.ResumeTiming();  // Resume benchmark timing.
 
-    CHECK_OK(graph.StartRun({}, {{"input", Adopt(header.release())}}));
+    ABSL_CHECK_OK(graph.StartRun({}, {{"input", Adopt(header.release())}}));
     for (auto& packet : input_packets) {
-      CHECK_OK(graph.AddPacketToInputStream("input", packet));
+      ABSL_CHECK_OK(graph.AddPacketToInputStream("input", packet));
     }
-    CHECK(!graph.HasError());
+    ABSL_CHECK(!graph.HasError());
-    CHECK_OK(graph.CloseAllInputStreams());
+    ABSL_CHECK_OK(graph.CloseAllInputStreams());
-    CHECK_OK(graph.WaitUntilIdle());
+    ABSL_CHECK_OK(graph.WaitUntilIdle());
   }
 }
 BENCHMARK(BM_TimeSeriesFramerCalculator);

mediapipe/calculators/audio/time_series_framer_calculator_test.cc

@@ -19,6 +19,7 @@
 #include <vector>
 
 #include "Eigen/Core"
+#include "absl/log/absl_log.h"
 #include "audio/dsp/window_functions.h"
 #include "mediapipe/calculators/audio/time_series_framer_calculator.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
@@ -186,11 +187,12 @@ class TimeSeriesFramerCalculatorTest
     const int num_unique_output_samples =
         round((output().packets.size() - 1) * frame_step_samples) +
         frame_duration_samples;
-    LOG(INFO) << "packets.size()=" << output().packets.size()
+    ABSL_LOG(INFO) << "packets.size()=" << output().packets.size()
-              << " frame_duration_samples=" << frame_duration_samples
+                   << " frame_duration_samples=" << frame_duration_samples
-              << " frame_step_samples=" << frame_step_samples
+                   << " frame_step_samples=" << frame_step_samples
-              << " num_input_samples_=" << num_input_samples_
+                   << " num_input_samples_=" << num_input_samples_
-              << " num_unique_output_samples=" << num_unique_output_samples;
+                   << " num_unique_output_samples="
+                   << num_unique_output_samples;
     const int num_padding_samples =
         num_unique_output_samples - num_input_samples_;
     if (options_.pad_final_packet()) {

mediapipe/calculators/core/BUILD

@@ -381,17 +381,6 @@ cc_library(
     alwayslink = 1,
 )
 
-cc_library(
-    name = "clip_detection_vector_size_calculator",
-    srcs = ["clip_detection_vector_size_calculator.cc"],
-    deps = [
-        ":clip_vector_size_calculator",
-        "//mediapipe/framework:calculator_framework",
-        "//mediapipe/framework/formats:detection_cc_proto",
-    ],
-    alwayslink = 1,
-)
-
 cc_test(
     name = "clip_vector_size_calculator_test",
     srcs = ["clip_vector_size_calculator_test.cc"],
@@ -593,6 +582,7 @@ cc_library(
         "//mediapipe/framework/port:logging",
         "//mediapipe/framework/port:status",
         "//mediapipe/framework/tool:options_util",
+        "@com_google_absl//absl/log:absl_check",
     ],
     alwayslink = 1,
 )
@@ -608,6 +598,7 @@ cc_test(
         "//mediapipe/framework/formats:video_stream_header",
         "//mediapipe/framework/port:gtest_main",
         "//mediapipe/framework/port:integral_types",
+        "@com_google_absl//absl/log:absl_check",
         "@com_google_absl//absl/strings",
     ],
 )
@@ -640,6 +631,7 @@ cc_library(
         "//mediapipe/framework:calculator_framework",
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/framework/port:status",
+        "@com_google_absl//absl/log:absl_log",
     ],
     alwayslink = 1,
 )
@@ -787,10 +779,11 @@ cc_library(
         "//mediapipe/framework/deps:random",
         "//mediapipe/framework/formats:video_stream_header",
         "//mediapipe/framework/port:integral_types",
-        "//mediapipe/framework/port:logging",
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/framework/port:status",
         "//mediapipe/framework/tool:options_util",
+        "@com_google_absl//absl/log:absl_check",
+        "@com_google_absl//absl/log:absl_log",
         "@com_google_absl//absl/strings",
     ],
     alwayslink = 1,
@@ -846,6 +839,7 @@ cc_test(
         "//mediapipe/framework/port:gtest_main",
         "//mediapipe/framework/port:logging",
         "//mediapipe/framework/tool:validate_type",
+        "@com_google_absl//absl/log:absl_check",
         "@eigen_archive//:eigen3",
     ],
 )
@@ -1033,6 +1027,7 @@ cc_library(
         "//mediapipe/framework:calculator_framework",
         "//mediapipe/framework/api2:node",
         "//mediapipe/framework/port:status",
+        "@com_google_absl//absl/log:absl_log",
     ],
     alwayslink = 1,
 )
@@ -1071,6 +1066,7 @@ cc_test(
         "//mediapipe/framework:calculator_runner",
         "//mediapipe/framework/port:gtest_main",
         "//mediapipe/framework/port:parse_text_proto",
+        "@com_google_absl//absl/log:absl_log",
     ],
 )
 
@@ -1117,6 +1113,7 @@ cc_library(
         "//mediapipe/framework/api2:node",
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/framework/port:status",
+        "@com_google_absl//absl/log:absl_log",
     ],
     alwayslink = 1,
 )

mediapipe/calculators/core/end_loop_calculator.cc

@@ -14,6 +14,8 @@
 
 #include "mediapipe/calculators/core/end_loop_calculator.h"
 
+#include <array>
+#include <utility>
 #include <vector>
 
 #include "mediapipe/framework/formats/classification.pb.h"
@@ -84,4 +86,8 @@ typedef EndLoopCalculator<std::vector<std::array<float, 16>>>
     EndLoopAffineMatrixCalculator;
 REGISTER_CALCULATOR(EndLoopAffineMatrixCalculator);
 
+typedef EndLoopCalculator<std::vector<std::pair<int, int>>>
+    EndLoopImageSizeCalculator;
+REGISTER_CALCULATOR(EndLoopImageSizeCalculator);
+
 }  // namespace mediapipe

mediapipe/calculators/core/gate_calculator_test.cc

@@ -12,6 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+#include "absl/log/absl_log.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/calculator_runner.h"
 #include "mediapipe/framework/port/gtest.h"
@@ -356,18 +357,18 @@ TEST_F(GateCalculatorTest, AllowWithStateChangeNoDataStreams) {
   RunTimeStepWithoutDataStream(kTimestampValue2, "ALLOW", true);
   constexpr int64_t kTimestampValue3 = 45;
   RunTimeStepWithoutDataStream(kTimestampValue3, "ALLOW", false);
-  LOG(INFO) << "a";
+  ABSL_LOG(INFO) << "a";
   const std::vector<Packet>& output =
       runner()->Outputs().Get("STATE_CHANGE", 0).packets;
-  LOG(INFO) << "s";
+  ABSL_LOG(INFO) << "s";
   ASSERT_EQ(2, output.size());
-  LOG(INFO) << "d";
+  ABSL_LOG(INFO) << "d";
   EXPECT_EQ(kTimestampValue1, output[0].Timestamp().Value());
   EXPECT_EQ(kTimestampValue3, output[1].Timestamp().Value());
-  LOG(INFO) << "f";
+  ABSL_LOG(INFO) << "f";
   EXPECT_EQ(true, output[0].Get<bool>());   // Allow.
   EXPECT_EQ(false, output[1].Get<bool>());  // Disallow.
-  LOG(INFO) << "g";
+  ABSL_LOG(INFO) << "g";
 }
 
 TEST_F(GateCalculatorTest, DisallowWithStateChange) {

mediapipe/calculators/core/immediate_mux_calculator.cc

@@ -12,6 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+#include "absl/log/absl_log.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/port/ret_check.h"
 #include "mediapipe/framework/port/status.h"
@@ -78,7 +79,7 @@ absl::Status ImmediateMuxCalculator::Process(CalculatorContext* cc) {
       if (packet.Timestamp() >= cc->Outputs().Index(0).NextTimestampBound()) {
         cc->Outputs().Index(0).AddPacket(packet);
       } else {
-        LOG_FIRST_N(WARNING, 5)
+        ABSL_LOG_FIRST_N(WARNING, 5)
             << "Dropping a packet with timestamp " << packet.Timestamp();
       }
       if (cc->Outputs().NumEntries() >= 2) {

mediapipe/calculators/core/matrix_multiply_calculator_test.cc

@@ -16,6 +16,7 @@
 #include <vector>
 
 #include "Eigen/Core"
+#include "absl/log/absl_check.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/calculator_runner.h"
 #include "mediapipe/framework/formats/matrix.h"
@@ -209,7 +210,7 @@ TEST(MatrixMultiplyCalculatorTest, Multiply) {
   MatrixFromTextProto(kSamplesText, &samples);
   Matrix expected;
   MatrixFromTextProto(kExpectedText, &expected);
-  CHECK_EQ(samples.cols(), expected.cols());
+  ABSL_CHECK_EQ(samples.cols(), expected.cols());
 
   for (int i = 0; i < samples.cols(); ++i) {
     // Take a column from samples and produce a packet with just that

mediapipe/calculators/core/merge_calculator.cc

@@ -12,6 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+#include "absl/log/absl_log.h"
 #include "mediapipe/framework/api2/node.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/port/ret_check.h"
@@ -53,7 +54,7 @@ class MergeCalculator : public Node {
   static absl::Status UpdateContract(CalculatorContract* cc) {
     RET_CHECK_GT(kIn(cc).Count(), 0) << "Needs at least one input stream";
     if (kIn(cc).Count() == 1) {
-      LOG(WARNING)
+      ABSL_LOG(WARNING)
           << "MergeCalculator expects multiple input streams to merge but is "
              "receiving only one. Make sure the calculator is configured "
              "correctly or consider removing this calculator to reduce "
@@ -72,8 +73,8 @@ class MergeCalculator : public Node {
       }
     }
 
-    LOG(WARNING) << "Empty input packets at timestamp "
+    ABSL_LOG(WARNING) << "Empty input packets at timestamp "
-                 << cc->InputTimestamp().Value();
+                      << cc->InputTimestamp().Value();
 
     return absl::OkStatus();
   }

mediapipe/calculators/core/packet_resampler_calculator.cc

@@ -16,6 +16,9 @@
 
 #include <memory>
 
+#include "absl/log/absl_check.h"
+#include "absl/log/absl_log.h"
+
 namespace {
 // Reflect an integer against the lower and upper bound of an interval.
 int64_t ReflectBetween(int64_t ts, int64_t ts_min, int64_t ts_max) {
@@ -177,7 +180,7 @@ PacketResamplerCalculator::GetSamplingStrategy(
     const PacketResamplerCalculatorOptions& options) {
   if (options.reproducible_sampling()) {
     if (!options.jitter_with_reflection()) {
-      LOG(WARNING)
+      ABSL_LOG(WARNING)
           << "reproducible_sampling enabled w/ jitter_with_reflection "
              "disabled. "
           << "reproducible_sampling always uses jitter with reflection, "
@@ -200,15 +203,15 @@ PacketResamplerCalculator::GetSamplingStrategy(
 
 Timestamp PacketResamplerCalculator::PeriodIndexToTimestamp(
     int64_t index) const {
-  CHECK_EQ(jitter_, 0.0);
+  ABSL_CHECK_EQ(jitter_, 0.0);
-  CHECK_NE(first_timestamp_, Timestamp::Unset());
+  ABSL_CHECK_NE(first_timestamp_, Timestamp::Unset());
   return first_timestamp_ + TimestampDiffFromSeconds(index / frame_rate_);
 }
 
 int64_t PacketResamplerCalculator::TimestampToPeriodIndex(
     Timestamp timestamp) const {
-  CHECK_EQ(jitter_, 0.0);
+  ABSL_CHECK_EQ(jitter_, 0.0);
-  CHECK_NE(first_timestamp_, Timestamp::Unset());
+  ABSL_CHECK_NE(first_timestamp_, Timestamp::Unset());
   return MathUtil::SafeRound<int64_t, double>(
       (timestamp - first_timestamp_).Seconds() * frame_rate_);
 }
@@ -229,13 +232,15 @@ absl::Status LegacyJitterWithReflectionStrategy::Open(CalculatorContext* cc) {
 
   if (resampler_options.output_header() !=
       PacketResamplerCalculatorOptions::NONE) {
-    LOG(WARNING) << "VideoHeader::frame_rate holds the target value and not "
+    ABSL_LOG(WARNING)
-                    "the actual value.";
+        << "VideoHeader::frame_rate holds the target value and not "
+           "the actual value.";
   }
 
   if (calculator_->flush_last_packet_) {
-    LOG(WARNING) << "PacketResamplerCalculatorOptions.flush_last_packet is "
+    ABSL_LOG(WARNING)
-                    "ignored, because we are adding jitter.";
+        << "PacketResamplerCalculatorOptions.flush_last_packet is "
+           "ignored, because we are adding jitter.";
   }
 
   const auto& seed = cc->InputSidePackets().Tag(kSeedTag).Get<std::string>();
@@ -254,7 +259,7 @@ absl::Status LegacyJitterWithReflectionStrategy::Open(CalculatorContext* cc) {
 }
 absl::Status LegacyJitterWithReflectionStrategy::Close(CalculatorContext* cc) {
   if (!packet_reservoir_->IsEmpty()) {
-    LOG(INFO) << "Emitting pack from reservoir.";
+    ABSL_LOG(INFO) << "Emitting pack from reservoir.";
     calculator_->OutputWithinLimits(cc, packet_reservoir_->GetSample());
   }
   return absl::OkStatus();
@@ -285,7 +290,7 @@ absl::Status LegacyJitterWithReflectionStrategy::Process(
 
   if (calculator_->frame_time_usec_ <
       (cc->InputTimestamp() - calculator_->last_packet_.Timestamp()).Value()) {
-    LOG_FIRST_N(WARNING, 2)
+    ABSL_LOG_FIRST_N(WARNING, 2)
         << "Adding jitter is not very useful when upsampling.";
   }
 
@@ -340,8 +345,8 @@ void LegacyJitterWithReflectionStrategy::UpdateNextOutputTimestampWithJitter() {
   next_output_timestamp_ = Timestamp(ReflectBetween(
       next_output_timestamp_.Value(), next_output_timestamp_min_.Value(),
       next_output_timestamp_max_.Value()));
-  CHECK_GE(next_output_timestamp_, next_output_timestamp_min_);
+  ABSL_CHECK_GE(next_output_timestamp_, next_output_timestamp_min_);
-  CHECK_LT(next_output_timestamp_, next_output_timestamp_max_);
+  ABSL_CHECK_LT(next_output_timestamp_, next_output_timestamp_max_);
 }
 
 absl::Status ReproducibleJitterWithReflectionStrategy::Open(
@@ -352,13 +357,15 @@ absl::Status ReproducibleJitterWithReflectionStrategy::Open(
 
   if (resampler_options.output_header() !=
       PacketResamplerCalculatorOptions::NONE) {
-    LOG(WARNING) << "VideoHeader::frame_rate holds the target value and not "
+    ABSL_LOG(WARNING)
-                    "the actual value.";
+        << "VideoHeader::frame_rate holds the target value and not "
+           "the actual value.";
   }
 
   if (calculator_->flush_last_packet_) {
-    LOG(WARNING) << "PacketResamplerCalculatorOptions.flush_last_packet is "
+    ABSL_LOG(WARNING)
-                    "ignored, because we are adding jitter.";
+        << "PacketResamplerCalculatorOptions.flush_last_packet is "
+           "ignored, because we are adding jitter.";
   }
 
   const auto& seed = cc->InputSidePackets().Tag(kSeedTag).Get<std::string>();
@@ -411,7 +418,7 @@ absl::Status ReproducibleJitterWithReflectionStrategy::Process(
     // Note, if the stream is upsampling, this could lead to the same packet
     // being emitted twice.  Upsampling and jitter doesn't make much sense
     // but does technically work.
-    LOG_FIRST_N(WARNING, 2)
+    ABSL_LOG_FIRST_N(WARNING, 2)
         << "Adding jitter is not very useful when upsampling.";
   }
 
@@ -499,13 +506,15 @@ absl::Status JitterWithoutReflectionStrategy::Open(CalculatorContext* cc) {
 
   if (resampler_options.output_header() !=
       PacketResamplerCalculatorOptions::NONE) {
-    LOG(WARNING) << "VideoHeader::frame_rate holds the target value and not "
+    ABSL_LOG(WARNING)
-                    "the actual value.";
+        << "VideoHeader::frame_rate holds the target value and not "
+           "the actual value.";
   }
 
   if (calculator_->flush_last_packet_) {
-    LOG(WARNING) << "PacketResamplerCalculatorOptions.flush_last_packet is "
+    ABSL_LOG(WARNING)
-                    "ignored, because we are adding jitter.";
+        << "PacketResamplerCalculatorOptions.flush_last_packet is "
+           "ignored, because we are adding jitter.";
   }
 
   const auto& seed = cc->InputSidePackets().Tag(kSeedTag).Get<std::string>();
@@ -555,7 +564,7 @@ absl::Status JitterWithoutReflectionStrategy::Process(CalculatorContext* cc) {
 
   if (calculator_->frame_time_usec_ <
       (cc->InputTimestamp() - calculator_->last_packet_.Timestamp()).Value()) {
-    LOG_FIRST_N(WARNING, 2)
+    ABSL_LOG_FIRST_N(WARNING, 2)
         << "Adding jitter is not very useful when upsampling.";
   }
 

mediapipe/calculators/core/packet_resampler_calculator.h

@@ -13,7 +13,6 @@
 #include "mediapipe/framework/deps/random_base.h"
 #include "mediapipe/framework/formats/video_stream_header.h"
 #include "mediapipe/framework/port/integral_types.h"
-#include "mediapipe/framework/port/logging.h"
 #include "mediapipe/framework/port/ret_check.h"
 #include "mediapipe/framework/port/status.h"
 #include "mediapipe/framework/port/status_macros.h"

mediapipe/calculators/core/packet_thinner_calculator.cc

@@ -17,6 +17,7 @@
 #include <cmath>  // for ceil
 #include <memory>
 
+#include "absl/log/absl_check.h"
 #include "mediapipe/calculators/core/packet_thinner_calculator.pb.h"
 #include "mediapipe/framework/calculator_context.h"
 #include "mediapipe/framework/calculator_framework.h"
@@ -160,8 +161,8 @@ absl::Status PacketThinnerCalculator::Open(CalculatorContext* cc) {
 
   thinner_type_ = options.thinner_type();
   // This check enables us to assume only two thinner types exist in Process()
-  CHECK(thinner_type_ == PacketThinnerCalculatorOptions::ASYNC ||
+  ABSL_CHECK(thinner_type_ == PacketThinnerCalculatorOptions::ASYNC ||
-        thinner_type_ == PacketThinnerCalculatorOptions::SYNC)
+             thinner_type_ == PacketThinnerCalculatorOptions::SYNC)
       << "Unsupported thinner type.";
 
   if (thinner_type_ == PacketThinnerCalculatorOptions::ASYNC) {
@@ -177,7 +178,8 @@ absl::Status PacketThinnerCalculator::Open(CalculatorContext* cc) {
   } else {
     period_ = TimestampDiff(options.period());
   }
-  CHECK_LT(TimestampDiff(0), period_) << "Specified period must be positive.";
+  ABSL_CHECK_LT(TimestampDiff(0), period_)
+      << "Specified period must be positive.";
 
   if (options.has_start_time()) {
     start_time_ = Timestamp(options.start_time());
@@ -189,7 +191,7 @@ absl::Status PacketThinnerCalculator::Open(CalculatorContext* cc) {
 
   end_time_ =
       options.has_end_time() ? Timestamp(options.end_time()) : Timestamp::Max();
-  CHECK_LT(start_time_, end_time_)
+  ABSL_CHECK_LT(start_time_, end_time_)
       << "Invalid PacketThinner: start_time must be earlier than end_time";
 
   sync_output_timestamps_ = options.sync_output_timestamps();
@@ -232,7 +234,7 @@ absl::Status PacketThinnerCalculator::Close(CalculatorContext* cc) {
   // Emit any saved packets before quitting.
   if (!saved_packet_.IsEmpty()) {
     // Only sync thinner should have saved packets.
-    CHECK_EQ(PacketThinnerCalculatorOptions::SYNC, thinner_type_);
+    ABSL_CHECK_EQ(PacketThinnerCalculatorOptions::SYNC, thinner_type_);
     if (sync_output_timestamps_) {
       cc->Outputs().Index(0).AddPacket(
           saved_packet_.At(NearestSyncTimestamp(saved_packet_.Timestamp())));
@@ -269,7 +271,7 @@ absl::Status PacketThinnerCalculator::SyncThinnerProcess(
     const Timestamp saved_sync = NearestSyncTimestamp(saved);
     const Timestamp now = cc->InputTimestamp();
     const Timestamp now_sync = NearestSyncTimestamp(now);
-    CHECK_LE(saved_sync, now_sync);
+    ABSL_CHECK_LE(saved_sync, now_sync);
     if (saved_sync == now_sync) {
       // Saved Packet is in same interval as current packet.
       // Replace saved packet with current if it is at least as
@@ -295,7 +297,7 @@ absl::Status PacketThinnerCalculator::SyncThinnerProcess(
 }
 
 Timestamp PacketThinnerCalculator::NearestSyncTimestamp(Timestamp now) const {
-  CHECK_NE(start_time_, Timestamp::Unset())
+  ABSL_CHECK_NE(start_time_, Timestamp::Unset())
       << "Method only valid for sync thinner calculator.";
 
   // Computation is done using int64 arithmetic.  No easy way to avoid
@@ -303,12 +305,12 @@ Timestamp PacketThinnerCalculator::NearestSyncTimestamp(Timestamp now) const {
   const int64_t now64 = now.Value();
   const int64_t start64 = start_time_.Value();
   const int64_t period64 = period_.Value();
-  CHECK_LE(0, period64);
+  ABSL_CHECK_LE(0, period64);
 
   // Round now64 to its closest interval (units of period64).
   int64_t sync64 =
       (now64 - start64 + period64 / 2) / period64 * period64 + start64;
-  CHECK_LE(abs(now64 - sync64), period64 / 2)
+  ABSL_CHECK_LE(abs(now64 - sync64), period64 / 2)
       << "start64: " << start64 << "; now64: " << now64
       << "; sync64: " << sync64;
 

mediapipe/calculators/core/packet_thinner_calculator_test.cc

@@ -16,6 +16,7 @@
 #include <string>
 #include <vector>
 
+#include "absl/log/absl_check.h"
 #include "absl/strings/str_cat.h"
 #include "mediapipe/calculators/core/packet_thinner_calculator.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
@@ -70,7 +71,7 @@ class SimpleRunner : public CalculatorRunner {
   }
 
   double GetFrameRate() const {
-    CHECK(!Outputs().Index(0).header.IsEmpty());
+    ABSL_CHECK(!Outputs().Index(0).header.IsEmpty());
     return Outputs().Index(0).header.Get<VideoHeader>().frame_rate;
   }
 };

mediapipe/calculators/core/sequence_shift_calculator.cc

@@ -14,6 +14,7 @@
 
 #include <deque>
 
+#include "absl/log/absl_log.h"
 #include "mediapipe/calculators/core/sequence_shift_calculator.pb.h"
 #include "mediapipe/framework/api2/node.h"
 #include "mediapipe/framework/calculator_framework.h"
@@ -101,7 +102,7 @@ void SequenceShiftCalculator::ProcessPositiveOffset(CalculatorContext* cc) {
     kOut(cc).Send(packet_cache_.front().At(cc->InputTimestamp()));
     packet_cache_.pop_front();
   } else if (emit_empty_packets_before_first_packet_) {
-    LOG(FATAL) << "Not supported yet";
+    ABSL_LOG(FATAL) << "Not supported yet";
   }
   // Store current packet for later output.
   packet_cache_.push_back(kIn(cc).packet());

mediapipe/calculators/image/BUILD

@@ -97,6 +97,7 @@ cc_library(
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/framework/port:source_location",
         "//mediapipe/framework/port:status",
+        "@com_google_absl//absl/log:absl_check",
     ],
     alwayslink = 1,
 )
@@ -125,6 +126,7 @@ cc_library(
         "//mediapipe/framework/port:opencv_imgcodecs",
         "//mediapipe/framework/port:opencv_imgproc",
         "//mediapipe/framework/port:status",
+        "@com_google_absl//absl/log:absl_check",
     ],
     alwayslink = 1,
 )
@@ -151,11 +153,11 @@ cc_library(
         "//mediapipe/framework/formats:image_format_cc_proto",
         "//mediapipe/framework/formats:image_frame",
         "//mediapipe/framework/formats:image_frame_opencv",
-        "//mediapipe/framework/port:logging",
         "//mediapipe/framework/port:opencv_core",
         "//mediapipe/framework/port:opencv_imgproc",
         "//mediapipe/framework/port:status",
         "//mediapipe/framework/port:vector",
+        "@com_google_absl//absl/log:absl_log",
     ] + select({
         "//mediapipe/gpu:disable_gpu": [],
         "//conditions:default": [
@@ -202,6 +204,7 @@ cc_library(
         "//mediapipe/framework/port:opencv_imgproc",
         "//mediapipe/framework/port:status",
         "//mediapipe/framework/port:vector",
+        "@com_google_absl//absl/log:absl_check",
         "@com_google_absl//absl/strings",
     ] + select({
         "//mediapipe/gpu:disable_gpu": [],
@@ -261,9 +264,12 @@ cc_library(
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/framework/port:status",
         "//mediapipe/gpu:scale_mode_cc_proto",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/strings",
     ] + select({
         "//mediapipe/gpu:disable_gpu": [],
         "//conditions:default": [
+            "//mediapipe/gpu:gl_base_hdr",
             "//mediapipe/gpu:gl_calculator_helper",
             "//mediapipe/gpu:gl_quad_renderer",
             "//mediapipe/gpu:gl_simple_shaders",
@@ -273,6 +279,36 @@ cc_library(
     alwayslink = 1,
 )
 
+cc_test(
+    name = "image_transformation_calculator_test",
+    srcs = ["image_transformation_calculator_test.cc"],
+    data = ["//mediapipe/calculators/image/testdata:test_images"],
+    tags = [
+        "desktop_only_test",
+    ],
+    deps = [
+        ":image_transformation_calculator",
+        "//mediapipe/framework:calculator_cc_proto",
+        "//mediapipe/framework:calculator_framework",
+        "//mediapipe/framework:calculator_runner",
+        "//mediapipe/framework/deps:file_path",
+        "//mediapipe/framework/formats:image_format_cc_proto",
+        "//mediapipe/framework/formats:image_frame",
+        "//mediapipe/framework/formats:image_frame_opencv",
+        "//mediapipe/framework/port:gtest",
+        "//mediapipe/framework/port:opencv_imgcodecs",
+        "//mediapipe/framework/port:opencv_imgproc",
+        "//mediapipe/framework/port:parse_text_proto",
+        "//mediapipe/gpu:gpu_buffer_to_image_frame_calculator",
+        "//mediapipe/gpu:image_frame_to_gpu_buffer_calculator",
+        "//third_party:opencv",
+        "@com_google_absl//absl/container:flat_hash_set",
+        "@com_google_absl//absl/flags:flag",
+        "@com_google_absl//absl/strings",
+        "@com_google_googletest//:gtest_main",
+    ],
+)
+
 cc_library(
     name = "image_cropping_calculator",
     srcs = ["image_cropping_calculator.cc"],
@@ -300,6 +336,7 @@ cc_library(
         "//mediapipe/framework/port:opencv_imgproc",
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/framework/port:status",
+        "@com_google_absl//absl/log:absl_log",
     ] + select({
         "//mediapipe/gpu:disable_gpu": [],
         "//conditions:default": [
@@ -396,6 +433,7 @@ cc_library(
         "//mediapipe/framework/port:logging",
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/framework/port:status",
+        "@com_google_absl//absl/log:absl_check",
         "@com_google_absl//absl/strings",
     ],
 )
@@ -420,6 +458,8 @@ cc_library(
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/framework/port:status",
         "//mediapipe/util:image_frame_util",
+        "@com_google_absl//absl/log:absl_check",
+        "@com_google_absl//absl/log:absl_log",
         "@com_google_absl//absl/strings",
         "@libyuv",
     ],
@@ -625,9 +665,9 @@ cc_library(
         "//mediapipe/framework/formats:image",
         "//mediapipe/framework/formats:image_format_cc_proto",
         "//mediapipe/framework/formats:image_frame",
-        "//mediapipe/framework/port:logging",
         "//mediapipe/framework/port:status",
         "//mediapipe/framework/port:vector",
+        "@com_google_absl//absl/log:absl_log",
     ] + select({
         "//mediapipe/gpu:disable_gpu": [],
         "//conditions:default": [
@@ -665,6 +705,7 @@ cc_test(
         "//mediapipe/framework/port:opencv_imgcodecs",
         "//mediapipe/framework/port:opencv_imgproc",
         "//mediapipe/framework/port:parse_text_proto",
+        "@com_google_absl//absl/log:absl_log",
     ],
 )
 
@@ -687,6 +728,7 @@ cc_library(
         "//mediapipe/gpu:gpu_buffer",
         "//mediapipe/gpu:gpu_origin_cc_proto",
         "//mediapipe/gpu:shader_util",
+        "@com_google_absl//absl/log:absl_log",
         "@com_google_absl//absl/memory",
         "@com_google_absl//absl/status",
         "@com_google_absl//absl/status:statusor",

mediapipe/calculators/image/affine_transformation_runner_gl.cc

@@ -20,6 +20,7 @@
 #include "Eigen/Core"
 #include "Eigen/Geometry"
 #include "Eigen/LU"
+#include "absl/log/absl_log.h"
 #include "absl/memory/memory.h"
 #include "absl/status/status.h"
 #include "absl/status/statusor.h"
@@ -53,6 +54,10 @@ bool IsMatrixVerticalFlipNeeded(GpuOrigin::Mode gpu_origin) {
 #endif  //  __APPLE__
     case GpuOrigin::TOP_LEFT:
       return false;
+    default:
+      ABSL_LOG(ERROR) << "Incorrect GpuOrigin: "
+                      << static_cast<int>(gpu_origin);
+      return true;
   }
 }
 
@@ -384,6 +389,8 @@ class GlTextureWarpAffineRunner
     glActiveTexture(GL_TEXTURE0);
     glBindTexture(GL_TEXTURE_2D, 0);
 
+    glFlush();
+
     return absl::OkStatus();
   }
 

mediapipe/calculators/image/bilateral_filter_calculator.cc

@@ -15,6 +15,7 @@
 #include <memory>
 #include <string>
 
+#include "absl/log/absl_check.h"
 #include "absl/strings/str_replace.h"
 #include "mediapipe/calculators/image/bilateral_filter_calculator.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
@@ -112,7 +113,7 @@ class BilateralFilterCalculator : public CalculatorBase {
 REGISTER_CALCULATOR(BilateralFilterCalculator);
 
 absl::Status BilateralFilterCalculator::GetContract(CalculatorContract* cc) {
-  CHECK_GE(cc->Inputs().NumEntries(), 1);
+  RET_CHECK_GE(cc->Inputs().NumEntries(), 1);
 
   if (cc->Inputs().HasTag(kInputFrameTag) &&
       cc->Inputs().HasTag(kInputFrameTagGpu)) {
@@ -183,8 +184,8 @@ absl::Status BilateralFilterCalculator::Open(CalculatorContext* cc) {
 
   sigma_color_ = options_.sigma_color();
   sigma_space_ = options_.sigma_space();
-  CHECK_GE(sigma_color_, 0.0);
+  ABSL_CHECK_GE(sigma_color_, 0.0);
-  CHECK_GE(sigma_space_, 0.0);
+  ABSL_CHECK_GE(sigma_space_, 0.0);
   if (!use_gpu_) sigma_color_ *= 255.0;
 
   if (use_gpu_) {

mediapipe/calculators/image/color_convert_calculator.cc

@@ -12,6 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+#include "absl/log/absl_check.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/formats/image_frame.h"
 #include "mediapipe/framework/formats/image_frame_opencv.h"
@@ -25,8 +26,8 @@
 namespace mediapipe {
 namespace {
 void SetColorChannel(int channel, uint8 value, cv::Mat* mat) {
-  CHECK(mat->depth() == CV_8U);
+  ABSL_CHECK(mat->depth() == CV_8U);
-  CHECK(channel < mat->channels());
+  ABSL_CHECK(channel < mat->channels());
   const int step = mat->channels();
   for (int r = 0; r < mat->rows; ++r) {
     uint8* row_ptr = mat->ptr<uint8>(r);

mediapipe/calculators/image/image_cropping_calculator.cc

@@ -16,6 +16,7 @@
 
 #include <cmath>
 
+#include "absl/log/absl_log.h"
 #include "mediapipe/framework/formats/image_frame.h"
 #include "mediapipe/framework/formats/image_frame_opencv.h"
 #include "mediapipe/framework/formats/rect.pb.h"
@@ -202,8 +203,9 @@ absl::Status ImageCroppingCalculator::ValidateBorderModeForGPU(
 
   switch (options.border_mode()) {
     case mediapipe::ImageCroppingCalculatorOptions::BORDER_ZERO:
-      LOG(WARNING) << "BORDER_ZERO mode is not supported by GPU "
+      ABSL_LOG(WARNING)
-                   << "implementation and will fall back into BORDER_REPLICATE";
+          << "BORDER_ZERO mode is not supported by GPU "
+          << "implementation and will fall back into BORDER_REPLICATE";
       break;
     case mediapipe::ImageCroppingCalculatorOptions::BORDER_REPLICATE:
       break;

mediapipe/calculators/image/image_transformation_calculator.cc

@@ -12,6 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+#include "absl/status/status.h"
 #include "mediapipe/calculators/image/image_transformation_calculator.pb.h"
 #include "mediapipe/calculators/image/rotation_mode.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
@@ -27,6 +28,7 @@
 #include "mediapipe/gpu/scale_mode.pb.h"
 
 #if !MEDIAPIPE_DISABLE_GPU
+#include "mediapipe/gpu/gl_base.h"
 #include "mediapipe/gpu/gl_calculator_helper.h"
 #include "mediapipe/gpu/gl_quad_renderer.h"
 #include "mediapipe/gpu/gl_simple_shaders.h"
@@ -60,42 +62,42 @@ constexpr char kVideoPrestreamTag[] = "VIDEO_PRESTREAM";
 
 int RotationModeToDegrees(mediapipe::RotationMode_Mode rotation) {
   switch (rotation) {
-    case mediapipe::RotationMode_Mode_UNKNOWN:
+    case mediapipe::RotationMode::UNKNOWN:
-    case mediapipe::RotationMode_Mode_ROTATION_0:
+    case mediapipe::RotationMode::ROTATION_0:
       return 0;
-    case mediapipe::RotationMode_Mode_ROTATION_90:
+    case mediapipe::RotationMode::ROTATION_90:
       return 90;
-    case mediapipe::RotationMode_Mode_ROTATION_180:
+    case mediapipe::RotationMode::ROTATION_180:
       return 180;
-    case mediapipe::RotationMode_Mode_ROTATION_270:
+    case mediapipe::RotationMode::ROTATION_270:
       return 270;
   }
 }
 mediapipe::RotationMode_Mode DegreesToRotationMode(int degrees) {
   switch (degrees) {
     case 0:
-      return mediapipe::RotationMode_Mode_ROTATION_0;
+      return mediapipe::RotationMode::ROTATION_0;
     case 90:
-      return mediapipe::RotationMode_Mode_ROTATION_90;
+      return mediapipe::RotationMode::ROTATION_90;
     case 180:
-      return mediapipe::RotationMode_Mode_ROTATION_180;
+      return mediapipe::RotationMode::ROTATION_180;
     case 270:
-      return mediapipe::RotationMode_Mode_ROTATION_270;
+      return mediapipe::RotationMode::ROTATION_270;
     default:
-      return mediapipe::RotationMode_Mode_UNKNOWN;
+      return mediapipe::RotationMode::UNKNOWN;
   }
 }
 mediapipe::ScaleMode_Mode ParseScaleMode(
     mediapipe::ScaleMode_Mode scale_mode,
     mediapipe::ScaleMode_Mode default_mode) {
   switch (scale_mode) {
-    case mediapipe::ScaleMode_Mode_DEFAULT:
+    case mediapipe::ScaleMode::DEFAULT:
       return default_mode;
-    case mediapipe::ScaleMode_Mode_STRETCH:
+    case mediapipe::ScaleMode::STRETCH:
       return scale_mode;
-    case mediapipe::ScaleMode_Mode_FIT:
+    case mediapipe::ScaleMode::FIT:
       return scale_mode;
-    case mediapipe::ScaleMode_Mode_FILL_AND_CROP:
+    case mediapipe::ScaleMode::FILL_AND_CROP:
       return scale_mode;
     default:
       return default_mode;
@@ -208,6 +210,8 @@ class ImageTransformationCalculator : public CalculatorBase {
 
   bool use_gpu_ = false;
   cv::Scalar padding_color_;
+  ImageTransformationCalculatorOptions::InterpolationMode interpolation_mode_;
+
 #if !MEDIAPIPE_DISABLE_GPU
   GlCalculatorHelper gpu_helper_;
   std::unique_ptr<QuadRenderer> rgb_renderer_;
@@ -343,6 +347,11 @@ absl::Status ImageTransformationCalculator::Open(CalculatorContext* cc) {
                               options_.padding_color().green(),
                               options_.padding_color().blue());
 
+  interpolation_mode_ = options_.interpolation_mode();
+  if (options_.interpolation_mode() ==
+      ImageTransformationCalculatorOptions::DEFAULT) {
+    interpolation_mode_ = ImageTransformationCalculatorOptions::LINEAR;
+  }
   if (use_gpu_) {
 #if !MEDIAPIPE_DISABLE_GPU
     // Let the helper access the GL context information.
@@ -457,26 +466,48 @@ absl::Status ImageTransformationCalculator::RenderCpu(CalculatorContext* cc) {
   ComputeOutputDimensions(input_width, input_height, &output_width,
                           &output_height);
 
+  int opencv_interpolation_mode = cv::INTER_LINEAR;
   if (output_width_ > 0 && output_height_ > 0) {
     cv::Mat scaled_mat;
-    if (scale_mode_ == mediapipe::ScaleMode_Mode_STRETCH) {
+    if (scale_mode_ == mediapipe::ScaleMode::STRETCH) {
-      int scale_flag =
+      if (interpolation_mode_ == ImageTransformationCalculatorOptions::LINEAR) {
-          input_mat.cols > output_width_ && input_mat.rows > output_height_
+        // Use INTER_AREA for downscaling if interpolation mode is set to
-              ? cv::INTER_AREA
+        // LINEAR.
-              : cv::INTER_LINEAR;
+        if (input_mat.cols > output_width_ && input_mat.rows > output_height_) {
+          opencv_interpolation_mode = cv::INTER_AREA;
+
+        } else {
+          opencv_interpolation_mode = cv::INTER_LINEAR;
+        }
+      } else {
+        opencv_interpolation_mode = cv::INTER_NEAREST;
+      }
       cv::resize(input_mat, scaled_mat, cv::Size(output_width_, output_height_),
-                 0, 0, scale_flag);
+                 0, 0, opencv_interpolation_mode);
     } else {
       const float scale =
           std::min(static_cast<float>(output_width_) / input_width,
                    static_cast<float>(output_height_) / input_height);
       const int target_width = std::round(input_width * scale);
       const int target_height = std::round(input_height * scale);
-      int scale_flag = scale < 1.0f ? cv::INTER_AREA : cv::INTER_LINEAR;
+
-      if (scale_mode_ == mediapipe::ScaleMode_Mode_FIT) {
+      if (interpolation_mode_ == ImageTransformationCalculatorOptions::LINEAR) {
+        // Use INTER_AREA for downscaling if interpolation mode is set to
+        // LINEAR.
+        if (scale < 1.0f) {
+          opencv_interpolation_mode = cv::INTER_AREA;
+        } else {
+          opencv_interpolation_mode = cv::INTER_LINEAR;
+        }
+      } else {
+        opencv_interpolation_mode = cv::INTER_NEAREST;
+      }
+
+      if (scale_mode_ == mediapipe::ScaleMode::FIT) {
         cv::Mat intermediate_mat;
         cv::resize(input_mat, intermediate_mat,
-                   cv::Size(target_width, target_height), 0, 0, scale_flag);
+                   cv::Size(target_width, target_height), 0, 0,
+                   opencv_interpolation_mode);
         const int top = (output_height_ - target_height) / 2;
         const int bottom = output_height_ - target_height - top;
         const int left = (output_width_ - target_width) / 2;
@@ -488,7 +519,7 @@ absl::Status ImageTransformationCalculator::RenderCpu(CalculatorContext* cc) {
                            padding_color_);
       } else {
         cv::resize(input_mat, scaled_mat, cv::Size(target_width, target_height),
-                   0, 0, scale_flag);
+                   0, 0, opencv_interpolation_mode);
         output_width = target_width;
         output_height = target_height;
       }
@@ -514,17 +545,17 @@ absl::Status ImageTransformationCalculator::RenderCpu(CalculatorContext* cc) {
     cv::warpAffine(input_mat, rotated_mat, rotation_mat, rotated_size);
   } else {
     switch (rotation_) {
-      case mediapipe::RotationMode_Mode_UNKNOWN:
+      case mediapipe::RotationMode::UNKNOWN:
-      case mediapipe::RotationMode_Mode_ROTATION_0:
+      case mediapipe::RotationMode::ROTATION_0:
         rotated_mat = input_mat;
         break;
-      case mediapipe::RotationMode_Mode_ROTATION_90:
+      case mediapipe::RotationMode::ROTATION_90:
         cv::rotate(input_mat, rotated_mat, cv::ROTATE_90_COUNTERCLOCKWISE);
         break;
-      case mediapipe::RotationMode_Mode_ROTATION_180:
+      case mediapipe::RotationMode::ROTATION_180:
         cv::rotate(input_mat, rotated_mat, cv::ROTATE_180);
         break;
-      case mediapipe::RotationMode_Mode_ROTATION_270:
+      case mediapipe::RotationMode::ROTATION_270:
         cv::rotate(input_mat, rotated_mat, cv::ROTATE_90_CLOCKWISE);
         break;
     }
@@ -561,7 +592,7 @@ absl::Status ImageTransformationCalculator::RenderGpu(CalculatorContext* cc) {
   ComputeOutputDimensions(input_width, input_height, &output_width,
                           &output_height);
 
-  if (scale_mode_ == mediapipe::ScaleMode_Mode_FILL_AND_CROP) {
+  if (scale_mode_ == mediapipe::ScaleMode::FILL_AND_CROP) {
     const float scale =
         std::min(static_cast<float>(output_width_) / input_width,
                  static_cast<float>(output_height_) / input_height);
@@ -628,6 +659,12 @@ absl::Status ImageTransformationCalculator::RenderGpu(CalculatorContext* cc) {
   glActiveTexture(GL_TEXTURE1);
   glBindTexture(src1.target(), src1.name());
 
+  if (interpolation_mode_ == ImageTransformationCalculatorOptions::NEAREST) {
+    // TODO: revert texture params.
+    glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+    glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+  }
+
   MP_RETURN_IF_ERROR(renderer->GlRender(
       src1.width(), src1.height(), dst.width(), dst.height(), scale_mode,
       rotation, flip_horizontally_, flip_vertically_,
@@ -652,8 +689,8 @@ void ImageTransformationCalculator::ComputeOutputDimensions(
   if (output_width_ > 0 && output_height_ > 0) {
     *output_width = output_width_;
     *output_height = output_height_;
-  } else if (rotation_ == mediapipe::RotationMode_Mode_ROTATION_90 ||
+  } else if (rotation_ == mediapipe::RotationMode::ROTATION_90 ||
-             rotation_ == mediapipe::RotationMode_Mode_ROTATION_270) {
+             rotation_ == mediapipe::RotationMode::ROTATION_270) {
     *output_width = input_height;
     *output_height = input_width;
   } else {
@@ -666,9 +703,9 @@ void ImageTransformationCalculator::ComputeOutputLetterboxPadding(
     int input_width, int input_height, int output_width, int output_height,
     std::array<float, 4>* padding) {
   padding->fill(0.f);
-  if (scale_mode_ == mediapipe::ScaleMode_Mode_FIT) {
+  if (scale_mode_ == mediapipe::ScaleMode::FIT) {
-    if (rotation_ == mediapipe::RotationMode_Mode_ROTATION_90 ||
+    if (rotation_ == mediapipe::RotationMode::ROTATION_90 ||
-        rotation_ == mediapipe::RotationMode_Mode_ROTATION_270) {
+        rotation_ == mediapipe::RotationMode::ROTATION_270) {
       std::swap(input_width, input_height);
     }
     const float input_aspect_ratio =

mediapipe/calculators/image/image_transformation_calculator.proto

@@ -54,4 +54,15 @@ message ImageTransformationCalculatorOptions {
   // The color for the padding. This option is only used when the scale mode is
   // FIT. Default is black. This is for CPU only.
   optional Color padding_color = 8;
+
+  // Interpolation method to use. Note that on CPU when LINEAR is specified,
+  // INTER_LINEAR is used for upscaling and INTER_AREA is used for downscaling.
+  enum InterpolationMode {
+    DEFAULT = 0;
+    LINEAR = 1;
+    NEAREST = 2;
+  }
+
+  // Mode DEFAULT will use LINEAR interpolation.
+  optional InterpolationMode interpolation_mode = 9;
 }

mediapipe/calculators/image/image_transformation_calculator_test.cc

@@ -0,0 +1,315 @@
+#include <string>
+#include <utility>
+#include <vector>
+
+#include "absl/container/flat_hash_set.h"
+#include "absl/flags/flag.h"
+#include "absl/strings/substitute.h"
+#include "mediapipe/framework/calculator.pb.h"
+#include "mediapipe/framework/calculator_framework.h"
+#include "mediapipe/framework/calculator_runner.h"
+#include "mediapipe/framework/deps/file_path.h"
+#include "mediapipe/framework/formats/image_format.pb.h"
+#include "mediapipe/framework/formats/image_frame.h"
+#include "mediapipe/framework/formats/image_frame_opencv.h"
+#include "mediapipe/framework/port/gtest.h"
+#include "mediapipe/framework/port/opencv_imgcodecs_inc.h"
+#include "mediapipe/framework/port/opencv_imgproc_inc.h"
+#include "mediapipe/framework/port/parse_text_proto.h"
+#include "testing/base/public/gmock.h"
+#include "testing/base/public/googletest.h"
+#include "third_party/OpenCV/core.hpp"  // IWYU pragma: keep
+#include "third_party/OpenCV/core/mat.hpp"
+
+namespace mediapipe {
+
+namespace {
+
+absl::flat_hash_set<int> computeUniqueValues(const cv::Mat& mat) {
+  // Compute the unique values in cv::Mat
+  absl::flat_hash_set<int> unique_values;
+  for (int i = 0; i < mat.rows; i++) {
+    for (int j = 0; j < mat.cols; j++) {
+      unique_values.insert(mat.at<unsigned char>(i, j));
+    }
+  }
+  return unique_values;
+}
+
+TEST(ImageTransformationCalculatorTest, NearestNeighborResizing) {
+  cv::Mat input_mat;
+  cv::cvtColor(cv::imread(file::JoinPath("./",
+                                         "/mediapipe/calculators/"
+                                         "image/testdata/binary_mask.png")),
+               input_mat, cv::COLOR_BGR2GRAY);
+  Packet input_image_packet = MakePacket<ImageFrame>(
+      ImageFormat::GRAY8, input_mat.size().width, input_mat.size().height);
+  input_mat.copyTo(formats::MatView(&(input_image_packet.Get<ImageFrame>())));
+
+  std::vector<std::pair<int, int>> output_dims{
+      {256, 333}, {512, 512}, {1024, 1024}};
+
+  for (auto& output_dim : output_dims) {
+    Packet input_output_dim_packet =
+        MakePacket<std::pair<int, int>>(output_dim);
+    std::vector<std::string> scale_modes{"FIT", "STRETCH"};
+    for (const auto& scale_mode : scale_modes) {
+      CalculatorGraphConfig::Node node_config =
+          ParseTextProtoOrDie<CalculatorGraphConfig::Node>(
+              absl::Substitute(R"(
+          calculator: "ImageTransformationCalculator"
+          input_stream: "IMAGE:input_image"
+          input_stream: "OUTPUT_DIMENSIONS:image_size"
+          output_stream: "IMAGE:output_image"
+          options: {
+            [mediapipe.ImageTransformationCalculatorOptions.ext]: {
+              scale_mode: $0
+              interpolation_mode: NEAREST
+            }
+          })",
+                               scale_mode));
+
+      CalculatorRunner runner(node_config);
+      runner.MutableInputs()->Tag("IMAGE").packets.push_back(
+          input_image_packet.At(Timestamp(0)));
+      runner.MutableInputs()
+          ->Tag("OUTPUT_DIMENSIONS")
+          .packets.push_back(input_output_dim_packet.At(Timestamp(0)));
+
+      MP_ASSERT_OK(runner.Run());
+      const auto& outputs = runner.Outputs();
+      ASSERT_EQ(outputs.NumEntries(), 1);
+      const std::vector<Packet>& packets = outputs.Tag("IMAGE").packets;
+      ASSERT_EQ(packets.size(), 1);
+      const auto& result = packets[0].Get<ImageFrame>();
+      ASSERT_EQ(output_dim.first, result.Width());
+      ASSERT_EQ(output_dim.second, result.Height());
+
+      auto unique_input_values = computeUniqueValues(input_mat);
+      auto unique_output_values =
+          computeUniqueValues(formats::MatView(&result));
+      EXPECT_THAT(unique_input_values,
+                  ::testing::ContainerEq(unique_output_values));
+    }
+  }
+}
+
+TEST(ImageTransformationCalculatorTest,
+     NearestNeighborResizingWorksForFloatInput) {
+  cv::Mat input_mat;
+  cv::cvtColor(cv::imread(file::JoinPath("./",
+                                         "/mediapipe/calculators/"
+                                         "image/testdata/binary_mask.png")),
+               input_mat, cv::COLOR_BGR2GRAY);
+  Packet input_image_packet = MakePacket<ImageFrame>(
+      ImageFormat::VEC32F1, input_mat.size().width, input_mat.size().height);
+  cv::Mat packet_mat_view =
+      formats::MatView(&(input_image_packet.Get<ImageFrame>()));
+  input_mat.convertTo(packet_mat_view, CV_32FC1, 1 / 255.f);
+
+  std::vector<std::pair<int, int>> output_dims{
+      {256, 333}, {512, 512}, {1024, 1024}};
+
+  for (auto& output_dim : output_dims) {
+    Packet input_output_dim_packet =
+        MakePacket<std::pair<int, int>>(output_dim);
+    std::vector<std::string> scale_modes{"FIT", "STRETCH"};
+    for (const auto& scale_mode : scale_modes) {
+      CalculatorGraphConfig::Node node_config =
+          ParseTextProtoOrDie<CalculatorGraphConfig::Node>(
+              absl::Substitute(R"(
+          calculator: "ImageTransformationCalculator"
+          input_stream: "IMAGE:input_image"
+          input_stream: "OUTPUT_DIMENSIONS:image_size"
+          output_stream: "IMAGE:output_image"
+          options: {
+            [mediapipe.ImageTransformationCalculatorOptions.ext]: {
+              scale_mode: $0
+              interpolation_mode: NEAREST
+            }
+          })",
+                               scale_mode));
+
+      CalculatorRunner runner(node_config);
+      runner.MutableInputs()->Tag("IMAGE").packets.push_back(
+          input_image_packet.At(Timestamp(0)));
+      runner.MutableInputs()
+          ->Tag("OUTPUT_DIMENSIONS")
+          .packets.push_back(input_output_dim_packet.At(Timestamp(0)));
+
+      MP_ASSERT_OK(runner.Run());
+      const auto& outputs = runner.Outputs();
+      ASSERT_EQ(outputs.NumEntries(), 1);
+      const std::vector<Packet>& packets = outputs.Tag("IMAGE").packets;
+      ASSERT_EQ(packets.size(), 1);
+      const auto& result = packets[0].Get<ImageFrame>();
+      ASSERT_EQ(output_dim.first, result.Width());
+      ASSERT_EQ(output_dim.second, result.Height());
+
+      auto unique_input_values = computeUniqueValues(packet_mat_view);
+      auto unique_output_values =
+          computeUniqueValues(formats::MatView(&result));
+      EXPECT_THAT(unique_input_values,
+                  ::testing::ContainerEq(unique_output_values));
+    }
+  }
+}
+
+TEST(ImageTransformationCalculatorTest, NearestNeighborResizingGpu) {
+  cv::Mat input_mat;
+  cv::cvtColor(cv::imread(file::JoinPath("./",
+                                         "/mediapipe/calculators/"
+                                         "image/testdata/binary_mask.png")),
+               input_mat, cv::COLOR_BGR2RGBA);
+
+  std::vector<std::pair<int, int>> output_dims{
+      {256, 333}, {512, 512}, {1024, 1024}};
+
+  for (auto& output_dim : output_dims) {
+    std::vector<std::string> scale_modes{"FIT"};  //, "STRETCH"};
+    for (const auto& scale_mode : scale_modes) {
+      CalculatorGraphConfig graph_config =
+          ParseTextProtoOrDie<CalculatorGraphConfig>(
+              absl::Substitute(R"(
+          input_stream: "input_image"
+          input_stream: "image_size"
+          output_stream: "output_image"
+
+          node {
+            calculator: "ImageFrameToGpuBufferCalculator"
+            input_stream: "input_image"
+            output_stream: "input_image_gpu"
+          }
+
+          node {
+            calculator: "ImageTransformationCalculator"
+            input_stream: "IMAGE_GPU:input_image_gpu"
+            input_stream: "OUTPUT_DIMENSIONS:image_size"
+            output_stream: "IMAGE_GPU:output_image_gpu"
+            options: {
+              [mediapipe.ImageTransformationCalculatorOptions.ext]: {
+                scale_mode: $0
+                interpolation_mode: NEAREST
+              }
+            }
+          }
+          node {
+            calculator: "GpuBufferToImageFrameCalculator"
+            input_stream: "output_image_gpu"
+            output_stream: "output_image"
+          })",
+                               scale_mode));
+      ImageFrame input_image(ImageFormat::SRGBA, input_mat.size().width,
+                             input_mat.size().height);
+      input_mat.copyTo(formats::MatView(&input_image));
+
+      std::vector<Packet> output_image_packets;
+      tool::AddVectorSink("output_image", &graph_config, &output_image_packets);
+
+      CalculatorGraph graph(graph_config);
+      MP_ASSERT_OK(graph.StartRun({}));
+
+      MP_ASSERT_OK(graph.AddPacketToInputStream(
+          "input_image",
+          MakePacket<ImageFrame>(std::move(input_image)).At(Timestamp(0))));
+      MP_ASSERT_OK(graph.AddPacketToInputStream(
+          "image_size",
+          MakePacket<std::pair<int, int>>(output_dim).At(Timestamp(0))));
+
+      MP_ASSERT_OK(graph.WaitUntilIdle());
+      ASSERT_THAT(output_image_packets, testing::SizeIs(1));
+
+      const auto& output_image = output_image_packets[0].Get<ImageFrame>();
+      ASSERT_EQ(output_dim.first, output_image.Width());
+      ASSERT_EQ(output_dim.second, output_image.Height());
+
+      auto unique_input_values = computeUniqueValues(input_mat);
+      auto unique_output_values =
+          computeUniqueValues(formats::MatView(&output_image));
+      EXPECT_THAT(unique_input_values,
+                  ::testing::ContainerEq(unique_output_values));
+    }
+  }
+}
+
+TEST(ImageTransformationCalculatorTest,
+     NearestNeighborResizingWorksForFloatTexture) {
+  cv::Mat input_mat;
+  cv::cvtColor(cv::imread(file::JoinPath("./",
+                                         "/mediapipe/calculators/"
+                                         "image/testdata/binary_mask.png")),
+               input_mat, cv::COLOR_BGR2GRAY);
+  Packet input_image_packet = MakePacket<ImageFrame>(
+      ImageFormat::VEC32F1, input_mat.size().width, input_mat.size().height);
+  cv::Mat packet_mat_view =
+      formats::MatView(&(input_image_packet.Get<ImageFrame>()));
+  input_mat.convertTo(packet_mat_view, CV_32FC1, 1 / 255.f);
+
+  std::vector<std::pair<int, int>> output_dims{
+      {256, 333}, {512, 512}, {1024, 1024}};
+
+  for (auto& output_dim : output_dims) {
+    std::vector<std::string> scale_modes{"FIT"};  //, "STRETCH"};
+    for (const auto& scale_mode : scale_modes) {
+      CalculatorGraphConfig graph_config =
+          ParseTextProtoOrDie<CalculatorGraphConfig>(
+              absl::Substitute(R"(
+          input_stream: "input_image"
+          input_stream: "image_size"
+          output_stream: "output_image"
+
+          node {
+            calculator: "ImageFrameToGpuBufferCalculator"
+            input_stream: "input_image"
+            output_stream: "input_image_gpu"
+          }
+
+          node {
+            calculator: "ImageTransformationCalculator"
+            input_stream: "IMAGE_GPU:input_image_gpu"
+            input_stream: "OUTPUT_DIMENSIONS:image_size"
+            output_stream: "IMAGE_GPU:output_image_gpu"
+            options: {
+              [mediapipe.ImageTransformationCalculatorOptions.ext]: {
+                scale_mode: $0
+                interpolation_mode: NEAREST
+              }
+            }
+          }
+          node {
+            calculator: "GpuBufferToImageFrameCalculator"
+            input_stream: "output_image_gpu"
+            output_stream: "output_image"
+          })",
+                               scale_mode));
+
+      std::vector<Packet> output_image_packets;
+      tool::AddVectorSink("output_image", &graph_config, &output_image_packets);
+
+      CalculatorGraph graph(graph_config);
+      MP_ASSERT_OK(graph.StartRun({}));
+
+      MP_ASSERT_OK(graph.AddPacketToInputStream(
+          "input_image", input_image_packet.At(Timestamp(0))));
+      MP_ASSERT_OK(graph.AddPacketToInputStream(
+          "image_size",
+          MakePacket<std::pair<int, int>>(output_dim).At(Timestamp(0))));
+
+      MP_ASSERT_OK(graph.WaitUntilIdle());
+      ASSERT_THAT(output_image_packets, testing::SizeIs(1));
+
+      const auto& output_image = output_image_packets[0].Get<ImageFrame>();
+      ASSERT_EQ(output_dim.first, output_image.Width());
+      ASSERT_EQ(output_dim.second, output_image.Height());
+
+      auto unique_input_values = computeUniqueValues(packet_mat_view);
+      auto unique_output_values =
+          computeUniqueValues(formats::MatView(&output_image));
+      EXPECT_THAT(unique_input_values,
+                  ::testing::ContainerEq(unique_output_values));
+    }
+  }
+}
+
+}  // namespace
+}  // namespace mediapipe

mediapipe/calculators/image/opencv_image_encoder_calculator.cc

@@ -12,6 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+#include "absl/log/absl_check.h"
 #include "mediapipe/calculators/image/opencv_image_encoder_calculator.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/formats/image_frame_opencv.h"
@@ -61,7 +62,7 @@ absl::Status OpenCvImageEncoderCalculator::Open(CalculatorContext* cc) {
 
 absl::Status OpenCvImageEncoderCalculator::Process(CalculatorContext* cc) {
   const ImageFrame& image_frame = cc->Inputs().Index(0).Get<ImageFrame>();
-  CHECK_EQ(1, image_frame.ByteDepth());
+  ABSL_CHECK_EQ(1, image_frame.ByteDepth());
 
   std::unique_ptr<OpenCvImageEncoderCalculatorResults> encoded_result =
       absl::make_unique<OpenCvImageEncoderCalculatorResults>();

mediapipe/calculators/image/scale_image_calculator.cc

@@ -18,6 +18,8 @@
 #include <memory>
 #include <string>
 
+#include "absl/log/absl_check.h"
+#include "absl/log/absl_log.h"
 #include "absl/strings/str_cat.h"
 #include "absl/strings/substitute.h"
 #include "libyuv/scale.h"
@@ -293,7 +295,7 @@ absl::Status ScaleImageCalculator::InitializeFrameInfo(CalculatorContext* cc) {
     header->width = output_width_;
     header->height = output_height_;
     header->format = output_format_;
-    LOG(INFO) << "OUTPUTTING HEADER on stream";
+    ABSL_LOG(INFO) << "OUTPUTTING HEADER on stream";
     cc->Outputs()
         .Tag("VIDEO_HEADER")
         .Add(header.release(), Timestamp::PreStream());
@@ -393,10 +395,11 @@ absl::Status ScaleImageCalculator::Open(CalculatorContext* cc) {
           .SetHeader(Adopt(output_header.release()));
       has_header_ = true;
     } else {
-      LOG(WARNING) << "Stream had a VideoHeader which didn't have sufficient "
+      ABSL_LOG(WARNING)
-                      "information.  "
+          << "Stream had a VideoHeader which didn't have sufficient "
-                      "Dropping VideoHeader and trying to deduce needed "
+             "information.  "
-                      "information.";
+             "Dropping VideoHeader and trying to deduce needed "
+             "information.";
       input_width_ = 0;
       input_height_ = 0;
       if (!options_.has_input_format()) {
@@ -507,7 +510,7 @@ absl::Status ScaleImageCalculator::ValidateImageFrame(
 
 absl::Status ScaleImageCalculator::ValidateYUVImage(CalculatorContext* cc,
                                                     const YUVImage& yuv_image) {
-  CHECK_EQ(input_format_, ImageFormat::YCBCR420P);
+  ABSL_CHECK_EQ(input_format_, ImageFormat::YCBCR420P);
   if (!has_header_) {
     if (input_width_ != yuv_image.width() ||
         input_height_ != yuv_image.height()) {

mediapipe/calculators/image/scale_image_utils.cc

@@ -18,6 +18,7 @@
 
 #include <string>
 
+#include "absl/log/absl_check.h"
 #include "absl/strings/str_split.h"
 #include "mediapipe/framework/port/logging.h"
 #include "mediapipe/framework/port/ret_check.h"
@@ -40,10 +41,10 @@ absl::Status FindCropDimensions(int input_width, int input_height,    //
                                 const std::string& max_aspect_ratio,  //
                                 int* crop_width, int* crop_height,    //
                                 int* col_start, int* row_start) {
-  CHECK(crop_width);
+  ABSL_CHECK(crop_width);
-  CHECK(crop_height);
+  ABSL_CHECK(crop_height);
-  CHECK(col_start);
+  ABSL_CHECK(col_start);
-  CHECK(row_start);
+  ABSL_CHECK(row_start);
 
   double min_aspect_ratio_q = 0.0;
   double max_aspect_ratio_q = 0.0;
@@ -83,8 +84,8 @@ absl::Status FindCropDimensions(int input_width, int input_height,    //
     }
   }
 
-  CHECK_LE(*crop_width, input_width);
+  ABSL_CHECK_LE(*crop_width, input_width);
-  CHECK_LE(*crop_height, input_height);
+  ABSL_CHECK_LE(*crop_height, input_height);
   return absl::OkStatus();
 }
 
@@ -96,8 +97,8 @@ absl::Status FindOutputDimensions(int input_width,             //
                                   bool preserve_aspect_ratio,  //
                                   int scale_to_multiple_of,    //
                                   int* output_width, int* output_height) {
-  CHECK(output_width);
+  ABSL_CHECK(output_width);
-  CHECK(output_height);
+  ABSL_CHECK(output_height);
 
   if (target_max_area > 0 && input_width * input_height > target_max_area) {
     preserve_aspect_ratio = true;

mediapipe/calculators/image/segmentation_smoothing_calculator.cc

@@ -15,13 +15,13 @@
 #include <algorithm>
 #include <memory>
 
+#include "absl/log/absl_log.h"
 #include "mediapipe/calculators/image/segmentation_smoothing_calculator.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/calculator_options.pb.h"
 #include "mediapipe/framework/formats/image.h"
 #include "mediapipe/framework/formats/image_format.pb.h"
 #include "mediapipe/framework/formats/image_frame.h"
-#include "mediapipe/framework/port/logging.h"
 #include "mediapipe/framework/port/status.h"
 #include "mediapipe/framework/port/vector.h"
 
@@ -110,7 +110,7 @@ REGISTER_CALCULATOR(SegmentationSmoothingCalculator);
 
 absl::Status SegmentationSmoothingCalculator::GetContract(
     CalculatorContract* cc) {
-  CHECK_GE(cc->Inputs().NumEntries(), 1);
+  RET_CHECK_GE(cc->Inputs().NumEntries(), 1);
 
   cc->Inputs().Tag(kCurrentMaskTag).Set<Image>();
   cc->Inputs().Tag(kPreviousMaskTag).Set<Image>();
@@ -273,7 +273,7 @@ absl::Status SegmentationSmoothingCalculator::RenderGpu(CalculatorContext* cc) {
 
   const auto& previous_frame = cc->Inputs().Tag(kPreviousMaskTag).Get<Image>();
   if (previous_frame.format() != current_frame.format()) {
-    LOG(ERROR) << "Warning: mixing input format types. ";
+    ABSL_LOG(ERROR) << "Warning: mixing input format types. ";
   }
   auto previous_texture = gpu_helper_.CreateSourceTexture(previous_frame);
 

mediapipe/calculators/image/segmentation_smoothing_calculator_test.cc

@@ -14,6 +14,7 @@
 
 #include <memory>
 
+#include "absl/log/absl_log.h"
 #include "mediapipe/calculators/image/segmentation_smoothing_calculator.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/calculator_runner.h"
@@ -169,7 +170,7 @@ void RunTest(bool use_gpu, float mix_ratio, cv::Mat& test_result) {
       }
     }
   } else {
-    LOG(ERROR) << "invalid ratio";
+    ABSL_LOG(ERROR) << "invalid ratio";
   }
 }
 

mediapipe/calculators/image/set_alpha_calculator.cc

@@ -14,13 +14,13 @@
 
 #include <memory>
 
+#include "absl/log/absl_log.h"
 #include "mediapipe/calculators/image/set_alpha_calculator.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/calculator_options.pb.h"
 #include "mediapipe/framework/formats/image_format.pb.h"
 #include "mediapipe/framework/formats/image_frame.h"
 #include "mediapipe/framework/formats/image_frame_opencv.h"
-#include "mediapipe/framework/port/logging.h"
 #include "mediapipe/framework/port/opencv_core_inc.h"
 #include "mediapipe/framework/port/opencv_imgproc_inc.h"
 #include "mediapipe/framework/port/status.h"
@@ -142,7 +142,7 @@ class SetAlphaCalculator : public CalculatorBase {
 REGISTER_CALCULATOR(SetAlphaCalculator);
 
 absl::Status SetAlphaCalculator::GetContract(CalculatorContract* cc) {
-  CHECK_GE(cc->Inputs().NumEntries(), 1);
+  RET_CHECK_GE(cc->Inputs().NumEntries(), 1);
 
   bool use_gpu = false;
 
@@ -268,7 +268,7 @@ absl::Status SetAlphaCalculator::RenderCpu(CalculatorContext* cc) {
   const auto& input_frame = cc->Inputs().Tag(kInputFrameTag).Get<ImageFrame>();
   const cv::Mat input_mat = formats::MatView(&input_frame);
   if (!(input_mat.type() == CV_8UC3 || input_mat.type() == CV_8UC4)) {
-    LOG(ERROR) << "Only 3 or 4 channel 8-bit input image supported";
+    ABSL_LOG(ERROR) << "Only 3 or 4 channel 8-bit input image supported";
   }
 
   // Setup destination image
@@ -328,7 +328,7 @@ absl::Status SetAlphaCalculator::RenderGpu(CalculatorContext* cc) {
       cc->Inputs().Tag(kInputFrameTagGpu).Get<mediapipe::GpuBuffer>();
   if (!(input_frame.format() == mediapipe::GpuBufferFormat::kBGRA32 ||
         input_frame.format() == mediapipe::GpuBufferFormat::kRGB24)) {
-    LOG(ERROR) << "Only RGB or RGBA input image supported";
+    ABSL_LOG(ERROR) << "Only RGB or RGBA input image supported";
   }
   auto input_texture = gpu_helper_.CreateSourceTexture(input_frame);
 

mediapipe/calculators/image/testdata/BUILD

@@ -18,6 +18,7 @@ licenses(["notice"])
 filegroup(
     name = "test_images",
     srcs = [
+        "binary_mask.png",
         "dino.jpg",
         "dino_quality_50.jpg",
         "dino_quality_80.jpg",

mediapipe/calculators/internal/BUILD

@@ -31,12 +31,14 @@ mediapipe_proto_library(
 cc_library(
     name = "callback_packet_calculator",
     srcs = ["callback_packet_calculator.cc"],
+    hdrs = ["callback_packet_calculator.h"],
     visibility = ["//mediapipe/framework:__subpackages__"],
     deps = [
         ":callback_packet_calculator_cc_proto",
         "//mediapipe/framework:calculator_base",
         "//mediapipe/framework:calculator_registry",
         "//mediapipe/framework:output_side_packet",
+        "@com_google_absl//absl/status",
     ],
     alwayslink = 1,
 )

mediapipe/calculators/internal/callback_packet_calculator.cc

@@ -11,10 +11,12 @@
 // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 // See the License for the specific language governing permissions and
 // limitations under the License.
+#include "mediapipe/calculators/internal/callback_packet_calculator.h"
 
 #include <functional>
 #include <string>
 
+#include "absl/status/status.h"
 #include "mediapipe/calculators/internal/callback_packet_calculator.pb.h"  // NOLINT
 #include "mediapipe/framework/calculator_base.h"
 #include "mediapipe/framework/calculator_registry.h"
@@ -39,64 +41,55 @@ void DumpPostStreamPacket(Packet* post_stream_packet, const Packet& packet) {
     *post_stream_packet = packet;
   }
 }
+
 }  // namespace
 
-// Creates a callback which takes a packet and stores it either in a
+absl::Status CallbackPacketCalculator::GetContract(CalculatorContract* cc) {
-// vector of packets or stores only the packet at PostStream timestamp.
+  const auto& options = cc->Options<CallbackPacketCalculatorOptions>();
-// The kind of callback is controlled by an option.  The callback is
+  switch (options.type()) {
-// a std::function and is directly usable by CallbackCalculator.
+    case CallbackPacketCalculatorOptions::VECTOR_PACKET:
-// Since the options for the packet generator include a serialized pointer
+    case CallbackPacketCalculatorOptions::POST_STREAM_PACKET:
-// value, the resulting callback is only valid on the original machine
+      cc->OutputSidePackets()
-// while that pointer is still alive.
+          .Index(0)
-class CallbackPacketCalculator : public CalculatorBase {
+          .Set<std::function<void(const Packet&)>>();
- public:
+      break;
-  static absl::Status GetContract(CalculatorContract* cc) {
+    default:
-    const auto& options = cc->Options<CallbackPacketCalculatorOptions>();
-    switch (options.type()) {
-      case CallbackPacketCalculatorOptions::VECTOR_PACKET:
-      case CallbackPacketCalculatorOptions::POST_STREAM_PACKET:
-        cc->OutputSidePackets()
-            .Index(0)
-            .Set<std::function<void(const Packet&)>>();
-        break;
-      default:
-        return mediapipe::InvalidArgumentErrorBuilder(MEDIAPIPE_LOC)
-               << "Invalid type of callback to produce.";
-    }
-    return absl::OkStatus();
-  }
-
-  absl::Status Open(CalculatorContext* cc) override {
-    const auto& options = cc->Options<CallbackPacketCalculatorOptions>();
-    void* ptr;
-    if (sscanf(options.pointer().c_str(), "%p", &ptr) != 1) {
       return mediapipe::InvalidArgumentErrorBuilder(MEDIAPIPE_LOC)
-             << "Stored pointer value in options is invalid.";
+             << "Invalid type of callback to produce.";
-    }
-    switch (options.type()) {
-      case CallbackPacketCalculatorOptions::VECTOR_PACKET:
-        cc->OutputSidePackets().Index(0).Set(
-            MakePacket<std::function<void(const Packet&)>>(std::bind(
-                &DumpToVector, reinterpret_cast<std::vector<Packet>*>(ptr),
-                std::placeholders::_1)));
-        break;
-      case CallbackPacketCalculatorOptions::POST_STREAM_PACKET:
-        cc->OutputSidePackets().Index(0).Set(
-            MakePacket<std::function<void(const Packet&)>>(
-                std::bind(&DumpPostStreamPacket, reinterpret_cast<Packet*>(ptr),
-                          std::placeholders::_1)));
-        break;
-      default:
-        return mediapipe::InvalidArgumentErrorBuilder(MEDIAPIPE_LOC)
-               << "Invalid type to dump into.";
-    }
-    return absl::OkStatus();
   }
+  return absl::OkStatus();
+}
 
-  absl::Status Process(CalculatorContext* cc) override {
+absl::Status CallbackPacketCalculator::Open(CalculatorContext* cc) {
-    return absl::OkStatus();
+  const auto& options = cc->Options<CallbackPacketCalculatorOptions>();
+  void* ptr;
+  if (sscanf(options.pointer().c_str(), "%p", &ptr) != 1) {
+    return mediapipe::InvalidArgumentErrorBuilder(MEDIAPIPE_LOC)
+           << "Stored pointer value in options is invalid.";
   }
-};
+  switch (options.type()) {
+    case CallbackPacketCalculatorOptions::VECTOR_PACKET:
+      cc->OutputSidePackets().Index(0).Set(
+          MakePacket<std::function<void(const Packet&)>>(std::bind(
+              &DumpToVector, reinterpret_cast<std::vector<Packet>*>(ptr),
+              std::placeholders::_1)));
+      break;
+    case CallbackPacketCalculatorOptions::POST_STREAM_PACKET:
+      cc->OutputSidePackets().Index(0).Set(
+          MakePacket<std::function<void(const Packet&)>>(
+              std::bind(&DumpPostStreamPacket, reinterpret_cast<Packet*>(ptr),
+                        std::placeholders::_1)));
+      break;
+    default:
+      return mediapipe::InvalidArgumentErrorBuilder(MEDIAPIPE_LOC)
+             << "Invalid type to dump into.";
+  }
+  return absl::OkStatus();
+}
+
+absl::Status CallbackPacketCalculator::Process(CalculatorContext* cc) {
+  return absl::OkStatus();
+}
 
 REGISTER_CALCULATOR(CallbackPacketCalculator);
 

mediapipe/calculators/internal/callback_packet_calculator.h

@@ -0,0 +1,39 @@
+// Copyright 2023 The MediaPipe Authors.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//      http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#ifndef MEDIAPIPE_CALCULATORS_INTERNAL_CALLBACK_PACKET_CALCULATOR_H_
+#define MEDIAPIPE_CALCULATORS_INTERNAL_CALLBACK_PACKET_CALCULATOR_H_
+
+#include "absl/status/status.h"
+#include "mediapipe/framework/calculator_base.h"
+
+namespace mediapipe {
+
+// Creates a callback which takes a packet and stores it either in a
+// vector of packets or stores only the packet at PostStream timestamp.
+// The kind of callback is controlled by an option.  The callback is
+// a std::function and is directly usable by CallbackCalculator.
+// Since the options for the packet generator include a serialized pointer
+// value, the resulting callback is only valid on the original machine
+// while that pointer is still alive.
+class CallbackPacketCalculator : public CalculatorBase {
+ public:
+  static absl::Status GetContract(CalculatorContract* cc);
+  absl::Status Open(CalculatorContext* cc) override;
+  absl::Status Process(CalculatorContext* cc) override;
+};
+
+}  // namespace mediapipe
+
+#endif  // MEDIAPIPE_CALCULATORS_INTERNAL_CALLBACK_PACKET_CALCULATOR_H_

mediapipe/calculators/tensor/BUILD

@@ -87,6 +87,7 @@ cc_library(
         "//mediapipe/framework/formats:time_series_header_cc_proto",
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/util:time_series_util",
+        "@com_google_absl//absl/log:absl_check",
         "@com_google_absl//absl/memory",
         "@com_google_absl//absl/status",
         "@com_google_absl//absl/status:statusor",
@@ -181,6 +182,7 @@ cc_library(
         "//mediapipe/framework:calculator_framework",
         "//mediapipe/framework/api2:node",
         "//mediapipe/framework/formats:tensor",
+        "@com_google_absl//absl/log:absl_check",
         "@com_google_absl//absl/status",
     ],
     alwayslink = 1,
@@ -198,6 +200,7 @@ cc_test(
         "//mediapipe/framework/formats:tensor",
         "//mediapipe/framework/port:gtest_main",
         "//mediapipe/framework/port:parse_text_proto",
+        "@com_google_absl//absl/log:absl_check",
         "@org_tensorflow//tensorflow/lite/c:common",
     ],
 )
@@ -445,6 +448,7 @@ cc_library(
         "//mediapipe/framework/deps:file_path",
         "//mediapipe/gpu:gl_calculator_helper",
         "//mediapipe/util/tflite:tflite_gpu_runner",
+        "@com_google_absl//absl/log:absl_log",
         "@com_google_absl//absl/memory",
         "@com_google_absl//absl/status",
         "@com_google_absl//absl/status:statusor",
@@ -474,6 +478,7 @@ cc_library(
         "//mediapipe/gpu:gpu_buffer",
         "//mediapipe/objc:mediapipe_framework_ios",
         "//mediapipe/util/tflite:config",
+        "@com_google_absl//absl/log:absl_log",
         "@com_google_absl//absl/memory",
         "@com_google_absl//absl/strings:str_format",
         "@org_tensorflow//tensorflow/lite/delegates/gpu:metal_delegate",
@@ -620,6 +625,7 @@ mediapipe_proto_library(
     deps = [
         "//mediapipe/framework:calculator_options_proto",
         "//mediapipe/framework:calculator_proto",
+        "//mediapipe/gpu:gpu_origin_proto",
     ],
 )
 
@@ -649,7 +655,18 @@ cc_library(
         "//mediapipe/framework/formats:matrix",
         "//mediapipe/framework/formats:tensor",
         "//mediapipe/framework/port:ret_check",
+        "//mediapipe/framework/port:status",
+        "//mediapipe/framework/port:statusor",
+        "//mediapipe/gpu:gpu_buffer_format",
+        "//mediapipe/gpu:gpu_origin_cc_proto",
         "//mediapipe/util:resource_util",
+        "@com_google_absl//absl/log",
+        "@com_google_absl//absl/log:absl_check",
+        "@com_google_absl//absl/log:absl_log",
+        "@com_google_absl//absl/log:check",
+        "@com_google_absl//absl/status",
+        "@com_google_absl//absl/status:statusor",
+        "@com_google_absl//absl/strings:str_format",
     ] + select({
         "//mediapipe/gpu:disable_gpu": [],
         "//conditions:default": ["tensor_converter_calculator_gpu_deps"],
@@ -699,9 +716,11 @@ cc_test(
         "//mediapipe/framework/formats:tensor",
         "//mediapipe/framework/port:gtest_main",
         "//mediapipe/framework/port:integral_types",
+        "//mediapipe/framework/port:opencv_core",
         "//mediapipe/framework/port:parse_text_proto",
         "//mediapipe/framework/tool:validate_type",
         "@com_google_absl//absl/memory",
+        "@com_google_absl//absl/status",
         "@com_google_absl//absl/strings",
     ],
 )
@@ -737,6 +756,8 @@ cc_library(
         "//mediapipe/framework/formats:tensor",
         "//mediapipe/framework/formats/object_detection:anchor_cc_proto",
         "//mediapipe/framework/port:ret_check",
+        "@com_google_absl//absl/log:absl_check",
+        "@com_google_absl//absl/log:absl_log",
         "@com_google_absl//absl/strings:str_format",
         "@com_google_absl//absl/types:span",
     ] + selects.with_or({
@@ -793,6 +814,7 @@ cc_library(
         "//mediapipe/framework/formats:landmark_cc_proto",
         "//mediapipe/framework/formats:tensor",
         "//mediapipe/framework/port:ret_check",
+        "@com_google_absl//absl/log:absl_check",
     ],
     alwayslink = 1,
 )
@@ -985,6 +1007,8 @@ cc_library(
         "//mediapipe/framework/port:status",
         "//mediapipe/framework/port:statusor",
         "//mediapipe/gpu:gpu_origin_cc_proto",
+        "@com_google_absl//absl/log:absl_check",
+        "@com_google_absl//absl/log:absl_log",
     ] + select({
         "//mediapipe/gpu:disable_gpu": [],
         "//conditions:default": [":image_to_tensor_calculator_gpu_deps"],
@@ -1077,6 +1101,7 @@ cc_test(
         "//mediapipe/framework/port:parse_text_proto",
         "//mediapipe/util:image_test_utils",
         "@com_google_absl//absl/flags:flag",
+        "@com_google_absl//absl/log:absl_check",
         "@com_google_absl//absl/memory",
         "@com_google_absl//absl/strings",
         "@com_google_absl//absl/strings:str_format",
@@ -1204,6 +1229,7 @@ cc_library(
             "//mediapipe/gpu:gl_calculator_helper",
             "//mediapipe/gpu:gl_simple_shaders",
             "//mediapipe/gpu:shader_util",
+            "@com_google_absl//absl/log:absl_log",
             "@com_google_absl//absl/strings",
         ],
     }),

mediapipe/calculators/tensor/audio_to_tensor_calculator.cc

@@ -20,6 +20,7 @@
 #include <utility>
 #include <vector>
 
+#include "absl/log/absl_check.h"
 #include "absl/memory/memory.h"
 #include "absl/status/status.h"
 #include "absl/status/statusor.h"
@@ -348,7 +349,7 @@ absl::Status AudioToTensorCalculator::Process(CalculatorContext* cc) {
     return absl::InvalidArgumentError(
         "The audio data should be stored in column-major.");
   }
-  CHECK(channels_match || mono_output);
+  ABSL_CHECK(channels_match || mono_output);
   const Matrix& input = channels_match ? input_frame
                                        // Mono mixdown.
                                        : input_frame.colwise().mean();
@@ -457,7 +458,7 @@ absl::Status AudioToTensorCalculator::SetupStreamingResampler(
 }
 
 void AudioToTensorCalculator::AppendZerosToSampleBuffer(int num_samples) {
-  CHECK_GE(num_samples, 0);  // Ensured by `UpdateContract`.
+  ABSL_CHECK_GE(num_samples, 0);  // Ensured by `UpdateContract`.
   if (num_samples == 0) {
     return;
   }

mediapipe/calculators/tensor/feedback_tensors_calculator_test.cc

@@ -18,6 +18,7 @@
 #include <utility>
 #include <vector>
 
+#include "absl/log/absl_check.h"
 #include "mediapipe/calculators/tensor/feedback_tensors_calculator.pb.h"
 #include "mediapipe/framework/calculator.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
@@ -65,7 +66,7 @@ template <typename T>
 Tensor MakeTensor(std::initializer_list<int> shape,
                   std::initializer_list<T> values) {
   Tensor tensor(TensorElementType<T>::value, shape);
-  CHECK_EQ(values.size(), tensor.shape().num_elements())
+  ABSL_CHECK_EQ(values.size(), tensor.shape().num_elements())
       << "The size of `values` is incompatible with `shape`";
   absl::c_copy(values, tensor.GetCpuWriteView().buffer<T>());
   return tensor;

mediapipe/calculators/tensor/image_to_tensor_calculator.cc

@@ -16,6 +16,7 @@
 #include <memory>
 #include <vector>
 
+#include "absl/log/absl_log.h"
 #include "mediapipe/calculators/tensor/image_to_tensor_calculator.pb.h"
 #include "mediapipe/calculators/tensor/image_to_tensor_converter.h"
 #include "mediapipe/calculators/tensor/image_to_tensor_utils.h"
@@ -284,9 +285,9 @@ class ImageToTensorCalculator : public Node {
                              cc, GetBorderMode(options_.border_mode()),
                              GetOutputTensorType(/*uses_gpu=*/false, params_)));
 #else
-        LOG(FATAL) << "Cannot create image to tensor CPU converter since "
+        ABSL_LOG(FATAL) << "Cannot create image to tensor CPU converter since "
-                      "MEDIAPIPE_DISABLE_OPENCV is defined and "
+                           "MEDIAPIPE_DISABLE_OPENCV is defined and "
-                      "MEDIAPIPE_ENABLE_HALIDE is not defined.";
+                           "MEDIAPIPE_ENABLE_HALIDE is not defined.";
 #endif  // !MEDIAPIPE_DISABLE_HALIDE
       }
     }

mediapipe/calculators/tensor/image_to_tensor_calculator_test.cc

@@ -18,6 +18,7 @@
 #include <vector>
 
 #include "absl/flags/flag.h"
+#include "absl/log/absl_check.h"
 #include "absl/memory/memory.h"
 #include "absl/strings/str_format.h"
 #include "absl/strings/substitute.h"
@@ -205,7 +206,7 @@ mediapipe::ImageFormat::Format GetImageFormat(int image_channels) {
   } else if (image_channels == 1) {
     return ImageFormat::GRAY8;
   }
-  CHECK(false) << "Unsupported input image channles: " << image_channels;
+  ABSL_CHECK(false) << "Unsupported input image channles: " << image_channels;
 }
 
 Packet MakeImageFramePacket(cv::Mat input) {

mediapipe/calculators/tensor/image_to_tensor_converter_gl_texture.cc

@@ -22,6 +22,7 @@
 #include <memory>
 #include <vector>
 
+#include "absl/log/absl_log.h"
 #include "absl/strings/str_cat.h"
 #include "mediapipe/calculators/tensor/image_to_tensor_converter.h"
 #include "mediapipe/calculators/tensor/image_to_tensor_converter_gl_utils.h"
@@ -259,7 +260,7 @@ class GlProcessor : public ImageToTensorConverter {
     // error. So in that case, we'll grab the transpose of our original matrix
     // and send that instead.
     const auto gl_context = mediapipe::GlContext::GetCurrent();
-    LOG_IF(FATAL, !gl_context) << "GlContext is not bound to the thread.";
+    ABSL_LOG_IF(FATAL, !gl_context) << "GlContext is not bound to the thread.";
     if (gl_context->GetGlVersion() == mediapipe::GlVersion::kGLES2) {
       GetTransposedRotatedSubRectToRectTransformMatrix(
           sub_rect, texture.width(), texture.height(), flip_horizontaly,

mediapipe/calculators/tensor/inference_calculator.proto

@@ -88,6 +88,20 @@ message InferenceCalculatorOptions {
       // serialized model is invalid or missing.
       optional string serialized_model_dir = 7;
 
+      enum CacheWritingBehavior {
+        // Do not write any caches.
+        NO_WRITE = 0;
+
+        // Try to write caches, log on failure.
+        TRY_WRITE = 1;
+
+        // Write caches or return an error if write fails.
+        WRITE_OR_ERROR = 2;
+      }
+      // Specifies how GPU caches are written to disk.
+      optional CacheWritingBehavior cache_writing_behavior = 10
+          [default = WRITE_OR_ERROR];
+
       // Unique token identifying the model. Used in conjunction with
       // "serialized_model_dir". It is the caller's responsibility to ensure
       // there is no clash of the tokens.

mediapipe/calculators/tensor/inference_calculator_gl_advanced.cc

@@ -12,6 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+#include <cstdint>
 #include <cstring>
 #include <memory>
 #include <string>
@@ -26,6 +27,7 @@
 #include "mediapipe/util/tflite/tflite_gpu_runner.h"
 
 #if defined(MEDIAPIPE_ANDROID) || defined(MEDIAPIPE_CHROMIUMOS)
+#include "absl/log/absl_log.h"
 #include "mediapipe/framework/deps/file_path.h"
 #include "mediapipe/util/android/file/base/file.h"
 #include "mediapipe/util/android/file/base/filesystem.h"
@@ -68,13 +70,21 @@ class InferenceCalculatorGlAdvancedImpl
         const mediapipe::InferenceCalculatorOptions::Delegate::Gpu&
             gpu_delegate_options);
     absl::Status ReadGpuCaches(tflite::gpu::TFLiteGPURunner* gpu_runner) const;
-    absl::Status SaveGpuCaches(tflite::gpu::TFLiteGPURunner* gpu_runner) const;
+    // Writes caches to disk based on |cache_writing_behavior_|.
+    absl::Status SaveGpuCachesBasedOnBehavior(
+        tflite::gpu::TFLiteGPURunner* gpu_runner) const;
+    bool UseSerializedModel() const { return use_serialized_model_; }
 
    private:
+    // Writes caches to disk, returns error on failure.
+    absl::Status SaveGpuCaches(tflite::gpu::TFLiteGPURunner* gpu_runner) const;
+
     bool use_kernel_caching_ = false;
     std::string cached_kernel_filename_;
     bool use_serialized_model_ = false;
     std::string serialized_model_path_;
+    mediapipe::InferenceCalculatorOptions::Delegate::Gpu::CacheWritingBehavior
+        cache_writing_behavior_;
   };
 
   // Helper class that wraps everything related to GPU inference acceleration.
@@ -150,8 +160,6 @@ InferenceCalculatorGlAdvancedImpl::GpuInferenceRunner::Process(
 }
 
 absl::Status InferenceCalculatorGlAdvancedImpl::GpuInferenceRunner::Close() {
-  MP_RETURN_IF_ERROR(
-      on_disk_cache_helper_.SaveGpuCaches(tflite_gpu_runner_.get()));
   return gpu_helper_.RunInGlContext([this]() -> absl::Status {
     tflite_gpu_runner_.reset();
     return absl::OkStatus();
@@ -226,9 +234,15 @@ InferenceCalculatorGlAdvancedImpl::GpuInferenceRunner::InitTFLiteGPURunner(
                          tflite_gpu_runner_->GetOutputShapes()[i].c};
   }
 
+  if (on_disk_cache_helper_.UseSerializedModel()) {
+    tflite_gpu_runner_->ForceOpenCLInitFromSerializedModel();
+  }
+
   MP_RETURN_IF_ERROR(
       on_disk_cache_helper_.ReadGpuCaches(tflite_gpu_runner_.get()));
-  return tflite_gpu_runner_->Build();
+  MP_RETURN_IF_ERROR(tflite_gpu_runner_->Build());
+  return on_disk_cache_helper_.SaveGpuCachesBasedOnBehavior(
+      tflite_gpu_runner_.get());
 }
 
 #if defined(MEDIAPIPE_ANDROID) || defined(MEDIAPIPE_CHROMIUMOS)
@@ -257,9 +271,36 @@ absl::Status InferenceCalculatorGlAdvancedImpl::OnDiskCacheHelper::Init(
         mediapipe::file::JoinPath(gpu_delegate_options.serialized_model_dir(),
                                   gpu_delegate_options.model_token());
   }
+  cache_writing_behavior_ = gpu_delegate_options.has_cache_writing_behavior()
+                                ? gpu_delegate_options.cache_writing_behavior()
+                                : mediapipe::InferenceCalculatorOptions::
+                                      Delegate::Gpu::WRITE_OR_ERROR;
   return absl::OkStatus();
 }
 
+absl::Status InferenceCalculatorGlAdvancedImpl::OnDiskCacheHelper::
+    SaveGpuCachesBasedOnBehavior(
+        tflite::gpu::TFLiteGPURunner* gpu_runner) const {
+  switch (cache_writing_behavior_) {
+    case mediapipe::InferenceCalculatorOptions::Delegate::Gpu::NO_WRITE:
+      return absl::OkStatus();
+    case mediapipe::InferenceCalculatorOptions::Delegate::Gpu::TRY_WRITE: {
+      auto status = SaveGpuCaches(gpu_runner);
+      if (!status.ok()) {
+        ABSL_LOG_FIRST_N(WARNING, 1) << "Failed to save gpu caches: " << status;
+      }
+      return absl::OkStatus();
+    }
+    case mediapipe::InferenceCalculatorOptions::Delegate::Gpu::WRITE_OR_ERROR:
+      return SaveGpuCaches(gpu_runner);
+    default:
+      ABSL_LOG_FIRST_N(ERROR, 1)
+          << "Unknown cache writing behavior: "
+          << static_cast<uint32_t>(cache_writing_behavior_);
+      return absl::InvalidArgumentError("Unknown cache writing behavior.");
+  }
+}
+
 absl::Status
 InferenceCalculatorGlAdvancedImpl::OnDiskCacheHelper::SaveGpuCaches(
     tflite::gpu::TFLiteGPURunner* gpu_runner) const {
@@ -314,6 +355,12 @@ absl::Status InferenceCalculatorGlAdvancedImpl::OnDiskCacheHelper::Init(
   return absl::OkStatus();
 }
 
+absl::Status InferenceCalculatorGlAdvancedImpl::OnDiskCacheHelper::
+    SaveGpuCachesBasedOnBehavior(
+        tflite::gpu::TFLiteGPURunner* gpu_runner) const {
+  return absl::OkStatus();
+}
+
 absl::Status
 InferenceCalculatorGlAdvancedImpl::OnDiskCacheHelper::ReadGpuCaches(
     tflite::gpu::TFLiteGPURunner* gpu_runner) const {

mediapipe/calculators/tensor/inference_calculator_metal.cc

@@ -21,6 +21,7 @@
 #include <string>
 #include <vector>
 
+#include "absl/log/absl_log.h"
 #include "absl/memory/memory.h"
 #include "absl/strings/str_format.h"
 #include "mediapipe/calculators/tensor/inference_calculator.h"
@@ -74,7 +75,7 @@ tflite::gpu::BHWC BhwcFromTensorShape(const Tensor::Shape& shape) {
       break;
     default:
       // Handles 0 and >4.
-      LOG(FATAL)
+      ABSL_LOG(FATAL)
           << "Dimensions size must be in range [1,4] for GPU inference, but "
           << shape.dims.size() << " is provided";
   }

mediapipe/calculators/tensor/inference_calculator_test.cc

@@ -16,7 +16,7 @@
 #include <string>
 #include <vector>
 
-#include "absl/log/check.h"
+#include "absl/log/absl_check.h"
 #include "absl/strings/str_cat.h"
 #include "absl/strings/str_replace.h"
 #include "absl/strings/string_view.h"

mediapipe/calculators/tensor/tensor_converter_calculator.cc

@@ -12,9 +12,15 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+#include <cstdint>
 #include <string>
 #include <vector>
 
+#include "absl/log/absl_check.h"
+#include "absl/log/absl_log.h"
+#include "absl/status/status.h"
+#include "absl/status/statusor.h"
+#include "absl/strings/str_format.h"
 #include "mediapipe/calculators/tensor/tensor_converter_calculator.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/formats/image_frame.h"
@@ -22,7 +28,8 @@
 #include "mediapipe/framework/formats/tensor.h"
 #include "mediapipe/framework/port.h"
 #include "mediapipe/framework/port/ret_check.h"
-#include "mediapipe/util/resource_util.h"
+#include "mediapipe/gpu/gpu_buffer_format.h"
+#include "mediapipe/gpu/gpu_origin.pb.h"
 
 #if !MEDIAPIPE_DISABLE_GPU
 #include "mediapipe/gpu/gpu_buffer.h"
@@ -43,12 +50,50 @@
 #endif  // !MEDIAPIPE_DISABLE_GPU
 
 namespace {
+
 constexpr int kWorkgroupSize = 8;  // Block size for GPU shader.
 // Commonly used to compute the number of blocks to launch in a kernel.
 int NumGroups(const int size, const int group_size) {  // NOLINT
   return (size + group_size - 1) / group_size;
 }
 
+absl::StatusOr<bool> ShouldFlipVertically(
+    const mediapipe::TensorConverterCalculatorOptions& options, bool use_gpu) {
+  if (options.has_flip_vertically() && options.has_gpu_origin()) {
+    return absl::FailedPreconditionError(absl::StrFormat(
+        "Cannot specify both flip_vertically and gpu_origin options"));
+  }
+
+  if (!options.has_gpu_origin()) {
+    // Fall back to flip_vertically.
+    return options.flip_vertically();
+  }
+
+  // Warn if gpu_origin is specified with a CPU input image.
+  // Those are always TOP_LEFT, so no flipping is necessary.
+  if (!use_gpu) {
+    ABSL_LOG(WARNING)
+        << "Ignoring gpu_origin option since IMAGE_GPU input is not specified";
+    return false;
+  }
+
+  switch (options.gpu_origin()) {
+    case mediapipe::GpuOrigin::TOP_LEFT:
+      return false;
+    case mediapipe::GpuOrigin::DEFAULT:
+    case mediapipe::GpuOrigin::CONVENTIONAL:
+      // TOP_LEFT on Metal, BOTTOM_LEFT on OpenGL.
+#ifdef __APPLE__
+      return false;
+#else
+      return true;
+#endif
+    default:
+      return absl::InvalidArgumentError(
+          absl::StrFormat("Unhandled GPU origin %i", options.gpu_origin()));
+  }
+}
+
 typedef Eigen::Matrix<float, Eigen::Dynamic, Eigen::Dynamic, Eigen::RowMajor>
     RowMajorMatrixXf;
 typedef Eigen::Matrix<float, Eigen::Dynamic, Eigen::Dynamic, Eigen::ColMajor>
@@ -58,6 +103,7 @@ constexpr char kImageFrameTag[] = "IMAGE";
 constexpr char kGpuBufferTag[] = "IMAGE_GPU";
 constexpr char kTensorsTag[] = "TENSORS";
 constexpr char kMatrixTag[] = "MATRIX";
+
 }  // namespace
 
 namespace mediapipe {
@@ -109,7 +155,7 @@ class TensorConverterCalculator : public CalculatorBase {
 
  private:
   absl::Status InitGpu(CalculatorContext* cc);
-  absl::Status LoadOptions(CalculatorContext* cc);
+  absl::Status LoadOptions(CalculatorContext* cc, bool use_gpu);
   template <class T>
   absl::Status NormalizeImage(const ImageFrame& image_frame,
                               bool flip_vertically, float* tensor_ptr);
@@ -145,7 +191,8 @@ absl::Status TensorConverterCalculator::GetContract(CalculatorContract* cc) {
   RET_CHECK(static_cast<int>(cc->Inputs().HasTag(kImageFrameTag)) +
                 static_cast<int>(cc->Inputs().HasTag(kGpuBufferTag)) +
                 static_cast<int>(cc->Inputs().HasTag(kMatrixTag)) ==
-            1);
+            1)
+      << "Only one input tag of {IMAGE, IMAGE_GPU, MATRIX} may be specified";
 
   if (cc->Inputs().HasTag(kImageFrameTag)) {
     cc->Inputs().Tag(kImageFrameTag).Set<ImageFrame>();
@@ -173,8 +220,6 @@ absl::Status TensorConverterCalculator::GetContract(CalculatorContract* cc) {
 absl::Status TensorConverterCalculator::Open(CalculatorContext* cc) {
   cc->SetOffset(TimestampDiff(0));
 
-  MP_RETURN_IF_ERROR(LoadOptions(cc));
-
 #if !MEDIAPIPE_DISABLE_GPU
   if (cc->Inputs().HasTag(kGpuBufferTag)) {
     use_gpu_ = true;
@@ -187,6 +232,8 @@ absl::Status TensorConverterCalculator::Open(CalculatorContext* cc) {
   }
 #endif  // !MEDIAPIPE_DISABLE_GPU
 
+  MP_RETURN_IF_ERROR(LoadOptions(cc, use_gpu_));
+
   return absl::OkStatus();
 }
 
@@ -378,23 +425,34 @@ absl::Status TensorConverterCalculator::InitGpu(CalculatorContext* cc) {
   // Get input image sizes.
   const auto& input =
       cc->Inputs().Tag(kGpuBufferTag).Get<mediapipe::GpuBuffer>();
-  mediapipe::ImageFormat::Format format =
+  mediapipe::GpuBufferFormat format = input.format();
-      mediapipe::ImageFormatForGpuBufferFormat(input.format());
   const bool include_alpha = (max_num_channels_ == 4);
   const bool single_channel = (max_num_channels_ == 1);
-  if (!(format == mediapipe::ImageFormat::GRAY8 ||
+
-        format == mediapipe::ImageFormat::SRGB ||
+  RET_CHECK(format == mediapipe::GpuBufferFormat::kBGRA32 ||
-        format == mediapipe::ImageFormat::SRGBA))
+            format == mediapipe::GpuBufferFormat::kRGB24 ||
-    RET_CHECK_FAIL() << "Unsupported GPU input format.";
+            format == mediapipe::GpuBufferFormat::kRGBA32 ||
-  if (include_alpha && (format != mediapipe::ImageFormat::SRGBA))
+            format == mediapipe::GpuBufferFormat::kRGBAFloat128 ||
-    RET_CHECK_FAIL() << "Num input channels is less than desired output.";
+            format == mediapipe::GpuBufferFormat::kRGBAHalf64 ||
+            format == mediapipe::GpuBufferFormat::kGrayFloat32 ||
+            format == mediapipe::GpuBufferFormat::kGrayHalf16 ||
+            format == mediapipe::GpuBufferFormat::kOneComponent8)
+      << "Unsupported GPU input format: " << static_cast<uint32_t>(format);
+  if (include_alpha) {
+    RET_CHECK(format == mediapipe::GpuBufferFormat::kBGRA32 ||
+              format == mediapipe::GpuBufferFormat::kRGBA32 ||
+              format == mediapipe::GpuBufferFormat::kRGBAFloat128 ||
+              format == mediapipe::GpuBufferFormat::kRGBAHalf64)
+        << "Num input channels is less than desired output, input format: "
+        << static_cast<uint32_t>(format);
+  }
 
 #if MEDIAPIPE_METAL_ENABLED
   id<MTLDevice> device = gpu_helper_.mtlDevice;
   // Shader to convert GL Texture to Metal Buffer,
   // with normalization to either: [0,1] or [-1,1].
   const std::string shader_source = absl::Substitute(
-      R"(
+      R"glsl(
   #include <metal_stdlib>
 
   using namespace metal;
@@ -413,7 +471,7 @@ absl::Status TensorConverterCalculator::InitGpu(CalculatorContext* cc) {
     $3  // g & b channels
     $4  // alpha channel
   }
-      )",
+      )glsl",
       /*$0=*/
       output_range_.has_value()
           ? absl::Substitute("pixel = pixel * half($0) + half($1);",
@@ -423,8 +481,8 @@ absl::Status TensorConverterCalculator::InitGpu(CalculatorContext* cc) {
       /*$1=*/max_num_channels_,
       /*$2=*/flip_vertically_ ? "(in_tex.get_height() - 1 - gid.y)" : "gid.y",
       /*$3=*/
-      single_channel ? "" : R"(out_buf[linear_index + 1] = pixel.y;
+      single_channel ? "" : R"glsl(out_buf[linear_index + 1] = pixel.y;
-               out_buf[linear_index + 2] = pixel.z;)",
+                                   out_buf[linear_index + 2] = pixel.z;)glsl",
       /*$4=*/include_alpha ? "out_buf[linear_index + 3] = pixel.w;" : "");
 
   NSString* library_source =
@@ -442,17 +500,17 @@ absl::Status TensorConverterCalculator::InitGpu(CalculatorContext* cc) {
   RET_CHECK(to_buffer_program_ != nil) << "Couldn't create pipeline state " <<
       [[error localizedDescription] UTF8String];
 #elif MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_30
-  MP_RETURN_IF_ERROR(gpu_helper_.RunInGlContext([this, &include_alpha,
+  MP_RETURN_IF_ERROR(
+      gpu_helper_.RunInGlContext([this, &include_alpha,
 #if MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_31
-                                                 &input,
+                                  &input,
 #endif  // MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_31
-                                                 &single_channel]()
+                                  &single_channel]() -> absl::Status {
-                                                    -> absl::Status {
 #if MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_31
-    // Shader to convert GL Texture to Shader Storage Buffer Object (SSBO),
+        // Shader to convert GL Texture to Shader Storage Buffer Object (SSBO),
-    // with normalization to either: [0,1] or [-1,1].
+        // with normalization to either: [0,1] or [-1,1].
-    const std::string shader_source = absl::Substitute(
+        const std::string shader_source = absl::Substitute(
-        R"( #version 310 es
+            R"glsl( #version 310 es
           layout(local_size_x = $0, local_size_y = $0) in;
           layout(binding = 0) uniform sampler2D input_texture;
           layout(std430, binding = 1) buffer Output {float elements[];} output_data;
@@ -466,38 +524,40 @@ absl::Status TensorConverterCalculator::InitGpu(CalculatorContext* cc) {
             output_data.elements[linear_index + 0] = pixel.x;  // r channel
             $5  // g & b channels
             $6  // alpha channel
-          })",
+          })glsl",
-        /*$0=*/kWorkgroupSize, /*$1=*/input.width(), /*$2=*/input.height(),
+            /*$0=*/kWorkgroupSize, /*$1=*/input.width(), /*$2=*/input.height(),
-        /*$3=*/
+            /*$3=*/
-        output_range_.has_value()
+            output_range_.has_value()
-            ? absl::Substitute("pixel = pixel * float($0) + float($1);",
+                ? absl::Substitute(
-                               (output_range_->second - output_range_->first),
+                      "pixel = pixel * float($0) + float($1);",
-                               output_range_->first)
+                      (output_range_->second - output_range_->first),
-            : "",
+                      output_range_->first)
-        /*$4=*/flip_vertically_ ? "(width_height.y - 1 - gid.y)" : "gid.y",
+                : "",
-        /*$5=*/
+            /*$4=*/flip_vertically_ ? "(width_height.y - 1 - gid.y)" : "gid.y",
-        single_channel ? ""
+            /*$5=*/
-                       : R"(output_data.elements[linear_index + 1] = pixel.y;
+            single_channel
-                            output_data.elements[linear_index + 2] = pixel.z;)",
+                ? ""
-        /*$6=*/
+                : R"glsl(output_data.elements[linear_index + 1] = pixel.y;
-        include_alpha ? "output_data.elements[linear_index + 3] = pixel.w;"
+                     output_data.elements[linear_index + 2] = pixel.z;)glsl",
-                      : "",
+            /*$6=*/
-        /*$7=*/max_num_channels_);
+            include_alpha ? "output_data.elements[linear_index + 3] = pixel.w;"
-    GLuint shader = glCreateShader(GL_COMPUTE_SHADER);
+                          : "",
-    const GLchar* sources[] = {shader_source.c_str()};
+            /*$7=*/max_num_channels_);
-    glShaderSource(shader, 1, sources, NULL);
+        GLuint shader = glCreateShader(GL_COMPUTE_SHADER);
-    glCompileShader(shader);
+        const GLchar* sources[] = {shader_source.c_str()};
-    GLint compiled = GL_FALSE;
+        glShaderSource(shader, 1, sources, NULL);
-    glGetShaderiv(shader, GL_COMPILE_STATUS, &compiled);
+        glCompileShader(shader);
-    RET_CHECK(compiled == GL_TRUE);
+        GLint compiled = GL_FALSE;
-    to_buffer_program_ = glCreateProgram();
+        glGetShaderiv(shader, GL_COMPILE_STATUS, &compiled);
-    glAttachShader(to_buffer_program_, shader);
+        RET_CHECK(compiled == GL_TRUE);
-    glDeleteShader(shader);
+        to_buffer_program_ = glCreateProgram();
-    glLinkProgram(to_buffer_program_);
+        glAttachShader(to_buffer_program_, shader);
+        glDeleteShader(shader);
+        glLinkProgram(to_buffer_program_);
 #else
-    // OpenGL ES 3.0 fragment shader Texture2d -> Texture2d conversion.
+        // OpenGL ES 3.0 fragment shader Texture2d -> Texture2d conversion.
-    const std::string shader_source = absl::Substitute(
+        const std::string shader_source = absl::Substitute(
-        R"(
+            R"glsl(
         #if __VERSION__ < 130
           #define in varying
         #endif  // __VERSION__ < 130
@@ -523,49 +583,51 @@ absl::Status TensorConverterCalculator::InitGpu(CalculatorContext* cc) {
             fragColor.r = pixel.r;  // r channel
             $3  // g & b channels
             $4  // alpha channel
-          })",
+          })glsl",
-        /*$0=*/single_channel ? "vec1" : "vec4",
+            /*$0=*/single_channel ? "vec1" : "vec4",
-        /*$1=*/
+            /*$1=*/
-        flip_vertically_
+            flip_vertically_
-            ? "vec2(sample_coordinate.x, 1.0 - sample_coordinate.y);"
+                ? "vec2(sample_coordinate.x, 1.0 - sample_coordinate.y);"
-            : "sample_coordinate;",
+                : "sample_coordinate;",
-        /*$2=*/output_range_.has_value()
+            /*$2=*/output_range_.has_value()
-            ? absl::Substitute("pixel = pixel * float($0) + float($1);",
+                ? absl::Substitute(
-                               (output_range_->second - output_range_->first),
+                      "pixel = pixel * float($0) + float($1);",
-                               output_range_->first)
+                      (output_range_->second - output_range_->first),
-            : "",
+                      output_range_->first)
-        /*$3=*/single_channel ? "" : R"(fragColor.g = pixel.g;
+                : "",
-                            fragColor.b = pixel.b;)",
+            /*$3=*/single_channel ? "" : R"glsl(fragColor.g = pixel.g;
-        /*$4=*/
+                                            fragColor.b = pixel.b;)glsl",
-        include_alpha ? "fragColor.a = pixel.a;"
+            /*$4=*/
-                      : (single_channel ? "" : "fragColor.a = 1.0;"));
+            include_alpha ? "fragColor.a = pixel.a;"
+                          : (single_channel ? "" : "fragColor.a = 1.0;"));
 
-    const GLint attr_location[NUM_ATTRIBUTES] = {
+        const GLint attr_location[NUM_ATTRIBUTES] = {
-        ATTRIB_VERTEX,
+            ATTRIB_VERTEX,
-        ATTRIB_TEXTURE_POSITION,
+            ATTRIB_TEXTURE_POSITION,
-    };
+        };
-    const GLchar* attr_name[NUM_ATTRIBUTES] = {
+        const GLchar* attr_name[NUM_ATTRIBUTES] = {
-        "position",
+            "position",
-        "texture_coordinate",
+            "texture_coordinate",
-    };
+        };
-    // shader program and params
+        // shader program and params
-    mediapipe::GlhCreateProgram(
+        mediapipe::GlhCreateProgram(
-        mediapipe::kBasicVertexShader, shader_source.c_str(), NUM_ATTRIBUTES,
+            mediapipe::kBasicVertexShader, shader_source.c_str(),
-        &attr_name[0], attr_location, &to_tex2d_program_);
+            NUM_ATTRIBUTES, &attr_name[0], attr_location, &to_tex2d_program_);
-    RET_CHECK(to_tex2d_program_) << "Problem initializing the program.";
+        RET_CHECK(to_tex2d_program_) << "Problem initializing the program.";
-    glUseProgram(to_tex2d_program_);
+        glUseProgram(to_tex2d_program_);
-    glUniform1i(glGetUniformLocation(to_tex2d_program_, "frame"), 1);
+        glUniform1i(glGetUniformLocation(to_tex2d_program_, "frame"), 1);
-    glGenFramebuffers(1, &framebuffer_);
+        glGenFramebuffers(1, &framebuffer_);
 
 #endif  // MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_31
-    return absl::OkStatus();
+        return absl::OkStatus();
-  }));
+      }));
 #endif  // MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_30
 #endif  // !MEDIAPIPE_DISABLE_GPU
   return absl::OkStatus();
 }
 
-absl::Status TensorConverterCalculator::LoadOptions(CalculatorContext* cc) {
+absl::Status TensorConverterCalculator::LoadOptions(CalculatorContext* cc,
+                                                    bool use_gpu) {
   // Get calculator options specified in the graph.
   const auto& options =
       cc->Options<::mediapipe::TensorConverterCalculatorOptions>();
@@ -582,7 +644,7 @@ absl::Status TensorConverterCalculator::LoadOptions(CalculatorContext* cc) {
   if (options.has_output_tensor_float_range()) {
     output_range_.emplace(options.output_tensor_float_range().min(),
                           options.output_tensor_float_range().max());
-    CHECK_GT(output_range_->second, output_range_->first);
+    ABSL_CHECK_GT(output_range_->second, output_range_->first);
   }
 
   // Custom div and sub values.
@@ -593,16 +655,16 @@ absl::Status TensorConverterCalculator::LoadOptions(CalculatorContext* cc) {
   }
 
   // Get y-flip mode.
-  flip_vertically_ = options.flip_vertically();
+  ASSIGN_OR_RETURN(flip_vertically_, ShouldFlipVertically(options, use_gpu));
 
   // Get row_major_matrix mode.
   row_major_matrix_ = options.row_major_matrix();
 
   // Get desired way to handle input channels.
   max_num_channels_ = options.max_num_channels();
-  CHECK_GE(max_num_channels_, 1);
+  ABSL_CHECK_GE(max_num_channels_, 1);
-  CHECK_LE(max_num_channels_, 4);
+  ABSL_CHECK_LE(max_num_channels_, 4);
-  CHECK_NE(max_num_channels_, 2);
+  ABSL_CHECK_NE(max_num_channels_, 2);
   return absl::OkStatus();
 }
 

mediapipe/calculators/tensor/tensor_converter_calculator.proto

@@ -3,6 +3,7 @@ syntax = "proto2";
 package mediapipe;
 
 import "mediapipe/framework/calculator.proto";
+import "mediapipe/gpu/gpu_origin.proto";
 
 // Full Example:
 //
@@ -43,8 +44,16 @@ message TensorConverterCalculatorOptions {
   // with a coordinate system where the origin is at the bottom-left corner
   // (e.g., in OpenGL) whereas the ML model expects an image with a top-left
   // origin.
+  // Prefer gpu_origin over this field when using GPU input images.
   optional bool flip_vertically = 2 [default = false];
 
+  // Determines when the input GPU image should be flipped vertically.
+  // See GpuOrigin.Mode for more information.
+  // Affects only IMAGE_GPU inputs.
+  // If unset, falls back to flip_vertically for backwards compatibility.
+  // Cannot set both gpu_origin and flip_vertically.
+  optional GpuOrigin.Mode gpu_origin = 10;
+
   // Controls how many channels of the input image get passed through to the
   // tensor. Valid values are 1,3,4 only. Ignored for iOS GPU.
   optional int32 max_num_channels = 3 [default = 3];

mediapipe/calculators/tensor/tensor_converter_calculator_test.cc

@@ -12,10 +12,15 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+#include <cmath>
+#include <cstdint>
+#include <memory>
 #include <random>
+#include <utility>
 #include <vector>
 
 #include "absl/memory/memory.h"
+#include "absl/status/status.h"
 #include "absl/strings/substitute.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/calculator_runner.h"
@@ -24,8 +29,10 @@
 #include "mediapipe/framework/formats/image_frame_opencv.h"
 #include "mediapipe/framework/formats/matrix.h"
 #include "mediapipe/framework/formats/tensor.h"
+#include "mediapipe/framework/port/gmock.h"
 #include "mediapipe/framework/port/gtest.h"
 #include "mediapipe/framework/port/integral_types.h"
+#include "mediapipe/framework/port/opencv_core_inc.h"
 #include "mediapipe/framework/port/parse_text_proto.h"
 #include "mediapipe/framework/port/status_matchers.h"  // NOLINT
 #include "mediapipe/framework/tool/validate_type.h"
@@ -40,7 +47,7 @@ constexpr char kTransposeOptionsString[] =
 }  // namespace
 
 using RandomEngine = std::mt19937_64;
-using testing::Eq;
+using ::testing::HasSubstr;
 const uint32_t kSeed = 1234;
 const int kNumSizes = 8;
 const int sizes[kNumSizes][2] = {{1, 1}, {12, 1}, {1, 9},   {2, 2},
@@ -53,7 +60,7 @@ class TensorConverterCalculatorTest : public ::testing::Test {
                        bool row_major_matrix = false) {
     RandomEngine random(kSeed);
     std::uniform_real_distribution<> uniform_dist(0, 1.0);
-    auto matrix = ::absl::make_unique<Matrix>();
+    auto matrix = std::make_unique<Matrix>();
     matrix->resize(num_rows, num_columns);
     if (row_major_matrix) {
       for (int y = 0; y < num_rows; ++y) {
@@ -101,7 +108,7 @@ TEST_F(TensorConverterCalculatorTest, RandomMatrixColMajor) {
     tool::AddVectorSink("tensor", &graph_config, &output_packets);
 
     // Run the graph.
-    graph_ = absl::make_unique<CalculatorGraph>();
+    graph_ = std::make_unique<CalculatorGraph>();
     MP_ASSERT_OK(graph_->Initialize(graph_config));
     MP_ASSERT_OK(graph_->StartRun({}));
 
@@ -110,12 +117,12 @@ TEST_F(TensorConverterCalculatorTest, RandomMatrixColMajor) {
 
     // Wait until the calculator done processing.
     MP_ASSERT_OK(graph_->WaitUntilIdle());
-    EXPECT_EQ(1, output_packets.size());
+    ASSERT_EQ(output_packets.size(), 1);
 
     // Get and process results.
     const std::vector<Tensor>& tensor_vec =
         output_packets[0].Get<std::vector<Tensor>>();
-    EXPECT_EQ(1, tensor_vec.size());
+    ASSERT_EQ(tensor_vec.size(), 1);
 
     const Tensor* tensor = &tensor_vec[0];
     EXPECT_EQ(Tensor::ElementType::kFloat32, tensor->element_type());
@@ -127,7 +134,7 @@ TEST_F(TensorConverterCalculatorTest, RandomMatrixColMajor) {
     auto tensor_buffer = view.buffer<float>();
     for (int i = 0; i < num_rows * num_columns; ++i) {
       const float expected = uniform_dist(random);
-      EXPECT_EQ(expected, tensor_buffer[i]) << "at i = " << i;
+      EXPECT_FLOAT_EQ(tensor_buffer[i], expected) << "at i = " << i;
     }
 
     // Fully close graph at end, otherwise calculator+tensors are destroyed
@@ -163,7 +170,7 @@ TEST_F(TensorConverterCalculatorTest, RandomMatrixRowMajor) {
     tool::AddVectorSink("tensor", &graph_config, &output_packets);
 
     // Run the graph.
-    graph_ = absl::make_unique<CalculatorGraph>();
+    graph_ = std::make_unique<CalculatorGraph>();
     MP_ASSERT_OK(graph_->Initialize(graph_config));
     MP_ASSERT_OK(graph_->StartRun({}));
 
@@ -172,12 +179,12 @@ TEST_F(TensorConverterCalculatorTest, RandomMatrixRowMajor) {
 
     // Wait until the calculator done processing.
     MP_ASSERT_OK(graph_->WaitUntilIdle());
-    EXPECT_EQ(1, output_packets.size());
+    ASSERT_EQ(output_packets.size(), 1);
 
     // Get and process results.
     const std::vector<Tensor>& tensor_vec =
         output_packets[0].Get<std::vector<Tensor>>();
-    EXPECT_EQ(1, tensor_vec.size());
+    ASSERT_EQ(tensor_vec.size(), 1);
 
     const Tensor* tensor = &tensor_vec[0];
     EXPECT_EQ(Tensor::ElementType::kFloat32, tensor->element_type());
@@ -189,7 +196,7 @@ TEST_F(TensorConverterCalculatorTest, RandomMatrixRowMajor) {
     auto tensor_buffer = view.buffer<float>();
     for (int i = 0; i < num_rows * num_columns; ++i) {
       const float expected = uniform_dist(random);
-      EXPECT_EQ(expected, tensor_buffer[i]) << "at i = " << i;
+      EXPECT_EQ(tensor_buffer[i], expected) << "at i = " << i;
     }
 
     // Fully close graph at end, otherwise calculator+tensors are destroyed
@@ -227,7 +234,7 @@ TEST_F(TensorConverterCalculatorTest, CustomDivAndSub) {
   // Run the graph.
   MP_ASSERT_OK(graph.Initialize(graph_config));
   MP_ASSERT_OK(graph.StartRun({}));
-  auto input_image = absl::make_unique<ImageFrame>(ImageFormat::GRAY8, 1, 1);
+  auto input_image = std::make_unique<ImageFrame>(ImageFormat::GRAY8, 1, 1);
   cv::Mat mat = mediapipe::formats::MatView(input_image.get());
   mat.at<uint8_t>(0, 0) = 200;
   MP_ASSERT_OK(graph.AddPacketToInputStream(
@@ -239,12 +246,12 @@ TEST_F(TensorConverterCalculatorTest, CustomDivAndSub) {
   // Get and process results.
   const std::vector<Tensor>& tensor_vec =
       output_packets[0].Get<std::vector<Tensor>>();
-  EXPECT_EQ(1, tensor_vec.size());
+  ASSERT_EQ(tensor_vec.size(), 1);
 
   const Tensor* tensor = &tensor_vec[0];
   EXPECT_EQ(Tensor::ElementType::kFloat32, tensor->element_type());
   auto view = tensor->GetCpuReadView();
-  EXPECT_FLOAT_EQ(67.0f, *view.buffer<float>());
+  EXPECT_FLOAT_EQ(*view.buffer<float>(), 67.0f);
 
   // Fully close graph at end, otherwise calculator+tensors are destroyed
   // after calling WaitUntilDone().
@@ -259,32 +266,29 @@ TEST_F(TensorConverterCalculatorTest, SetOutputRange) {
   for (std::pair<float, float> range : range_values) {
     CalculatorGraph graph;
     CalculatorGraphConfig graph_config =
-        mediapipe::ParseTextProtoOrDie<CalculatorGraphConfig>(
+        mediapipe::ParseTextProtoOrDie<CalculatorGraphConfig>(absl::Substitute(
-            absl::Substitute(R"(
+            R"pb(
-        input_stream: "input_image"
+              input_stream: "input_image"
-        node {
+              node {
-          calculator: "TensorConverterCalculator"
+                calculator: "TensorConverterCalculator"
-          input_stream: "IMAGE:input_image"
+                input_stream: "IMAGE:input_image"
-          output_stream: "TENSORS:tensor"
+                output_stream: "TENSORS:tensor"
-          options {
+                options {
-            [mediapipe.TensorConverterCalculatorOptions.ext] {
+                  [mediapipe.TensorConverterCalculatorOptions.ext] {
-              output_tensor_float_range {
+                    output_tensor_float_range { min: $0 max: $1 }
-                min: $0
+                  }
-                max: $1
+                }
               }
-            }
+            )pb",
-          }
+            /*$0=*/range.first,
-        }
+            /*$1=*/range.second));
-        )",
-                             /*$0=*/range.first,
-                             /*$1=*/range.second));
     std::vector<Packet> output_packets;
     tool::AddVectorSink("tensor", &graph_config, &output_packets);
 
     // Run the graph.
     MP_ASSERT_OK(graph.Initialize(graph_config));
     MP_ASSERT_OK(graph.StartRun({}));
-    auto input_image = absl::make_unique<ImageFrame>(ImageFormat::GRAY8, 1, 1);
+    auto input_image = std::make_unique<ImageFrame>(ImageFormat::GRAY8, 1, 1);
     cv::Mat mat = mediapipe::formats::MatView(input_image.get());
     mat.at<uint8_t>(0, 0) = 200;
     MP_ASSERT_OK(graph.AddPacketToInputStream(
@@ -292,26 +296,23 @@ TEST_F(TensorConverterCalculatorTest, SetOutputRange) {
 
     // Wait until the calculator finishes processing.
     MP_ASSERT_OK(graph.WaitUntilIdle());
-    EXPECT_THAT(output_packets.size(), Eq(1));
+    ASSERT_EQ(output_packets.size(), 1);
 
     // Get and process results.
     const std::vector<Tensor>& tensor_vec =
         output_packets[0].Get<std::vector<Tensor>>();
-    EXPECT_THAT(tensor_vec.size(), Eq(1));
+    ASSERT_EQ(tensor_vec.size(), 1);
 
     const Tensor* tensor = &tensor_vec[0];
 
     // Calculate the expected normalized value:
-    float normalized_value =
+    float expected_value =
         range.first + (200 * (range.second - range.first)) / 255.0;
 
-    EXPECT_THAT(tensor->element_type(), Eq(Tensor::ElementType::kFloat32));
+    EXPECT_EQ(tensor->element_type(), Tensor::ElementType::kFloat32);
     auto view = tensor->GetCpuReadView();
-    float dataf = *view.buffer<float>();
+    float actual_value = *view.buffer<float>();
-    EXPECT_THAT(
+    EXPECT_FLOAT_EQ(actual_value, expected_value);
-        normalized_value,
-        testing::FloatNear(dataf, 2.0f * std::abs(dataf) *
-                                      std::numeric_limits<float>::epsilon()));
 
     // Fully close graph at end, otherwise calculator+tensors are destroyed
     // after calling WaitUntilDone().
@@ -320,4 +321,153 @@ TEST_F(TensorConverterCalculatorTest, SetOutputRange) {
   }
 }
 
+TEST_F(TensorConverterCalculatorTest, FlipVertically) {
+  CalculatorGraph graph;
+  CalculatorGraphConfig graph_config =
+      mediapipe::ParseTextProtoOrDie<CalculatorGraphConfig>(R"pb(
+        input_stream: "input_image"
+        node {
+          calculator: "TensorConverterCalculator"
+          input_stream: "IMAGE:input_image"
+          output_stream: "TENSORS:tensor"
+          options {
+            [mediapipe.TensorConverterCalculatorOptions.ext] {
+              flip_vertically: true
+              output_tensor_float_range { min: 0 max: 255 }
+            }
+          }
+        }
+      )pb");
+  std::vector<Packet> output_packets;
+  tool::AddVectorSink("tensor", &graph_config, &output_packets);
+
+  // Run the graph.
+  MP_ASSERT_OK(graph.Initialize(graph_config));
+  MP_ASSERT_OK(graph.StartRun({}));
+  auto input_image = std::make_unique<ImageFrame>(ImageFormat::GRAY8, 1, 2);
+  cv::Mat mat = mediapipe::formats::MatView(input_image.get());
+  constexpr uint8_t kY0Value = 100;
+  constexpr uint8_t kY1Value = 200;
+  mat.at<uint8_t>(0, 0) = kY0Value;
+  mat.at<uint8_t>(1, 0) = kY1Value;  // Note: y, x!
+  MP_ASSERT_OK(graph.AddPacketToInputStream(
+      "input_image", Adopt(input_image.release()).At(Timestamp(0))));
+
+  // Wait until the calculator finishes processing.
+  MP_ASSERT_OK(graph.WaitUntilIdle());
+  ASSERT_EQ(output_packets.size(), 1);
+
+  // Get and process results.
+  const std::vector<Tensor>& tensor_vec =
+      output_packets[0].Get<std::vector<Tensor>>();
+  ASSERT_EQ(tensor_vec.size(), 1);
+
+  const Tensor* tensor = &tensor_vec[0];
+
+  EXPECT_EQ(tensor->element_type(), Tensor::ElementType::kFloat32);
+  const float* dataf = tensor->GetCpuReadView().buffer<float>();
+  EXPECT_EQ(static_cast<int>(roundf(dataf[0])), kY1Value);  // Y0, Y1 flipped!
+  EXPECT_EQ(static_cast<int>(roundf(dataf[1])), kY0Value);
+
+  // Fully close graph at end, otherwise calculator+tensors are destroyed
+  // after calling WaitUntilDone().
+  MP_ASSERT_OK(graph.CloseInputStream("input_image"));
+  MP_ASSERT_OK(graph.WaitUntilDone());
+}
+
+TEST_F(TensorConverterCalculatorTest,
+       CannotSpecifyBothFlipVerticallyAndGpuOrigin) {
+  CalculatorGraph graph;
+  CalculatorGraphConfig graph_config =
+      mediapipe::ParseTextProtoOrDie<CalculatorGraphConfig>(R"pb(
+        input_stream: "input_image"
+        node {
+          calculator: "TensorConverterCalculator"
+          input_stream: "IMAGE:input_image"
+          output_stream: "TENSORS:tensor"
+          options {
+            [mediapipe.TensorConverterCalculatorOptions.ext] {
+              flip_vertically: true
+              gpu_origin: TOP_LEFT
+              output_tensor_float_range { min: 0 max: 255 }
+            }
+          }
+        }
+      )pb");
+  std::vector<Packet> output_packets;
+  tool::AddVectorSink("tensor", &graph_config, &output_packets);
+
+  // Run the graph.
+  MP_ASSERT_OK(graph.Initialize(graph_config));
+  MP_ASSERT_OK(graph.StartRun({}));
+  auto input_image = std::make_unique<ImageFrame>(ImageFormat::GRAY8, 1, 1);
+  MP_ASSERT_OK(graph.AddPacketToInputStream(
+      "input_image", Adopt(input_image.release()).At(Timestamp(0))));
+
+  // Processing should fail as we specified both flip_vertically and gpu_origin.
+  absl::Status status = graph.WaitUntilIdle();
+  EXPECT_FALSE(status.ok());
+  EXPECT_THAT(status.message(), HasSubstr("flip_vertically and gpu_origin"));
+  EXPECT_EQ(output_packets.size(), 0);
+
+  // Fully close graph at end, otherwise calculator+tensors are destroyed
+  // after calling WaitUntilDone().
+  MP_ASSERT_OK(graph.CloseInputStream("input_image"));
+  EXPECT_FALSE(graph.WaitUntilDone().ok());
+}
+
+TEST_F(TensorConverterCalculatorTest, GpuOriginIsIgnoredWithCpuImage) {
+  CalculatorGraph graph;
+  CalculatorGraphConfig graph_config =
+      mediapipe::ParseTextProtoOrDie<CalculatorGraphConfig>(R"pb(
+        input_stream: "input_image"
+        node {
+          calculator: "TensorConverterCalculator"
+          input_stream: "IMAGE:input_image"
+          output_stream: "TENSORS:tensor"
+          options {
+            [mediapipe.TensorConverterCalculatorOptions.ext] {
+              gpu_origin: CONVENTIONAL
+              output_tensor_float_range { min: 0 max: 255 }
+            }
+          }
+        }
+      )pb");
+  std::vector<Packet> output_packets;
+  tool::AddVectorSink("tensor", &graph_config, &output_packets);
+
+  // Run the graph.
+  MP_ASSERT_OK(graph.Initialize(graph_config));
+  MP_ASSERT_OK(graph.StartRun({}));
+  auto input_image = std::make_unique<ImageFrame>(ImageFormat::GRAY8, 1, 2);
+  cv::Mat mat = mediapipe::formats::MatView(input_image.get());
+  constexpr uint8_t kY0Value = 100;
+  constexpr uint8_t kY1Value = 200;
+  mat.at<uint8_t>(0, 0) = kY0Value;
+  mat.at<uint8_t>(1, 0) = kY1Value;  // Note: y, x!
+  MP_ASSERT_OK(graph.AddPacketToInputStream(
+      "input_image", Adopt(input_image.release()).At(Timestamp(0))));
+
+  // Wait until the calculator finishes processing.
+  MP_ASSERT_OK(graph.WaitUntilIdle());
+  ASSERT_EQ(output_packets.size(), 1);
+
+  // Get and process results.
+  const std::vector<Tensor>& tensor_vec =
+      output_packets[0].Get<std::vector<Tensor>>();
+  ASSERT_EQ(tensor_vec.size(), 1);
+
+  const Tensor* tensor = &tensor_vec[0];
+
+  EXPECT_EQ(tensor->element_type(), Tensor::ElementType::kFloat32);
+  const float* dataf = tensor->GetCpuReadView().buffer<float>();
+  EXPECT_EQ(static_cast<int>(roundf(dataf[0])), kY0Value);  // Not flipped!
+  EXPECT_EQ(static_cast<int>(roundf(dataf[1])), kY1Value);
+
+  // Fully close graph at end, otherwise calculator+tensors are destroyed
+  // after calling WaitUntilDone().
+  MP_ASSERT_OK(graph.CloseInputStream("input_image"));
+  MP_ASSERT_OK(graph.WaitUntilDone());
+}
+
 }  // namespace mediapipe

mediapipe/calculators/tensor/tensors_to_detections_calculator.cc

@@ -15,6 +15,7 @@
 #include <unordered_map>
 #include <vector>
 
+#include "absl/log/absl_log.h"
 #include "absl/strings/str_format.h"
 #include "absl/types/span.h"
 #include "mediapipe/calculators/tensor/tensors_to_detections_calculator.pb.h"
@@ -83,7 +84,7 @@ void ConvertRawValuesToAnchors(const float* raw_anchors, int num_boxes,
 
 void ConvertAnchorsToRawValues(const std::vector<Anchor>& anchors,
                                int num_boxes, float* raw_anchors) {
-  CHECK_EQ(anchors.size(), num_boxes);
+  ABSL_CHECK_EQ(anchors.size(), num_boxes);
   int box = 0;
   for (const auto& anchor : anchors) {
     raw_anchors[box * kNumCoordsPerBox + 0] = anchor.y_center();
@@ -329,7 +330,7 @@ absl::Status TensorsToDetectionsCalculator::Process(CalculatorContext* cc) {
     } else if (status.code() == absl::StatusCode::kFailedPrecondition) {
       // For initialization error because of hardware limitation, fallback to
       // CPU processing.
-      LOG(WARNING) << status.message();
+      ABSL_LOG(WARNING) << status.message();
     } else {
       // For other error, let the error propagates.
       return status;
@@ -668,7 +669,7 @@ absl::Status TensorsToDetectionsCalculator::ProcessGPU(
                                          output_detections));
 
 #else
-  LOG(ERROR) << "GPU input on non-Android not supported yet.";
+  ABSL_LOG(ERROR) << "GPU input on non-Android not supported yet.";
 #endif  // !defined(MEDIAPIPE_DISABLE_GL_COMPUTE)
   return absl::OkStatus();
 }
@@ -703,18 +704,18 @@ absl::Status TensorsToDetectionsCalculator::LoadOptions(CalculatorContext* cc) {
   num_boxes_ = options_.num_boxes();
   num_coords_ = options_.num_coords();
   box_output_format_ = GetBoxFormat(options_);
-  CHECK_NE(options_.max_results(), 0)
+  ABSL_CHECK_NE(options_.max_results(), 0)
       << "The maximum number of the top-scored detection results must be "
          "non-zero.";
   max_results_ = options_.max_results();
 
   // Currently only support 2D when num_values_per_keypoint equals to 2.
-  CHECK_EQ(options_.num_values_per_keypoint(), 2);
+  ABSL_CHECK_EQ(options_.num_values_per_keypoint(), 2);
 
   // Check if the output size is equal to the requested boxes and keypoints.
-  CHECK_EQ(options_.num_keypoints() * options_.num_values_per_keypoint() +
+  ABSL_CHECK_EQ(options_.num_keypoints() * options_.num_values_per_keypoint() +
-               kNumCoordsPerBox,
+                    kNumCoordsPerBox,
-           num_coords_);
+                num_coords_);
 
   if (kSideInIgnoreClasses(cc).IsConnected()) {
     RET_CHECK(!kSideInIgnoreClasses(cc).IsEmpty());
@@ -1154,11 +1155,12 @@ void main() {
     }
     // TODO support better filtering.
     if (class_index_set_.is_allowlist) {
-      CHECK_EQ(class_index_set_.values.size(),
+      ABSL_CHECK_EQ(class_index_set_.values.size(),
-               IsClassIndexAllowed(0) ? num_classes_ : num_classes_ - 1)
+                    IsClassIndexAllowed(0) ? num_classes_ : num_classes_ - 1)
           << "Only all classes  >= class 0  or  >= class 1";
     } else {
-      CHECK_EQ(class_index_set_.values.size(), IsClassIndexAllowed(0) ? 0 : 1)
+      ABSL_CHECK_EQ(class_index_set_.values.size(),
+                    IsClassIndexAllowed(0) ? 0 : 1)
           << "Only ignore class 0 is allowed";
     }
 
@@ -1379,11 +1381,12 @@ kernel void scoreKernel(
 
   // TODO support better filtering.
   if (class_index_set_.is_allowlist) {
-    CHECK_EQ(class_index_set_.values.size(),
+    ABSL_CHECK_EQ(class_index_set_.values.size(),
-             IsClassIndexAllowed(0) ? num_classes_ : num_classes_ - 1)
+                  IsClassIndexAllowed(0) ? num_classes_ : num_classes_ - 1)
         << "Only all classes  >= class 0  or  >= class 1";
   } else {
-    CHECK_EQ(class_index_set_.values.size(), IsClassIndexAllowed(0) ? 0 : 1)
+    ABSL_CHECK_EQ(class_index_set_.values.size(),
+                  IsClassIndexAllowed(0) ? 0 : 1)
         << "Only ignore class 0 is allowed";
   }
 

mediapipe/calculators/tensor/tensors_to_landmarks_calculator.cc

@@ -142,7 +142,7 @@ absl::Status TensorsToLandmarksCalculator::Process(CalculatorContext* cc) {
   RET_CHECK(input_tensors[0].element_type() == Tensor::ElementType::kFloat32);
   int num_values = input_tensors[0].shape().num_elements();
   const int num_dimensions = num_values / num_landmarks_;
-  CHECK_GT(num_dimensions, 0);
+  ABSL_CHECK_GT(num_dimensions, 0);
 
   auto view = input_tensors[0].GetCpuReadView();
   auto raw_landmarks = view.buffer<float>();

mediapipe/calculators/tensorflow/BUILD

@@ -13,6 +13,7 @@
 # limitations under the License.
 #
 
+# Placeholder: load py_proto_library
 load("//mediapipe/framework/port:build_config.bzl", "mediapipe_cc_proto_library", "mediapipe_proto_library")
 
 licenses(["notice"])
@@ -314,6 +315,7 @@ cc_library(
         "//mediapipe/framework/formats:time_series_header_cc_proto",
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/framework/port:status",
+        "@com_google_absl//absl/log:absl_check",
     ] + select({
         "//conditions:default": [
             "@org_tensorflow//tensorflow/core:framework",
@@ -366,18 +368,18 @@ cc_library(
     name = "pack_media_sequence_calculator",
     srcs = ["pack_media_sequence_calculator.cc"],
     deps = [
+        ":pack_media_sequence_calculator_cc_proto",
         "//mediapipe/calculators/image:opencv_image_encoder_calculator_cc_proto",
-        "//mediapipe/calculators/tensorflow:pack_media_sequence_calculator_cc_proto",
         "//mediapipe/framework:calculator_framework",
         "//mediapipe/framework/formats:detection_cc_proto",
         "//mediapipe/framework/formats:location",
         "//mediapipe/framework/formats:location_opencv",
         "//mediapipe/framework/port:opencv_imgcodecs",
         "//mediapipe/framework/port:ret_check",
-        "//mediapipe/framework/port:status",
         "//mediapipe/util/sequence:media_sequence",
         "//mediapipe/util/sequence:media_sequence_util",
         "@com_google_absl//absl/container:flat_hash_map",
+        "@com_google_absl//absl/status",
         "@com_google_absl//absl/strings",
         "@org_tensorflow//tensorflow/core:protos_all_cc",
     ],
@@ -406,8 +408,13 @@ cc_library(
     alwayslink = 1,
 )
 
-# This dependency removed tensorflow_jellyfish_deps and xprofilez_with_server because they failed
+# This dependency removed the following 3 targets because they failed Boq conformance test:
-# Boq conformance test. Weigh your use case to see if this will work for you.
+#
+# tensorflow_jellyfish_deps
+# jfprof_lib
+# xprofilez_with_server
+#
+# If you need them plz consider tensorflow_inference_calculator_no_envelope_loader.
 cc_library(
     name = "tensorflow_inference_calculator_for_boq",
     srcs = ["tensorflow_inference_calculator.cc"],
@@ -424,7 +431,7 @@ cc_library(
         "//mediapipe/framework/port:status",
         "//mediapipe/framework/tool:status_util",
         "@com_google_absl//absl/base:core_headers",
-        "@com_google_absl//absl/log:check",
+        "@com_google_absl//absl/log:absl_check",
         "@com_google_absl//absl/memory",
         "@com_google_absl//absl/strings",
         "@com_google_absl//absl/synchronization",
@@ -483,10 +490,10 @@ cc_library(
         "//mediapipe/calculators/tensorflow:tensorflow_session_from_frozen_graph_calculator_cc_proto",
         "//mediapipe/framework:calculator_framework",
         "//mediapipe/framework/deps:clock",
-        "//mediapipe/framework/port:logging",
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/framework/port:status",
         "//mediapipe/framework/tool:status_util",
+        "@com_google_absl//absl/log:absl_log",
         "@org_tensorflow//tensorflow/core:protos_all_cc",
     ] + select({
         "//conditions:default": [
@@ -514,10 +521,10 @@ cc_library(
         ":tensorflow_session_from_frozen_graph_generator_cc_proto",
         "//mediapipe/framework:calculator_framework",
         "//mediapipe/framework/deps:clock",
-        "//mediapipe/framework/port:logging",
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/framework/port:status",
         "//mediapipe/framework/tool:status_util",
+        "@com_google_absl//absl/log:absl_log",
         "@org_tensorflow//tensorflow/core:protos_all_cc",
     ] + select({
         "//conditions:default": [
@@ -550,6 +557,7 @@ cc_library(
         "//mediapipe/framework/deps:file_path",
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/framework/port:status",
+        "@com_google_absl//absl/log:absl_log",
         "@com_google_absl//absl/strings",
         "@org_tensorflow//tensorflow/cc/saved_model:constants",
         "@org_tensorflow//tensorflow/cc/saved_model:loader_lite",
@@ -627,6 +635,7 @@ cc_library(
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/framework/port:status",
         "//mediapipe/framework/tool:status_util",
+        "@com_google_absl//absl/log:absl_log",
         "@com_google_absl//absl/status",
         "@com_google_absl//absl/strings",
         "@org_tensorflow//tensorflow/cc/saved_model:constants",
@@ -648,6 +657,7 @@ cc_library(
         "//mediapipe/framework:calculator_framework",
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/framework/port:status",
+        "@com_google_absl//absl/log:absl_log",
         "@org_tensorflow//tensorflow/core:framework",
     ],
     alwayslink = 1,
@@ -662,6 +672,7 @@ cc_library(
         "//mediapipe/framework/formats:image_frame",
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/framework/port:status",
+        "@com_google_absl//absl/log:absl_check",
         "@org_tensorflow//tensorflow/core:framework",
     ],
     alwayslink = 1,
@@ -677,6 +688,7 @@ cc_library(
         "//mediapipe/framework/formats:time_series_header_cc_proto",
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/framework/port:status",
+        "@com_google_absl//absl/log:absl_check",
     ] + select({
         "//conditions:default": [
             "@org_tensorflow//tensorflow/core:framework",
@@ -711,6 +723,7 @@ cc_library(
         "//mediapipe/framework:calculator_framework",
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/framework/port:status",
+        "@org_tensorflow//tensorflow/core/platform:bfloat16",
     ] + select({
         "//conditions:default": [
             "@org_tensorflow//tensorflow/core:framework",
@@ -773,6 +786,7 @@ cc_library(
         "//mediapipe/framework/port:status",
         "//mediapipe/util:audio_decoder_cc_proto",
         "//mediapipe/util/sequence:media_sequence",
+        "@com_google_absl//absl/log:absl_log",
         "@com_google_absl//absl/strings",
         "@org_tensorflow//tensorflow/core:protos_all_cc",
     ],
@@ -787,6 +801,8 @@ cc_library(
         "//mediapipe/framework:calculator_framework",
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/framework/port:status",
+        "@com_google_absl//absl/log:absl_check",
+        "@com_google_absl//absl/log:absl_log",
         "@org_tensorflow//tensorflow/core:framework",
     ],
     alwayslink = 1,
@@ -800,6 +816,7 @@ cc_library(
         "//mediapipe/framework:calculator_framework",
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/framework/port:status",
+        "@com_google_absl//absl/log:absl_log",
         "@org_tensorflow//tensorflow/core:framework",
     ],
     alwayslink = 1,
@@ -813,6 +830,7 @@ cc_library(
         "//mediapipe/framework:calculator_framework",
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/framework/port:status",
+        "@com_google_absl//absl/log:absl_log",
         "@org_tensorflow//tensorflow/core:framework",
     ],
     alwayslink = 1,
@@ -826,6 +844,8 @@ cc_library(
         "//mediapipe/framework:calculator_framework",
         "//mediapipe/framework:packet",
         "//mediapipe/framework/port:status",
+        "@com_google_absl//absl/log:absl_check",
+        "@com_google_absl//absl/log:absl_log",
         "@org_tensorflow//tensorflow/core:protos_all_cc",
     ],
     alwayslink = 1,
@@ -920,22 +940,24 @@ cc_test(
     srcs = ["pack_media_sequence_calculator_test.cc"],
     deps = [
         ":pack_media_sequence_calculator",
+        ":pack_media_sequence_calculator_cc_proto",
         "//mediapipe/calculators/image:opencv_image_encoder_calculator_cc_proto",
-        "//mediapipe/calculators/tensorflow:pack_media_sequence_calculator_cc_proto",
         "//mediapipe/framework:calculator_framework",
         "//mediapipe/framework:calculator_runner",
+        "//mediapipe/framework:packet",
         "//mediapipe/framework:timestamp",
         "//mediapipe/framework/formats:detection_cc_proto",
-        "//mediapipe/framework/formats:image_frame",
-        "//mediapipe/framework/formats:image_frame_opencv",
         "//mediapipe/framework/formats:location",
         "//mediapipe/framework/formats:location_opencv",
         "//mediapipe/framework/port:gtest_main",
         "//mediapipe/framework/port:opencv_imgcodecs",
         "//mediapipe/util/sequence:media_sequence",
-        "@com_google_absl//absl/container:flat_hash_map",
+        "//mediapipe/util/sequence:media_sequence_util",
+        "@com_google_absl//absl/log:absl_check",
         "@com_google_absl//absl/memory",
+        "@com_google_absl//absl/status",
         "@com_google_absl//absl/strings",
+        "@com_google_googletest//:gtest_main",
         "@org_tensorflow//tensorflow/core:protos_all_cc",
     ],
 )
@@ -1118,6 +1140,7 @@ cc_test(
         "//mediapipe/util:packet_test_util",
         "@org_tensorflow//tensorflow/core:framework",
         "@org_tensorflow//tensorflow/core:protos_all_cc",
+        "@org_tensorflow//tensorflow/core/platform:bfloat16",
     ],
 )
 
@@ -1163,6 +1186,7 @@ cc_test(
         "//mediapipe/framework/port:rectangle",
         "//mediapipe/util:audio_decoder_cc_proto",
         "//mediapipe/util/sequence:media_sequence",
+        "@com_google_absl//absl/log:absl_log",
         "@com_google_absl//absl/memory",
         "@com_google_absl//absl/strings",
         "@org_tensorflow//tensorflow/core:protos_all_cc",
@@ -1244,6 +1268,8 @@ cc_test(
         "//mediapipe/framework/tool:sink",
         "//mediapipe/framework/tool:validate_type",
         "@com_google_absl//absl/flags:flag",
+        "@com_google_absl//absl/log:absl_check",
+        "@com_google_absl//absl/log:absl_log",
     ] + select({
         "//conditions:default": [
             "@org_tensorflow//tensorflow/core:direct_session",

mediapipe/calculators/tensorflow/matrix_to_tensor_calculator.cc

@@ -12,6 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+#include "absl/log/absl_check.h"
 #include "mediapipe/calculators/tensorflow/matrix_to_tensor_calculator_options.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/formats/matrix.h"
@@ -28,7 +29,7 @@ namespace mediapipe {
 namespace {
 absl::Status FillTimeSeriesHeaderIfValid(const Packet& header_packet,
                                          TimeSeriesHeader* header) {
-  CHECK(header);
+  ABSL_CHECK(header);
   if (header_packet.IsEmpty()) {
     return absl::UnknownError("No header found.");
   }

mediapipe/calculators/tensorflow/pack_media_sequence_calculator.cc

@@ -12,21 +12,23 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+#include <cstdint>
+#include <optional>
 #include <string>
 #include <vector>
 
 #include "absl/container/flat_hash_map.h"
+#include "absl/status/status.h"
 #include "absl/strings/match.h"
+#include "absl/strings/strip.h"
 #include "mediapipe/calculators/image/opencv_image_encoder_calculator.pb.h"
 #include "mediapipe/calculators/tensorflow/pack_media_sequence_calculator.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/formats/detection.pb.h"
 #include "mediapipe/framework/formats/location.h"
 #include "mediapipe/framework/formats/location_opencv.h"
-#include "mediapipe/framework/port/canonical_errors.h"
 #include "mediapipe/framework/port/opencv_imgcodecs_inc.h"
 #include "mediapipe/framework/port/ret_check.h"
-#include "mediapipe/framework/port/status.h"
 #include "mediapipe/util/sequence/media_sequence.h"
 #include "mediapipe/util/sequence/media_sequence_util.h"
 #include "tensorflow/core/example/example.pb.h"
@@ -36,7 +38,11 @@ namespace mediapipe {
 
 const char kSequenceExampleTag[] = "SEQUENCE_EXAMPLE";
 const char kImageTag[] = "IMAGE";
+const char kImageLabelPrefixTag[] = "IMAGE_LABEL_";
+const char kClipLabelPrefixTag[] = "CLIP_LABEL_";
 const char kFloatContextFeaturePrefixTag[] = "FLOAT_CONTEXT_FEATURE_";
+const char kIntsContextFeaturePrefixTag[] = "INTS_CONTEXT_FEATURE_";
+const char kBytesContextFeaturePrefixTag[] = "BYTES_CONTEXT_FEATURE_";
 const char kFloatFeaturePrefixTag[] = "FLOAT_FEATURE_";
 const char kIntFeaturePrefixTag[] = "INT_FEATURE_";
 const char kBytesFeaturePrefixTag[] = "BYTES_FEATURE_";
@@ -44,6 +50,7 @@ const char kForwardFlowEncodedTag[] = "FORWARD_FLOW_ENCODED";
 const char kBBoxTag[] = "BBOX";
 const char kKeypointsTag[] = "KEYPOINTS";
 const char kSegmentationMaskTag[] = "CLASS_SEGMENTATION";
+const char kClipMediaIdTag[] = "CLIP_MEDIA_ID";
 
 namespace tf = ::tensorflow;
 namespace mpms = mediapipe::mediasequence;
@@ -55,16 +62,23 @@ namespace mpms = mediapipe::mediasequence;
 // context features can be supplied verbatim in the calculator's options. The
 // SequenceExample will conform to the description in media_sequence.h.
 //
-// The supported input stream tags are "IMAGE", which stores the encoded
+// The supported input stream tags are:
-// images from the OpenCVImageEncoderCalculator, "FORWARD_FLOW_ENCODED", which
+// * "IMAGE", which stores the encoded images from the
-// stores the encoded optical flow from the same calculator, "BBOX" which stores
+//   OpenCVImageEncoderCalculator,
-// bounding boxes from vector<Detections>, and streams with the
+// * "IMAGE_LABEL", which stores whole image labels from Detection,
-// "FLOAT_FEATURE_${NAME}" pattern, which stores the values from vector<float>'s
+// * "FORWARD_FLOW_ENCODED", which stores the encoded optical flow from the same
-// associated with the name ${NAME}. "KEYPOINTS" stores a map of 2D keypoints
+//   calculator,
-// from flat_hash_map<string, vector<pair<float, float>>>. "IMAGE_${NAME}",
+// * "BBOX" which stores bounding boxes from vector<Detections>,
-// "BBOX_${NAME}", and "KEYPOINTS_${NAME}" will also store prefixed versions of
+// * streams with the "FLOAT_FEATURE_${NAME}" pattern, which stores the values
-// each stream, which allows for multiple image streams to be included. However,
+//   from vector<float>'s associated with the name ${NAME},
-// the default names are suppored by more tools.
+// * "KEYPOINTS" stores a map of 2D keypoints from flat_hash_map<string,
+//   vector<pair<float, float>>>,
+// * "CLIP_MEDIA_ID", which stores the clip's media ID as a string.
+// * "CLIP_LABEL_${NAME}" which stores sparse feature labels, ID and scores in
+//   mediapipe::Detection.
+// "IMAGE_${NAME}", "BBOX_${NAME}", and "KEYPOINTS_${NAME}" will also store
+// prefixed versions of each stream, which allows for multiple image streams to
+// be included. However, the default names are suppored by more tools.
 //
 // Example config:
 // node {
@@ -100,6 +114,9 @@ class PackMediaSequenceCalculator : public CalculatorBase {
   static absl::Status GetContract(CalculatorContract* cc) {
     RET_CHECK(cc->InputSidePackets().HasTag(kSequenceExampleTag));
     cc->InputSidePackets().Tag(kSequenceExampleTag).Set<tf::SequenceExample>();
+    if (cc->InputSidePackets().HasTag(kClipMediaIdTag)) {
+      cc->InputSidePackets().Tag(kClipMediaIdTag).Set<std::string>();
+    }
 
     if (cc->Inputs().HasTag(kForwardFlowEncodedTag)) {
       cc->Inputs()
@@ -112,6 +129,10 @@ class PackMediaSequenceCalculator : public CalculatorBase {
 
     for (const auto& tag : cc->Inputs().GetTags()) {
       if (absl::StartsWith(tag, kImageTag)) {
+        if (absl::StartsWith(tag, kImageLabelPrefixTag)) {
+          cc->Inputs().Tag(tag).Set<Detection>();
+          continue;
+        }
         std::string key = "";
         if (tag != kImageTag) {
           int tag_length = sizeof(kImageTag) / sizeof(*kImageTag) - 1;
@@ -150,9 +171,18 @@ class PackMediaSequenceCalculator : public CalculatorBase {
         }
         cc->Inputs().Tag(tag).Set<std::vector<Detection>>();
       }
+      if (absl::StartsWith(tag, kClipLabelPrefixTag)) {
+        cc->Inputs().Tag(tag).Set<Detection>();
+      }
       if (absl::StartsWith(tag, kFloatContextFeaturePrefixTag)) {
         cc->Inputs().Tag(tag).Set<std::vector<float>>();
       }
+      if (absl::StartsWith(tag, kIntsContextFeaturePrefixTag)) {
+        cc->Inputs().Tag(tag).Set<std::vector<int64_t>>();
+      }
+      if (absl::StartsWith(tag, kBytesContextFeaturePrefixTag)) {
+        cc->Inputs().Tag(tag).Set<std::vector<std::string>>();
+      }
       if (absl::StartsWith(tag, kFloatFeaturePrefixTag)) {
         cc->Inputs().Tag(tag).Set<std::vector<float>>();
       }
@@ -164,8 +194,8 @@ class PackMediaSequenceCalculator : public CalculatorBase {
       }
     }
 
-    CHECK(cc->Outputs().HasTag(kSequenceExampleTag) ||
+    RET_CHECK(cc->Outputs().HasTag(kSequenceExampleTag) ||
-          cc->OutputSidePackets().HasTag(kSequenceExampleTag))
+              cc->OutputSidePackets().HasTag(kSequenceExampleTag))
         << "Neither the output stream nor the output side packet is set to "
            "output the sequence example.";
     if (cc->Outputs().HasTag(kSequenceExampleTag)) {
@@ -184,6 +214,11 @@ class PackMediaSequenceCalculator : public CalculatorBase {
         cc->InputSidePackets()
             .Tag(kSequenceExampleTag)
             .Get<tf::SequenceExample>());
+    if (cc->InputSidePackets().HasTag(kClipMediaIdTag) &&
+        !cc->InputSidePackets().Tag(kClipMediaIdTag).IsEmpty()) {
+      clip_media_id_ =
+          cc->InputSidePackets().Tag(kClipMediaIdTag).Get<std::string>();
+    }
 
     const auto& context_features =
         cc->Options<PackMediaSequenceCalculatorOptions>().context_feature_map();
@@ -197,8 +232,19 @@ class PackMediaSequenceCalculator : public CalculatorBase {
     replace_keypoints_ = false;
     if (cc->Options<PackMediaSequenceCalculatorOptions>()
             .replace_data_instead_of_append()) {
+      // Clear the existing values under the same key.
       for (const auto& tag : cc->Inputs().GetTags()) {
         if (absl::StartsWith(tag, kImageTag)) {
+          if (absl::StartsWith(tag, kImageLabelPrefixTag)) {
+            std::string key =
+                std::string(absl::StripPrefix(tag, kImageLabelPrefixTag));
+            mpms::ClearImageLabelString(key, sequence_.get());
+            mpms::ClearImageLabelConfidence(key, sequence_.get());
+            if (!key.empty() || mpms::HasImageEncoded(*sequence_)) {
+              mpms::ClearImageTimestamp(key, sequence_.get());
+            }
+            continue;
+          }
           std::string key = "";
           if (tag != kImageTag) {
             int tag_length = sizeof(kImageTag) / sizeof(*kImageTag) - 1;
@@ -227,12 +273,41 @@ class PackMediaSequenceCalculator : public CalculatorBase {
           mpms::ClearBBoxNumRegions(key, sequence_.get());
           mpms::ClearBBoxLabelString(key, sequence_.get());
           mpms::ClearBBoxLabelIndex(key, sequence_.get());
+          mpms::ClearBBoxLabelConfidence(key, sequence_.get());
           mpms::ClearBBoxClassString(key, sequence_.get());
           mpms::ClearBBoxClassIndex(key, sequence_.get());
           mpms::ClearBBoxTrackString(key, sequence_.get());
           mpms::ClearBBoxTrackIndex(key, sequence_.get());
           mpms::ClearUnmodifiedBBoxTimestamp(key, sequence_.get());
         }
+        if (absl::StartsWith(tag, kClipLabelPrefixTag)) {
+          const std::string& key = tag.substr(
+              sizeof(kClipLabelPrefixTag) / sizeof(*kClipLabelPrefixTag) - 1);
+          mpms::ClearClipLabelIndex(key, sequence_.get());
+          mpms::ClearClipLabelString(key, sequence_.get());
+          mpms::ClearClipLabelConfidence(key, sequence_.get());
+        }
+        if (absl::StartsWith(tag, kFloatContextFeaturePrefixTag)) {
+          const std::string& key =
+              tag.substr(sizeof(kFloatContextFeaturePrefixTag) /
+                             sizeof(*kFloatContextFeaturePrefixTag) -
+                         1);
+          mpms::ClearContextFeatureFloats(key, sequence_.get());
+        }
+        if (absl::StartsWith(tag, kIntsContextFeaturePrefixTag)) {
+          const std::string& key =
+              tag.substr(sizeof(kIntsContextFeaturePrefixTag) /
+                             sizeof(*kIntsContextFeaturePrefixTag) -
+                         1);
+          mpms::ClearContextFeatureInts(key, sequence_.get());
+        }
+        if (absl::StartsWith(tag, kBytesContextFeaturePrefixTag)) {
+          const std::string& key =
+              tag.substr(sizeof(kBytesContextFeaturePrefixTag) /
+                             sizeof(*kBytesContextFeaturePrefixTag) -
+                         1);
+          mpms::ClearContextFeatureBytes(key, sequence_.get());
+        }
         if (absl::StartsWith(tag, kFloatFeaturePrefixTag)) {
           std::string key = tag.substr(sizeof(kFloatFeaturePrefixTag) /
                                            sizeof(*kFloatFeaturePrefixTag) -
@@ -343,6 +418,34 @@ class PackMediaSequenceCalculator : public CalculatorBase {
       if (absl::StartsWith(tag, kImageTag) &&
           !cc->Inputs().Tag(tag).IsEmpty()) {
         std::string key = "";
+        if (absl::StartsWith(tag, kImageLabelPrefixTag)) {
+          std::string key =
+              std::string(absl::StripPrefix(tag, kImageLabelPrefixTag));
+          const auto& detection = cc->Inputs().Tag(tag).Get<Detection>();
+          if (detection.label().empty()) continue;
+          RET_CHECK(detection.label_size() == detection.score_size())
+              << "Wrong image label data format: " << detection.label_size()
+              << " vs " << detection.score_size();
+          if (!detection.label_id().empty()) {
+            RET_CHECK(detection.label_id_size() == detection.label_size())
+                << "Wrong image label ID format: " << detection.label_id_size()
+                << " vs " << detection.label_size();
+          }
+          std::vector<std::string> labels(detection.label().begin(),
+                                          detection.label().end());
+          std::vector<float> confidences(detection.score().begin(),
+                                         detection.score().end());
+          std::vector<int32_t> ids(detection.label_id().begin(),
+                                   detection.label_id().end());
+          if (!key.empty() || mpms::HasImageEncoded(*sequence_)) {
+            mpms::AddImageTimestamp(key, cc->InputTimestamp().Value(),
+                                    sequence_.get());
+          }
+          mpms::AddImageLabelString(key, labels, sequence_.get());
+          mpms::AddImageLabelConfidence(key, confidences, sequence_.get());
+          if (!ids.empty()) mpms::AddImageLabelIndex(key, ids, sequence_.get());
+          continue;
+        }
         if (tag != kImageTag) {
           int tag_length = sizeof(kImageTag) / sizeof(*kImageTag) - 1;
           if (tag[tag_length] == '_') {
@@ -393,6 +496,7 @@ class PackMediaSequenceCalculator : public CalculatorBase {
             mpms::ClearBBoxNumRegions(prefix, sequence_.get());
             mpms::ClearBBoxLabelString(prefix, sequence_.get());
             mpms::ClearBBoxLabelIndex(prefix, sequence_.get());
+            mpms::ClearBBoxLabelConfidence(prefix, sequence_.get());
             mpms::ClearBBoxClassString(prefix, sequence_.get());
             mpms::ClearBBoxClassIndex(prefix, sequence_.get());
             mpms::ClearBBoxTrackString(prefix, sequence_.get());
@@ -405,6 +509,33 @@ class PackMediaSequenceCalculator : public CalculatorBase {
         }
         replace_keypoints_ = false;
       }
+      if (absl::StartsWith(tag, kClipLabelPrefixTag) &&
+          !cc->Inputs().Tag(tag).IsEmpty()) {
+        const std::string& key = tag.substr(
+            sizeof(kClipLabelPrefixTag) / sizeof(*kClipLabelPrefixTag) - 1);
+        const Detection& detection = cc->Inputs().Tag(tag).Get<Detection>();
+        if (detection.label().size() != detection.score().size()) {
+          return absl::InvalidArgumentError(
+              "Different size of detection.label and detection.score");
+        }
+        // Allow empty label_ids, but if label_ids is not empty, it should have
+        // the same size as the label and score fields.
+        if (!detection.label_id().empty()) {
+          if (detection.label_id().size() != detection.label().size()) {
+            return absl::InvalidArgumentError(
+                "Different size of detection.label_id and detection.label");
+          }
+        }
+        for (int i = 0; i < detection.label().size(); ++i) {
+          if (!detection.label_id().empty()) {
+            mpms::AddClipLabelIndex(key, detection.label_id(i),
+                                    sequence_.get());
+          }
+          mpms::AddClipLabelString(key, detection.label(i), sequence_.get());
+          mpms::AddClipLabelConfidence(key, detection.score(i),
+                                       sequence_.get());
+        }
+      }
       if (absl::StartsWith(tag, kFloatContextFeaturePrefixTag) &&
           !cc->Inputs().Tag(tag).IsEmpty()) {
         std::string key =
@@ -412,9 +543,36 @@ class PackMediaSequenceCalculator : public CalculatorBase {
                            sizeof(*kFloatContextFeaturePrefixTag) -
                        1);
         RET_CHECK_EQ(cc->InputTimestamp(), Timestamp::PostStream());
-        mpms::SetContextFeatureFloats(
+        for (const auto& value :
-            key, cc->Inputs().Tag(tag).Get<std::vector<float>>(),
+             cc->Inputs().Tag(tag).Get<std::vector<float>>()) {
-            sequence_.get());
+          mpms::AddContextFeatureFloats(key, value, sequence_.get());
+        }
+      }
+      if (absl::StartsWith(tag, kIntsContextFeaturePrefixTag) &&
+          !cc->Inputs().Tag(tag).IsEmpty()) {
+        const std::string& key =
+            tag.substr(sizeof(kIntsContextFeaturePrefixTag) /
+                           sizeof(*kIntsContextFeaturePrefixTag) -
+                       1);
+        // To ensure only one packet is provided for this tag.
+        RET_CHECK_EQ(cc->InputTimestamp(), Timestamp::PostStream());
+        for (const auto& value :
+             cc->Inputs().Tag(tag).Get<std::vector<int64_t>>()) {
+          mpms::AddContextFeatureInts(key, value, sequence_.get());
+        }
+      }
+      if (absl::StartsWith(tag, kBytesContextFeaturePrefixTag) &&
+          !cc->Inputs().Tag(tag).IsEmpty()) {
+        const std::string& key =
+            tag.substr(sizeof(kBytesContextFeaturePrefixTag) /
+                           sizeof(*kBytesContextFeaturePrefixTag) -
+                       1);
+        // To ensure only one packet is provided for this tag.
+        RET_CHECK_EQ(cc->InputTimestamp(), Timestamp::PostStream());
+        for (const auto& value :
+             cc->Inputs().Tag(tag).Get<std::vector<std::string>>()) {
+          mpms::AddContextFeatureBytes(key, value, sequence_.get());
+        }
       }
       if (absl::StartsWith(tag, kFloatFeaturePrefixTag) &&
           !cc->Inputs().Tag(tag).IsEmpty()) {
@@ -460,6 +618,7 @@ class PackMediaSequenceCalculator : public CalculatorBase {
         }
         std::vector<Location> predicted_locations;
         std::vector<std::string> predicted_class_strings;
+        std::vector<float> predicted_class_confidences;
         std::vector<int> predicted_label_ids;
         for (auto& detection :
              cc->Inputs().Tag(tag).Get<std::vector<Detection>>()) {
@@ -488,6 +647,9 @@ class PackMediaSequenceCalculator : public CalculatorBase {
             if (detection.label_id_size() > 0) {
               predicted_label_ids.push_back(detection.label_id(0));
             }
+            if (detection.score_size() > 0) {
+              predicted_class_confidences.push_back(detection.score(0));
+            }
           }
         }
         if (!predicted_locations.empty()) {
@@ -501,6 +663,10 @@ class PackMediaSequenceCalculator : public CalculatorBase {
           if (!predicted_label_ids.empty()) {
             mpms::AddBBoxLabelIndex(key, predicted_label_ids, sequence_.get());
           }
+          if (!predicted_class_confidences.empty()) {
+            mpms::AddBBoxLabelConfidence(key, predicted_class_confidences,
+                                         sequence_.get());
+          }
         }
       }
     }
@@ -548,10 +714,14 @@ class PackMediaSequenceCalculator : public CalculatorBase {
         }
       }
     }
+    if (clip_media_id_.has_value()) {
+      mpms::SetClipMediaId(*clip_media_id_, sequence_.get());
+    }
     return absl::OkStatus();
   }
 
   std::unique_ptr<tf::SequenceExample> sequence_;
+  std::optional<std::string> clip_media_id_ = std::nullopt;
   std::map<std::string, bool> features_present_;
   bool replace_keypoints_;
 };

mediapipe/calculators/tensorflow/pack_media_sequence_calculator_test.cc

@@ -12,28 +12,32 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include <algorithm>
+#include <cstdint>
+#include <memory>
+#include <string>
+#include <vector>
 
-#include "absl/container/flat_hash_map.h"
+#include "absl/log/absl_check.h"
 #include "absl/memory/memory.h"
-#include "absl/strings/numbers.h"
+#include "absl/status/status.h"
+#include "absl/strings/str_cat.h"
 #include "mediapipe/calculators/image/opencv_image_encoder_calculator.pb.h"
 #include "mediapipe/calculators/tensorflow/pack_media_sequence_calculator.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/calculator_runner.h"
 #include "mediapipe/framework/formats/detection.pb.h"
-#include "mediapipe/framework/formats/image_frame.h"
-#include "mediapipe/framework/formats/image_frame_opencv.h"
 #include "mediapipe/framework/formats/location.h"
 #include "mediapipe/framework/formats/location_opencv.h"
-#include "mediapipe/framework/port/gmock.h"
+#include "mediapipe/framework/packet.h"
-#include "mediapipe/framework/port/gtest.h"
 #include "mediapipe/framework/port/opencv_imgcodecs_inc.h"
 #include "mediapipe/framework/port/status_matchers.h"
 #include "mediapipe/framework/timestamp.h"
 #include "mediapipe/util/sequence/media_sequence.h"
+#include "mediapipe/util/sequence/media_sequence_util.h"
 #include "tensorflow/core/example/example.pb.h"
 #include "tensorflow/core/example/feature.pb.h"
+#include "testing/base/public/gmock.h"
+#include "testing/base/public/gunit.h"
 
 namespace mediapipe {
 namespace {
@@ -55,13 +59,22 @@ constexpr char kBytesFeatureTestTag[] = "BYTES_FEATURE_TEST";
 constexpr char kForwardFlowEncodedTag[] = "FORWARD_FLOW_ENCODED";
 constexpr char kFloatContextFeatureOtherTag[] = "FLOAT_CONTEXT_FEATURE_OTHER";
 constexpr char kFloatContextFeatureTestTag[] = "FLOAT_CONTEXT_FEATURE_TEST";
+constexpr char kIntsContextFeatureTestTag[] = "INTS_CONTEXT_FEATURE_TEST";
+constexpr char kIntsContextFeatureOtherTag[] = "INTS_CONTEXT_FEATURE_OTHER";
+constexpr char kBytesContextFeatureTestTag[] = "BYTES_CONTEXT_FEATURE_TEST";
+constexpr char kBytesContextFeatureOtherTag[] = "BYTES_CONTEXT_FEATURE_OTHER";
 constexpr char kFloatFeatureOtherTag[] = "FLOAT_FEATURE_OTHER";
 constexpr char kFloatFeatureTestTag[] = "FLOAT_FEATURE_TEST";
 constexpr char kIntFeatureOtherTag[] = "INT_FEATURE_OTHER";
 constexpr char kIntFeatureTestTag[] = "INT_FEATURE_TEST";
+constexpr char kImageLabelTestTag[] = "IMAGE_LABEL_TEST";
+constexpr char kImageLabelOtherTag[] = "IMAGE_LABEL_OTHER";
 constexpr char kImagePrefixTag[] = "IMAGE_PREFIX";
 constexpr char kSequenceExampleTag[] = "SEQUENCE_EXAMPLE";
 constexpr char kImageTag[] = "IMAGE";
+constexpr char kClipMediaIdTag[] = "CLIP_MEDIA_ID";
+constexpr char kClipLabelTestTag[] = "CLIP_LABEL_TEST";
+constexpr char kClipLabelOtherTag[] = "CLIP_LABEL_OTHER";
 
 class PackMediaSequenceCalculatorTest : public ::testing::Test {
  protected:
@@ -69,10 +82,14 @@ class PackMediaSequenceCalculatorTest : public ::testing::Test {
                        const tf::Features& features,
                        const bool output_only_if_all_present,
                        const bool replace_instead_of_append,
-                       const bool output_as_zero_timestamp = false) {
+                       const bool output_as_zero_timestamp = false,
+                       const std::vector<std::string>& input_side_packets = {
+                           "SEQUENCE_EXAMPLE:input_sequence"}) {
     CalculatorGraphConfig::Node config;
     config.set_calculator("PackMediaSequenceCalculator");
-    config.add_input_side_packet("SEQUENCE_EXAMPLE:input_sequence");
+    for (const std::string& side_packet : input_side_packets) {
+      config.add_input_side_packet(side_packet);
+    }
     config.add_output_stream("SEQUENCE_EXAMPLE:output_sequence");
     for (const std::string& stream : input_streams) {
       config.add_input_stream(stream);
@@ -96,7 +113,8 @@ TEST_F(PackMediaSequenceCalculatorTest, PacksTwoImages) {
   mpms::SetClipMediaId(test_video_id, input_sequence.get());
   cv::Mat image(2, 3, CV_8UC3, cv::Scalar(0, 0, 255));
   std::vector<uchar> bytes;
-  ASSERT_TRUE(cv::imencode(".jpg", image, bytes, {80}));
+  ASSERT_TRUE(
+      cv::imencode(".jpg", image, bytes, {cv::IMWRITE_HDR_COMPRESSION, 1}));
   OpenCvImageEncoderCalculatorResults encoded_image;
   encoded_image.set_encoded_image(bytes.data(), bytes.size());
   encoded_image.set_width(2);
@@ -139,7 +157,8 @@ TEST_F(PackMediaSequenceCalculatorTest, PacksTwoPrefixedImages) {
   mpms::SetClipMediaId(test_video_id, input_sequence.get());
   cv::Mat image(2, 3, CV_8UC3, cv::Scalar(0, 0, 255));
   std::vector<uchar> bytes;
-  ASSERT_TRUE(cv::imencode(".jpg", image, bytes, {80}));
+  ASSERT_TRUE(
+      cv::imencode(".jpg", image, bytes, {cv::IMWRITE_HDR_COMPRESSION, 1}));
   OpenCvImageEncoderCalculatorResults encoded_image;
   encoded_image.set_encoded_image(bytes.data(), bytes.size());
   encoded_image.set_width(2);
@@ -312,6 +331,76 @@ TEST_F(PackMediaSequenceCalculatorTest, PacksTwoBytesLists) {
   }
 }
 
+TEST_F(PackMediaSequenceCalculatorTest, PacksTwoImageLabels) {
+  SetUpCalculator(
+      {"IMAGE_LABEL_TEST:test_labels", "IMAGE_LABEL_OTHER:test_labels2"}, {},
+      false, true);
+  auto input_sequence = ::absl::make_unique<tf::SequenceExample>();
+
+  int num_timesteps = 2;
+  for (int i = 0; i < num_timesteps; ++i) {
+    Detection detection1;
+    detection1.add_label(absl::StrCat("foo", 2 << i));
+    detection1.add_label_id(i);
+    detection1.add_score(0.1 * i);
+    detection1.add_label(absl::StrCat("foo", 2 << i));
+    detection1.add_label_id(i);
+    detection1.add_score(0.1 * i);
+    auto label_ptr1 = ::absl::make_unique<Detection>(detection1);
+    runner_->MutableInputs()
+        ->Tag(kImageLabelTestTag)
+        .packets.push_back(Adopt(label_ptr1.release()).At(Timestamp(i)));
+    Detection detection2;
+    detection2.add_label(absl::StrCat("bar", 2 << i));
+    detection2.add_score(0.2 * i);
+    detection2.add_label(absl::StrCat("bar", 2 << i));
+    detection2.add_score(0.2 * i);
+    auto label_ptr2 = ::absl::make_unique<Detection>(detection2);
+    runner_->MutableInputs()
+        ->Tag(kImageLabelOtherTag)
+        .packets.push_back(Adopt(label_ptr2.release()).At(Timestamp(i)));
+  }
+  runner_->MutableSidePackets()->Tag(kSequenceExampleTag) =
+      Adopt(input_sequence.release());
+
+  MP_ASSERT_OK(runner_->Run());
+
+  const std::vector<Packet>& output_packets =
+      runner_->Outputs().Tag(kSequenceExampleTag).packets;
+  ASSERT_EQ(1, output_packets.size());
+  const tf::SequenceExample& output_sequence =
+      output_packets[0].Get<tf::SequenceExample>();
+
+  ASSERT_EQ(num_timesteps,
+            mpms::GetImageTimestampSize("TEST", output_sequence));
+  ASSERT_EQ(num_timesteps,
+            mpms::GetImageLabelStringSize("TEST", output_sequence));
+  ASSERT_EQ(num_timesteps,
+            mpms::GetImageLabelConfidenceSize("TEST", output_sequence));
+  ASSERT_EQ(num_timesteps,
+            mpms::GetImageTimestampSize("OTHER", output_sequence));
+  ASSERT_EQ(num_timesteps,
+            mpms::GetImageLabelStringSize("OTHER", output_sequence));
+  ASSERT_EQ(num_timesteps,
+            mpms::GetImageLabelConfidenceSize("OTHER", output_sequence));
+  for (int i = 0; i < num_timesteps; ++i) {
+    ASSERT_EQ(i, mpms::GetImageTimestampAt("TEST", output_sequence, i));
+    ASSERT_THAT(mpms::GetImageLabelStringAt("TEST", output_sequence, i),
+                ::testing::ElementsAreArray(
+                    std::vector<std::string>(2, absl::StrCat("foo", 2 << i))));
+    ASSERT_THAT(mpms::GetImageLabelIndexAt("TEST", output_sequence, i),
+                ::testing::ElementsAreArray(std::vector<int32_t>(2, i)));
+    ASSERT_THAT(mpms::GetImageLabelConfidenceAt("TEST", output_sequence, i),
+                ::testing::ElementsAreArray(std::vector<float>(2, 0.1 * i)));
+    ASSERT_EQ(i, mpms::GetImageTimestampAt("OTHER", output_sequence, i));
+    ASSERT_THAT(mpms::GetImageLabelStringAt("OTHER", output_sequence, i),
+                ::testing::ElementsAreArray(
+                    std::vector<std::string>(2, absl::StrCat("bar", 2 << i))));
+    ASSERT_THAT(mpms::GetImageLabelConfidenceAt("OTHER", output_sequence, i),
+                ::testing::ElementsAreArray(std::vector<float>(2, 0.2 * i)));
+  }
+}
+
 TEST_F(PackMediaSequenceCalculatorTest, OutputAsZeroTimestamp) {
   SetUpCalculator({"FLOAT_FEATURE_TEST:test"}, {}, false, true, true);
   auto input_sequence = ::absl::make_unique<tf::SequenceExample>();
@@ -367,6 +456,315 @@ TEST_F(PackMediaSequenceCalculatorTest, PacksTwoContextFloatLists) {
               testing::ElementsAre(4, 4));
 }
 
+TEST_F(PackMediaSequenceCalculatorTest, ReplaceTwoContextFloatLists) {
+  SetUpCalculator(
+      /*input_streams=*/{"FLOAT_CONTEXT_FEATURE_TEST:test",
+                         "FLOAT_CONTEXT_FEATURE_OTHER:test2"},
+      /*features=*/{},
+      /*output_only_if_all_present=*/false, /*replace_instead_of_append=*/true);
+  auto input_sequence = std::make_unique<tf::SequenceExample>();
+  mpms::SetContextFeatureFloats("TEST", {2, 3}, input_sequence.get());
+  mpms::SetContextFeatureFloats("OTHER", {2, 4}, input_sequence.get());
+
+  const std::vector<float> vf_1 = {5, 6};
+  runner_->MutableInputs()
+      ->Tag(kFloatContextFeatureTestTag)
+      .packets.push_back(
+          MakePacket<std::vector<float>>(vf_1).At(Timestamp::PostStream()));
+  const std::vector<float> vf_2 = {7, 8};
+  runner_->MutableInputs()
+      ->Tag(kFloatContextFeatureOtherTag)
+      .packets.push_back(
+          MakePacket<std::vector<float>>(vf_2).At(Timestamp::PostStream()));
+
+  runner_->MutableSidePackets()->Tag(kSequenceExampleTag) =
+      Adopt(input_sequence.release());
+
+  MP_ASSERT_OK(runner_->Run());
+
+  const std::vector<Packet>& output_packets =
+      runner_->Outputs().Tag(kSequenceExampleTag).packets;
+  ASSERT_EQ(1, output_packets.size());
+  const tf::SequenceExample& output_sequence =
+      output_packets[0].Get<tf::SequenceExample>();
+
+  ASSERT_THAT(mpms::GetContextFeatureFloats("TEST", output_sequence),
+              testing::ElementsAre(5, 6));
+  ASSERT_THAT(mpms::GetContextFeatureFloats("OTHER", output_sequence),
+              testing::ElementsAre(7, 8));
+}
+
+TEST_F(PackMediaSequenceCalculatorTest, AppendTwoContextFloatLists) {
+  SetUpCalculator(
+      /*input_streams=*/{"FLOAT_CONTEXT_FEATURE_TEST:test",
+                         "FLOAT_CONTEXT_FEATURE_OTHER:test2"},
+      /*features=*/{},
+      /*output_only_if_all_present=*/false,
+      /*replace_instead_of_append=*/false);
+  auto input_sequence = std::make_unique<tf::SequenceExample>();
+  mpms::SetContextFeatureFloats("TEST", {2, 3}, input_sequence.get());
+  mpms::SetContextFeatureFloats("OTHER", {2, 4}, input_sequence.get());
+
+  const std::vector<float> vf_1 = {5, 6};
+  runner_->MutableInputs()
+      ->Tag(kFloatContextFeatureTestTag)
+      .packets.push_back(
+          MakePacket<std::vector<float>>(vf_1).At(Timestamp::PostStream()));
+  const std::vector<float> vf_2 = {7, 8};
+  runner_->MutableInputs()
+      ->Tag(kFloatContextFeatureOtherTag)
+      .packets.push_back(
+          MakePacket<std::vector<float>>(vf_2).At(Timestamp::PostStream()));
+
+  runner_->MutableSidePackets()->Tag(kSequenceExampleTag) =
+      Adopt(input_sequence.release());
+
+  MP_ASSERT_OK(runner_->Run());
+
+  const std::vector<Packet>& output_packets =
+      runner_->Outputs().Tag(kSequenceExampleTag).packets;
+  ASSERT_EQ(1, output_packets.size());
+  const tf::SequenceExample& output_sequence =
+      output_packets[0].Get<tf::SequenceExample>();
+
+  EXPECT_THAT(mpms::GetContextFeatureFloats("TEST", output_sequence),
+              testing::ElementsAre(2, 3, 5, 6));
+  EXPECT_THAT(mpms::GetContextFeatureFloats("OTHER", output_sequence),
+              testing::ElementsAre(2, 4, 7, 8));
+}
+
+TEST_F(PackMediaSequenceCalculatorTest, PackTwoContextIntLists) {
+  SetUpCalculator(
+      /*input_streams=*/{"INTS_CONTEXT_FEATURE_TEST:test",
+                         "INTS_CONTEXT_FEATURE_OTHER:test2"},
+      /*features=*/{},
+      /*output_only_if_all_present=*/false, /*replace_instead_of_append=*/true);
+  auto input_sequence = absl::make_unique<tf::SequenceExample>();
+
+  const std::vector<int64_t> vi_1 = {2, 3};
+  runner_->MutableInputs()
+      ->Tag(kIntsContextFeatureTestTag)
+      .packets.push_back(
+          MakePacket<std::vector<int64_t>>(vi_1).At(Timestamp::PostStream()));
+  const std::vector<int64_t> vi_2 = {2, 4};
+  runner_->MutableInputs()
+      ->Tag(kIntsContextFeatureOtherTag)
+      .packets.push_back(
+          MakePacket<std::vector<int64_t>>(vi_2).At(Timestamp::PostStream()));
+
+  runner_->MutableSidePackets()->Tag(kSequenceExampleTag) =
+      Adopt(input_sequence.release());
+
+  MP_ASSERT_OK(runner_->Run());
+
+  const std::vector<Packet>& output_packets =
+      runner_->Outputs().Tag(kSequenceExampleTag).packets;
+  ASSERT_EQ(1, output_packets.size());
+  const tf::SequenceExample& output_sequence =
+      output_packets[0].Get<tf::SequenceExample>();
+
+  ASSERT_THAT(mpms::GetContextFeatureInts("TEST", output_sequence),
+              testing::ElementsAre(2, 3));
+  ASSERT_THAT(mpms::GetContextFeatureInts("OTHER", output_sequence),
+              testing::ElementsAre(2, 4));
+}
+
+TEST_F(PackMediaSequenceCalculatorTest, ReplaceTwoContextIntLists) {
+  SetUpCalculator(
+      /*input_streams=*/{"INTS_CONTEXT_FEATURE_TEST:test",
+                         "INTS_CONTEXT_FEATURE_OTHER:test2"},
+      /*features=*/{},
+      /*output_only_if_all_present=*/false, /*replace_instead_of_append=*/true);
+  auto input_sequence = absl::make_unique<tf::SequenceExample>();
+  mpms::SetContextFeatureInts("TEST", {2, 3}, input_sequence.get());
+  mpms::SetContextFeatureInts("OTHER", {2, 4}, input_sequence.get());
+
+  const std::vector<int64_t> vi_1 = {5, 6};
+  runner_->MutableInputs()
+      ->Tag(kIntsContextFeatureTestTag)
+      .packets.push_back(
+          MakePacket<std::vector<int64_t>>(vi_1).At(Timestamp::PostStream()));
+  const std::vector<int64_t> vi_2 = {7, 8};
+  runner_->MutableInputs()
+      ->Tag(kIntsContextFeatureOtherTag)
+      .packets.push_back(
+          MakePacket<std::vector<int64_t>>(vi_2).At(Timestamp::PostStream()));
+
+  runner_->MutableSidePackets()->Tag(kSequenceExampleTag) =
+      Adopt(input_sequence.release());
+
+  MP_ASSERT_OK(runner_->Run());
+
+  const std::vector<Packet>& output_packets =
+      runner_->Outputs().Tag(kSequenceExampleTag).packets;
+  ASSERT_EQ(1, output_packets.size());
+  const tf::SequenceExample& output_sequence =
+      output_packets[0].Get<tf::SequenceExample>();
+
+  ASSERT_THAT(mpms::GetContextFeatureInts("TEST", output_sequence),
+              testing::ElementsAre(5, 6));
+  ASSERT_THAT(mpms::GetContextFeatureInts("OTHER", output_sequence),
+              testing::ElementsAre(7, 8));
+}
+
+TEST_F(PackMediaSequenceCalculatorTest, AppendTwoContextIntLists) {
+  SetUpCalculator(
+      /*input_streams=*/{"INTS_CONTEXT_FEATURE_TEST:test",
+                         "INTS_CONTEXT_FEATURE_OTHER:test2"},
+      /*features=*/{},
+      /*output_only_if_all_present=*/false,
+      /*replace_instead_of_append=*/false);
+  auto input_sequence = absl::make_unique<tf::SequenceExample>();
+  mpms::SetContextFeatureInts("TEST", {2, 3}, input_sequence.get());
+  mpms::SetContextFeatureInts("OTHER", {2, 4}, input_sequence.get());
+
+  const std::vector<int64_t> vi_1 = {5, 6};
+  runner_->MutableInputs()
+      ->Tag(kIntsContextFeatureTestTag)
+      .packets.push_back(
+          MakePacket<std::vector<int64_t>>(vi_1).At(Timestamp::PostStream()));
+  const std::vector<int64_t> vi_2 = {7, 8};
+  runner_->MutableInputs()
+      ->Tag(kIntsContextFeatureOtherTag)
+      .packets.push_back(
+          MakePacket<std::vector<int64_t>>(vi_2).At(Timestamp::PostStream()));
+
+  runner_->MutableSidePackets()->Tag(kSequenceExampleTag) =
+      Adopt(input_sequence.release());
+
+  MP_ASSERT_OK(runner_->Run());
+
+  const std::vector<Packet>& output_packets =
+      runner_->Outputs().Tag(kSequenceExampleTag).packets;
+  ASSERT_EQ(1, output_packets.size());
+  const tf::SequenceExample& output_sequence =
+      output_packets[0].Get<tf::SequenceExample>();
+
+  ASSERT_THAT(mpms::GetContextFeatureInts("TEST", output_sequence),
+              testing::ElementsAre(2, 3, 5, 6));
+  ASSERT_THAT(mpms::GetContextFeatureInts("OTHER", output_sequence),
+              testing::ElementsAre(2, 4, 7, 8));
+}
+
+TEST_F(PackMediaSequenceCalculatorTest, PackTwoContextByteLists) {
+  SetUpCalculator(
+      /*input_streams=*/{"BYTES_CONTEXT_FEATURE_TEST:test",
+                         "BYTES_CONTEXT_FEATURE_OTHER:test2"},
+      /*features=*/{},
+      /*output_only_if_all_present=*/false, /*replace_instead_of_append=*/true);
+  auto input_sequence = absl::make_unique<tf::SequenceExample>();
+
+  const std::vector<std::string> vb_1 = {"value_1", "value_2"};
+  runner_->MutableInputs()
+      ->Tag(kBytesContextFeatureTestTag)
+      .packets.push_back(MakePacket<std::vector<std::string>>(vb_1).At(
+          Timestamp::PostStream()));
+  const std::vector<std::string> vb_2 = {"value_3", "value_4"};
+  runner_->MutableInputs()
+      ->Tag(kBytesContextFeatureOtherTag)
+      .packets.push_back(MakePacket<std::vector<std::string>>(vb_2).At(
+          Timestamp::PostStream()));
+
+  runner_->MutableSidePackets()->Tag(kSequenceExampleTag) =
+      Adopt(input_sequence.release());
+
+  MP_ASSERT_OK(runner_->Run());
+
+  const std::vector<Packet>& output_packets =
+      runner_->Outputs().Tag(kSequenceExampleTag).packets;
+  ASSERT_EQ(1, output_packets.size());
+  const tf::SequenceExample& output_sequence =
+      output_packets[0].Get<tf::SequenceExample>();
+
+  ASSERT_THAT(mpms::GetContextFeatureBytes("TEST", output_sequence),
+              testing::ElementsAre("value_1", "value_2"));
+  ASSERT_THAT(mpms::GetContextFeatureBytes("OTHER", output_sequence),
+              testing::ElementsAre("value_3", "value_4"));
+}
+
+TEST_F(PackMediaSequenceCalculatorTest, ReplaceTwoContextByteLists) {
+  SetUpCalculator(
+      /*input_streams=*/{"BYTES_CONTEXT_FEATURE_TEST:test",
+                         "BYTES_CONTEXT_FEATURE_OTHER:test2"},
+      /*features=*/{},
+      /*output_only_if_all_present=*/false, /*replace_instead_of_append=*/true);
+  auto input_sequence = absl::make_unique<tf::SequenceExample>();
+  mpms::SetContextFeatureBytes("TEST", {"existing_value_1", "existing_value_2"},
+                               input_sequence.get());
+  mpms::SetContextFeatureBytes(
+      "OTHER", {"existing_value_3", "existing_value_4"}, input_sequence.get());
+
+  const std::vector<std::string> vb_1 = {"value_1", "value_2"};
+  runner_->MutableInputs()
+      ->Tag(kBytesContextFeatureTestTag)
+      .packets.push_back(MakePacket<std::vector<std::string>>(vb_1).At(
+          Timestamp::PostStream()));
+  const std::vector<std::string> vb_2 = {"value_3", "value_4"};
+  runner_->MutableInputs()
+      ->Tag(kBytesContextFeatureOtherTag)
+      .packets.push_back(MakePacket<std::vector<std::string>>(vb_2).At(
+          Timestamp::PostStream()));
+
+  runner_->MutableSidePackets()->Tag(kSequenceExampleTag) =
+      Adopt(input_sequence.release());
+
+  MP_ASSERT_OK(runner_->Run());
+
+  const std::vector<Packet>& output_packets =
+      runner_->Outputs().Tag(kSequenceExampleTag).packets;
+  ASSERT_EQ(1, output_packets.size());
+  const tf::SequenceExample& output_sequence =
+      output_packets[0].Get<tf::SequenceExample>();
+
+  ASSERT_THAT(mpms::GetContextFeatureBytes("TEST", output_sequence),
+              testing::ElementsAre("value_1", "value_2"));
+  ASSERT_THAT(mpms::GetContextFeatureBytes("OTHER", output_sequence),
+              testing::ElementsAre("value_3", "value_4"));
+}
+
+TEST_F(PackMediaSequenceCalculatorTest, AppendTwoContextByteLists) {
+  SetUpCalculator(
+      /*input_streams=*/{"BYTES_CONTEXT_FEATURE_TEST:test",
+                         "BYTES_CONTEXT_FEATURE_OTHER:test2"},
+      /*features=*/{},
+      /*output_only_if_all_present=*/false,
+      /*replace_instead_of_append=*/false);
+  auto input_sequence = absl::make_unique<tf::SequenceExample>();
+  mpms::SetContextFeatureBytes("TEST", {"existing_value_1", "existing_value_2"},
+                               input_sequence.get());
+  mpms::SetContextFeatureBytes(
+      "OTHER", {"existing_value_3", "existing_value_4"}, input_sequence.get());
+
+  const std::vector<std::string> vb_1 = {"value_1", "value_2"};
+  runner_->MutableInputs()
+      ->Tag(kBytesContextFeatureTestTag)
+      .packets.push_back(MakePacket<std::vector<std::string>>(vb_1).At(
+          Timestamp::PostStream()));
+  const std::vector<std::string> vb_2 = {"value_3", "value_4"};
+  runner_->MutableInputs()
+      ->Tag(kBytesContextFeatureOtherTag)
+      .packets.push_back(MakePacket<std::vector<std::string>>(vb_2).At(
+          Timestamp::PostStream()));
+
+  runner_->MutableSidePackets()->Tag(kSequenceExampleTag) =
+      Adopt(input_sequence.release());
+
+  MP_ASSERT_OK(runner_->Run());
+
+  const std::vector<Packet>& output_packets =
+      runner_->Outputs().Tag(kSequenceExampleTag).packets;
+  ASSERT_EQ(1, output_packets.size());
+  const tf::SequenceExample& output_sequence =
+      output_packets[0].Get<tf::SequenceExample>();
+
+  ASSERT_THAT(mpms::GetContextFeatureBytes("TEST", output_sequence),
+              testing::ElementsAre("existing_value_1", "existing_value_2",
+                                   "value_1", "value_2"));
+  ASSERT_THAT(mpms::GetContextFeatureBytes("OTHER", output_sequence),
+              testing::ElementsAre("existing_value_3", "existing_value_4",
+                                   "value_3", "value_4"));
+}
+
 TEST_F(PackMediaSequenceCalculatorTest, PacksAdditionalContext) {
   tf::Features context;
   (*context.mutable_feature())["TEST"].mutable_bytes_list()->add_value("YES");
@@ -378,7 +776,8 @@ TEST_F(PackMediaSequenceCalculatorTest, PacksAdditionalContext) {
       Adopt(input_sequence.release());
   cv::Mat image(2, 3, CV_8UC3, cv::Scalar(0, 0, 255));
   std::vector<uchar> bytes;
-  ASSERT_TRUE(cv::imencode(".jpg", image, bytes, {80}));
+  ASSERT_TRUE(
+      cv::imencode(".jpg", image, bytes, {cv::IMWRITE_HDR_COMPRESSION, 1}));
   OpenCvImageEncoderCalculatorResults encoded_image;
   encoded_image.set_encoded_image(bytes.data(), bytes.size());
   auto image_ptr =
@@ -410,7 +809,8 @@ TEST_F(PackMediaSequenceCalculatorTest, PacksTwoForwardFlowEncodeds) {
 
   cv::Mat image(2, 3, CV_8UC3, cv::Scalar(0, 0, 255));
   std::vector<uchar> bytes;
-  ASSERT_TRUE(cv::imencode(".jpg", image, bytes, {80}));
+  ASSERT_TRUE(
+      cv::imencode(".jpg", image, bytes, {cv::IMWRITE_HDR_COMPRESSION, 1}));
   std::string test_flow_string(bytes.begin(), bytes.end());
   OpenCvImageEncoderCalculatorResults encoded_flow;
   encoded_flow.set_encoded_image(test_flow_string);
@@ -526,6 +926,10 @@ TEST_F(PackMediaSequenceCalculatorTest, PacksTwoBBoxDetections) {
     auto class_indices = mpms::GetPredictedBBoxLabelIndexAt(output_sequence, i);
     ASSERT_EQ(0, class_indices[0]);
     ASSERT_EQ(1, class_indices[1]);
+    auto class_scores =
+        mpms::GetPredictedBBoxLabelConfidenceAt(output_sequence, i);
+    ASSERT_FLOAT_EQ(0.5, class_scores[0]);
+    ASSERT_FLOAT_EQ(0.75, class_scores[1]);
   }
 }
 
@@ -618,7 +1022,8 @@ TEST_F(PackMediaSequenceCalculatorTest, PacksBBoxWithImages) {
   }
   cv::Mat image(height, width, CV_8UC3, cv::Scalar(0, 0, 255));
   std::vector<uchar> bytes;
-  ASSERT_TRUE(cv::imencode(".jpg", image, bytes, {80}));
+  ASSERT_TRUE(
+      cv::imencode(".jpg", image, bytes, {cv::IMWRITE_HDR_COMPRESSION, 1}));
   OpenCvImageEncoderCalculatorResults encoded_image;
   encoded_image.set_encoded_image(bytes.data(), bytes.size());
   encoded_image.set_width(width);
@@ -667,6 +1072,10 @@ TEST_F(PackMediaSequenceCalculatorTest, PacksBBoxWithImages) {
     auto class_indices = mpms::GetPredictedBBoxLabelIndexAt(output_sequence, i);
     ASSERT_EQ(0, class_indices[0]);
     ASSERT_EQ(1, class_indices[1]);
+    auto class_scores =
+        mpms::GetPredictedBBoxLabelConfidenceAt(output_sequence, i);
+    ASSERT_FLOAT_EQ(0.5, class_scores[0]);
+    ASSERT_FLOAT_EQ(0.75, class_scores[1]);
   }
 }
 
@@ -757,6 +1166,365 @@ TEST_F(PackMediaSequenceCalculatorTest, PacksTwoMaskDetections) {
               testing::ElementsAreArray(::std::vector<std::string>({"mask"})));
 }
 
+TEST_F(PackMediaSequenceCalculatorTest, PackTwoClipLabels) {
+  SetUpCalculator(
+      /*input_streams=*/{"CLIP_LABEL_TEST:test", "CLIP_LABEL_OTHER:test2"},
+      /*features=*/{}, /*output_only_if_all_present=*/false,
+      /*replace_instead_of_append=*/true);
+  auto input_sequence = ::absl::make_unique<tf::SequenceExample>();
+
+  Detection detection_1;
+  detection_1.add_label("label_1");
+  detection_1.add_label("label_2");
+  detection_1.add_label_id(1);
+  detection_1.add_label_id(2);
+  detection_1.add_score(0.1);
+  detection_1.add_score(0.2);
+  runner_->MutableInputs()
+      ->Tag(kClipLabelTestTag)
+      .packets.push_back(MakePacket<Detection>(detection_1).At(Timestamp(1)));
+  // No label ID for detection_2.
+  Detection detection_2;
+  detection_2.add_label("label_3");
+  detection_2.add_label("label_4");
+  detection_2.add_score(0.3);
+  detection_2.add_score(0.4);
+  runner_->MutableInputs()
+      ->Tag(kClipLabelOtherTag)
+      .packets.push_back(MakePacket<Detection>(detection_2).At(Timestamp(2)));
+  runner_->MutableSidePackets()->Tag(kSequenceExampleTag) =
+      Adopt(input_sequence.release());
+
+  MP_ASSERT_OK(runner_->Run());
+
+  const std::vector<Packet>& output_packets =
+      runner_->Outputs().Tag(kSequenceExampleTag).packets;
+  ASSERT_EQ(1, output_packets.size());
+  const tf::SequenceExample& output_sequence =
+      output_packets[0].Get<tf::SequenceExample>();
+
+  ASSERT_THAT(mpms::GetClipLabelString("TEST", output_sequence),
+              testing::ElementsAre("label_1", "label_2"));
+  ASSERT_THAT(mpms::GetClipLabelIndex("TEST", output_sequence),
+              testing::ElementsAre(1, 2));
+  ASSERT_THAT(mpms::GetClipLabelConfidence("TEST", output_sequence),
+              testing::ElementsAre(0.1, 0.2));
+  ASSERT_THAT(mpms::GetClipLabelString("OTHER", output_sequence),
+              testing::ElementsAre("label_3", "label_4"));
+  ASSERT_FALSE(mpms::HasClipLabelIndex("OTHER", output_sequence));
+  ASSERT_THAT(mpms::GetClipLabelConfidence("OTHER", output_sequence),
+              testing::ElementsAre(0.3, 0.4));
+}
+
+TEST_F(PackMediaSequenceCalculatorTest,
+       PackTwoClipLabels_DifferentLabelScoreSize) {
+  SetUpCalculator(
+      /*input_streams=*/{"CLIP_LABEL_TEST:test", "CLIP_LABEL_OTHER:test2"},
+      /*features=*/{}, /*output_only_if_all_present=*/false,
+      /*replace_instead_of_append=*/true);
+  auto input_sequence = ::absl::make_unique<tf::SequenceExample>();
+
+  // 2 labels and 1 score in detection_1.
+  Detection detection_1;
+  detection_1.add_label("label_1");
+  detection_1.add_label("label_2");
+  detection_1.add_score(0.1);
+  runner_->MutableInputs()
+      ->Tag(kClipLabelTestTag)
+      .packets.push_back(MakePacket<Detection>(detection_1).At(Timestamp(1)));
+  Detection detection_2;
+  detection_2.add_label("label_3");
+  detection_2.add_label("label_4");
+  detection_2.add_score(0.3);
+  detection_2.add_score(0.4);
+  runner_->MutableInputs()
+      ->Tag(kClipLabelOtherTag)
+      .packets.push_back(MakePacket<Detection>(detection_2).At(Timestamp(2)));
+  runner_->MutableSidePackets()->Tag(kSequenceExampleTag) =
+      Adopt(input_sequence.release());
+
+  ASSERT_THAT(
+      runner_->Run(),
+      testing::status::StatusIs(
+          absl::StatusCode::kInvalidArgument,
+          testing::HasSubstr(
+              "Different size of detection.label and detection.score")));
+}
+
+TEST_F(PackMediaSequenceCalculatorTest,
+       PackTwoClipLabels_DifferentLabelIdSize) {
+  SetUpCalculator(
+      /*input_streams=*/{"CLIP_LABEL_TEST:test", "CLIP_LABEL_OTHER:test2"},
+      /*features=*/{}, /*output_only_if_all_present=*/false,
+      /*replace_instead_of_append=*/true);
+  auto input_sequence = ::absl::make_unique<tf::SequenceExample>();
+
+  // 2 labels and 1 label_id in detection_1.
+  Detection detection_1;
+  detection_1.add_label("label_1");
+  detection_1.add_label("label_2");
+  detection_1.add_label_id(1);
+  detection_1.add_score(0.1);
+  detection_1.add_score(0.2);
+  runner_->MutableInputs()
+      ->Tag(kClipLabelTestTag)
+      .packets.push_back(MakePacket<Detection>(detection_1).At(Timestamp(1)));
+  Detection detection_2;
+  detection_2.add_label("label_3");
+  detection_2.add_label("label_4");
+  detection_2.add_score(0.3);
+  detection_2.add_score(0.4);
+  runner_->MutableInputs()
+      ->Tag(kClipLabelOtherTag)
+      .packets.push_back(MakePacket<Detection>(detection_2).At(Timestamp(2)));
+  runner_->MutableSidePackets()->Tag(kSequenceExampleTag) =
+      Adopt(input_sequence.release());
+
+  ASSERT_THAT(
+      runner_->Run(),
+      testing::status::StatusIs(
+          absl::StatusCode::kInvalidArgument,
+          testing::HasSubstr(
+              "Different size of detection.label_id and detection.label")));
+}
+
+TEST_F(PackMediaSequenceCalculatorTest, ReplaceTwoClipLabels) {
+  // Replace existing clip/label/string and clip/label/confidence values for
+  // the prefixes.
+  SetUpCalculator(
+      /*input_streams=*/{"CLIP_LABEL_TEST:test", "CLIP_LABEL_OTHER:test2"},
+      /*features=*/{}, /*output_only_if_all_present=*/false,
+      /*replace_instead_of_append=*/true);
+  auto input_sequence = ::absl::make_unique<tf::SequenceExample>();
+  mpms::SetClipLabelString("TEST", {"old_label_1", "old_label_2"},
+                           input_sequence.get());
+  mpms::SetClipLabelConfidence("TEST", {0.1, 0.2}, input_sequence.get());
+  mpms::SetClipLabelString("OTHER", {"old_label_3", "old_label_4"},
+                           input_sequence.get());
+  mpms::SetClipLabelConfidence("OTHER", {0.3, 0.4}, input_sequence.get());
+
+  Detection detection_1;
+  detection_1.add_label("label_1");
+  detection_1.add_label("label_2");
+  detection_1.add_label_id(1);
+  detection_1.add_label_id(2);
+  detection_1.add_score(0.9);
+  detection_1.add_score(0.8);
+  runner_->MutableInputs()
+      ->Tag(kClipLabelTestTag)
+      .packets.push_back(MakePacket<Detection>(detection_1).At(Timestamp(1)));
+  Detection detection_2;
+  detection_2.add_label("label_3");
+  detection_2.add_label("label_4");
+  detection_2.add_label_id(3);
+  detection_2.add_label_id(4);
+  detection_2.add_score(0.7);
+  detection_2.add_score(0.6);
+  runner_->MutableInputs()
+      ->Tag(kClipLabelOtherTag)
+      .packets.push_back(MakePacket<Detection>(detection_2).At(Timestamp(2)));
+  runner_->MutableSidePackets()->Tag(kSequenceExampleTag) =
+      Adopt(input_sequence.release());
+
+  MP_ASSERT_OK(runner_->Run());
+
+  const std::vector<Packet>& output_packets =
+      runner_->Outputs().Tag(kSequenceExampleTag).packets;
+  ASSERT_EQ(1, output_packets.size());
+  const tf::SequenceExample& output_sequence =
+      output_packets[0].Get<tf::SequenceExample>();
+
+  ASSERT_THAT(mpms::GetClipLabelString("TEST", output_sequence),
+              testing::ElementsAre("label_1", "label_2"));
+  ASSERT_THAT(mpms::GetClipLabelIndex("TEST", output_sequence),
+              testing::ElementsAre(1, 2));
+  ASSERT_THAT(mpms::GetClipLabelConfidence("TEST", output_sequence),
+              testing::ElementsAre(0.9, 0.8));
+  ASSERT_THAT(mpms::GetClipLabelString("OTHER", output_sequence),
+              testing::ElementsAre("label_3", "label_4"));
+  ASSERT_THAT(mpms::GetClipLabelIndex("OTHER", output_sequence),
+              testing::ElementsAre(3, 4));
+  ASSERT_THAT(mpms::GetClipLabelConfidence("OTHER", output_sequence),
+              testing::ElementsAre(0.7, 0.6));
+}
+
+TEST_F(PackMediaSequenceCalculatorTest, AppendTwoClipLabels) {
+  // Append to the existing clip/label/string and clip/label/confidence values
+  // for the prefixes.
+  SetUpCalculator(
+      /*input_streams=*/{"CLIP_LABEL_TEST:test", "CLIP_LABEL_OTHER:test2"},
+      /*features=*/{}, /*output_only_if_all_present=*/false,
+      /*replace_instead_of_append=*/false);
+  auto input_sequence = ::absl::make_unique<tf::SequenceExample>();
+  mpms::SetClipLabelString("TEST", {"old_label_1", "old_label_2"},
+                           input_sequence.get());
+  mpms::SetClipLabelIndex("TEST", {1, 2}, input_sequence.get());
+  mpms::SetClipLabelConfidence("TEST", {0.1, 0.2}, input_sequence.get());
+  mpms::SetClipLabelString("OTHER", {"old_label_3", "old_label_4"},
+                           input_sequence.get());
+  mpms::SetClipLabelIndex("OTHER", {3, 4}, input_sequence.get());
+  mpms::SetClipLabelConfidence("OTHER", {0.3, 0.4}, input_sequence.get());
+
+  Detection detection_1;
+  detection_1.add_label("label_1");
+  detection_1.add_label("label_2");
+  detection_1.add_label_id(9);
+  detection_1.add_label_id(8);
+  detection_1.add_score(0.9);
+  detection_1.add_score(0.8);
+  runner_->MutableInputs()
+      ->Tag(kClipLabelTestTag)
+      .packets.push_back(MakePacket<Detection>(detection_1).At(Timestamp(1)));
+  Detection detection_2;
+  detection_2.add_label("label_3");
+  detection_2.add_label("label_4");
+  detection_2.add_label_id(7);
+  detection_2.add_label_id(6);
+  detection_2.add_score(0.7);
+  detection_2.add_score(0.6);
+  runner_->MutableInputs()
+      ->Tag(kClipLabelOtherTag)
+      .packets.push_back(MakePacket<Detection>(detection_2).At(Timestamp(2)));
+  runner_->MutableSidePackets()->Tag(kSequenceExampleTag) =
+      Adopt(input_sequence.release());
+
+  MP_ASSERT_OK(runner_->Run());
+
+  const std::vector<Packet>& output_packets =
+      runner_->Outputs().Tag(kSequenceExampleTag).packets;
+  ASSERT_EQ(1, output_packets.size());
+  const tf::SequenceExample& output_sequence =
+      output_packets[0].Get<tf::SequenceExample>();
+
+  ASSERT_THAT(
+      mpms::GetClipLabelString("TEST", output_sequence),
+      testing::ElementsAre("old_label_1", "old_label_2", "label_1", "label_2"));
+  ASSERT_THAT(mpms::GetClipLabelIndex("TEST", output_sequence),
+              testing::ElementsAre(1, 2, 9, 8));
+  ASSERT_THAT(mpms::GetClipLabelConfidence("TEST", output_sequence),
+              testing::ElementsAre(0.1, 0.2, 0.9, 0.8));
+  ASSERT_THAT(
+      mpms::GetClipLabelString("OTHER", output_sequence),
+      testing::ElementsAre("old_label_3", "old_label_4", "label_3", "label_4"));
+  ASSERT_THAT(mpms::GetClipLabelIndex("OTHER", output_sequence),
+              testing::ElementsAre(3, 4, 7, 6));
+  ASSERT_THAT(mpms::GetClipLabelConfidence("OTHER", output_sequence),
+              testing::ElementsAre(0.3, 0.4, 0.7, 0.6));
+}
+
+TEST_F(PackMediaSequenceCalculatorTest,
+       DifferentClipLabelScoreAndConfidenceSize) {
+  SetUpCalculator(
+      /*input_streams=*/{"CLIP_LABEL_TEST:test", "CLIP_LABEL_OTHER:test2"},
+      /*features=*/{}, /*output_only_if_all_present=*/false,
+      /*replace_instead_of_append=*/true);
+  auto input_sequence = ::absl::make_unique<tf::SequenceExample>();
+
+  Detection detection_1;
+  // 2 labels and 1 score.
+  detection_1.add_label("label_1");
+  detection_1.add_label("label_2");
+  detection_1.add_score(0.1);
+  runner_->MutableInputs()
+      ->Tag(kClipLabelTestTag)
+      .packets.push_back(MakePacket<Detection>(detection_1).At(Timestamp(1)));
+  Detection detection_2;
+  detection_2.add_label("label_3");
+  detection_2.add_label("label_4");
+  detection_2.add_score(0.3);
+  detection_2.add_score(0.4);
+  runner_->MutableInputs()
+      ->Tag(kClipLabelOtherTag)
+      .packets.push_back(MakePacket<Detection>(detection_2).At(Timestamp(2)));
+  runner_->MutableSidePackets()->Tag(kSequenceExampleTag) =
+      Adopt(input_sequence.release());
+
+  ASSERT_THAT(runner_->Run(),
+              testing::status::StatusIs(absl::StatusCode::kInvalidArgument));
+}
+
+TEST_F(PackMediaSequenceCalculatorTest, AddClipMediaId) {
+  SetUpCalculator(
+      /*input_streams=*/{"FLOAT_FEATURE_TEST:test",
+                         "FLOAT_FEATURE_OTHER:test2"},
+      /*features=*/{},
+      /*output_only_if_all_present=*/false,
+      /*replace_instead_of_append=*/true,
+      /*output_as_zero_timestamp=*/false, /*input_side_packets=*/
+      {"SEQUENCE_EXAMPLE:input_sequence", "CLIP_MEDIA_ID:video_id"});
+  auto input_sequence = absl::make_unique<tf::SequenceExample>();
+  const std::string test_video_id = "test_video_id";
+
+  int num_timesteps = 2;
+  for (int i = 0; i < num_timesteps; ++i) {
+    auto vf_ptr = ::absl::make_unique<std::vector<float>>(2, 2 << i);
+    runner_->MutableInputs()
+        ->Tag(kFloatFeatureTestTag)
+        .packets.push_back(Adopt(vf_ptr.release()).At(Timestamp(i)));
+    vf_ptr = ::absl::make_unique<std::vector<float>>(2, 2 << i);
+    runner_->MutableInputs()
+        ->Tag(kFloatFeatureOtherTag)
+        .packets.push_back(Adopt(vf_ptr.release()).At(Timestamp(i)));
+  }
+
+  runner_->MutableSidePackets()->Tag(kClipMediaIdTag) =
+      MakePacket<std::string>(test_video_id);
+  runner_->MutableSidePackets()->Tag(kSequenceExampleTag) =
+      Adopt(input_sequence.release());
+
+  MP_ASSERT_OK(runner_->Run());
+
+  const std::vector<Packet>& output_packets =
+      runner_->Outputs().Tag(kSequenceExampleTag).packets;
+  ASSERT_EQ(1, output_packets.size());
+  const tf::SequenceExample& output_sequence =
+      output_packets[0].Get<tf::SequenceExample>();
+
+  ASSERT_EQ(test_video_id, mpms::GetClipMediaId(output_sequence));
+}
+
+TEST_F(PackMediaSequenceCalculatorTest, ReplaceClipMediaId) {
+  SetUpCalculator(
+      /*input_streams=*/{"FLOAT_FEATURE_TEST:test",
+                         "FLOAT_FEATURE_OTHER:test2"},
+      /*features=*/{},
+      /*output_only_if_all_present=*/false,
+      /*replace_instead_of_append=*/true,
+      /*output_as_zero_timestamp=*/false, /*input_side_packets=*/
+      {"SEQUENCE_EXAMPLE:input_sequence", "CLIP_MEDIA_ID:video_id"});
+  auto input_sequence = absl::make_unique<tf::SequenceExample>();
+  const std::string existing_video_id = "existing_video_id";
+  mpms::SetClipMediaId(existing_video_id, input_sequence.get());
+  const std::string test_video_id = "test_video_id";
+
+  int num_timesteps = 2;
+  for (int i = 0; i < num_timesteps; ++i) {
+    auto vf_ptr = ::absl::make_unique<std::vector<float>>(2, 2 << i);
+    runner_->MutableInputs()
+        ->Tag(kFloatFeatureTestTag)
+        .packets.push_back(Adopt(vf_ptr.release()).At(Timestamp(i)));
+    vf_ptr = ::absl::make_unique<std::vector<float>>(2, 2 << i);
+    runner_->MutableInputs()
+        ->Tag(kFloatFeatureOtherTag)
+        .packets.push_back(Adopt(vf_ptr.release()).At(Timestamp(i)));
+  }
+
+  runner_->MutableSidePackets()->Tag(kClipMediaIdTag) =
+      MakePacket<std::string>(test_video_id).At(Timestamp(0));
+  runner_->MutableSidePackets()->Tag(kSequenceExampleTag) =
+      Adopt(input_sequence.release());
+
+  MP_ASSERT_OK(runner_->Run());
+
+  const std::vector<Packet>& output_packets =
+      runner_->Outputs().Tag(kSequenceExampleTag).packets;
+  ASSERT_EQ(1, output_packets.size());
+  const tf::SequenceExample& output_sequence =
+      output_packets[0].Get<tf::SequenceExample>();
+
+  ASSERT_EQ(test_video_id, mpms::GetClipMediaId(output_sequence));
+}
+
 TEST_F(PackMediaSequenceCalculatorTest, MissingStreamOK) {
   SetUpCalculator(
       {"FORWARD_FLOW_ENCODED:flow", "FLOAT_FEATURE_I3D_FLOW:feature"}, {},
@@ -767,7 +1535,8 @@ TEST_F(PackMediaSequenceCalculatorTest, MissingStreamOK) {
 
   cv::Mat image(2, 3, CV_8UC3, cv::Scalar(0, 0, 255));
   std::vector<uchar> bytes;
-  ASSERT_TRUE(cv::imencode(".jpg", image, bytes, {80}));
+  ASSERT_TRUE(
+      cv::imencode(".jpg", image, bytes, {cv::IMWRITE_HDR_COMPRESSION, 1}));
   std::string test_flow_string(bytes.begin(), bytes.end());
   OpenCvImageEncoderCalculatorResults encoded_flow;
   encoded_flow.set_encoded_image(test_flow_string);
@@ -813,7 +1582,8 @@ TEST_F(PackMediaSequenceCalculatorTest, MissingStreamNotOK) {
   mpms::SetClipMediaId(test_video_id, input_sequence.get());
   cv::Mat image(2, 3, CV_8UC3, cv::Scalar(0, 0, 255));
   std::vector<uchar> bytes;
-  ASSERT_TRUE(cv::imencode(".jpg", image, bytes, {80}));
+  ASSERT_TRUE(
+      cv::imencode(".jpg", image, bytes, {cv::IMWRITE_HDR_COMPRESSION, 1}));
   std::string test_flow_string(bytes.begin(), bytes.end());
   OpenCvImageEncoderCalculatorResults encoded_flow;
   encoded_flow.set_encoded_image(test_flow_string);
@@ -970,7 +1740,8 @@ TEST_F(PackMediaSequenceCalculatorTest, TestReconcilingAnnotations) {
   auto input_sequence = ::absl::make_unique<tf::SequenceExample>();
   cv::Mat image(2, 3, CV_8UC3, cv::Scalar(0, 0, 255));
   std::vector<uchar> bytes;
-  ASSERT_TRUE(cv::imencode(".jpg", image, bytes, {80}));
+  ASSERT_TRUE(
+      cv::imencode(".jpg", image, bytes, {cv::IMWRITE_HDR_COMPRESSION, 1}));
   OpenCvImageEncoderCalculatorResults encoded_image;
   encoded_image.set_encoded_image(bytes.data(), bytes.size());
   encoded_image.set_width(2);
@@ -1021,7 +1792,8 @@ TEST_F(PackMediaSequenceCalculatorTest, TestOverwritingAndReconciling) {
   auto input_sequence = ::absl::make_unique<tf::SequenceExample>();
   cv::Mat image(2, 3, CV_8UC3, cv::Scalar(0, 0, 255));
   std::vector<uchar> bytes;
-  ASSERT_TRUE(cv::imencode(".jpg", image, bytes, {80}));
+  ASSERT_TRUE(
+      cv::imencode(".jpg", image, bytes, {cv::IMWRITE_HDR_COMPRESSION, 1}));
   OpenCvImageEncoderCalculatorResults encoded_image;
   encoded_image.set_encoded_image(bytes.data(), bytes.size());
   int height = 2;
@@ -1057,6 +1829,7 @@ TEST_F(PackMediaSequenceCalculatorTest, TestOverwritingAndReconciling) {
     mpms::AddBBoxNumRegions(-1, input_sequence.get());
     mpms::AddBBoxLabelString({"anything"}, input_sequence.get());
     mpms::AddBBoxLabelIndex({-1}, input_sequence.get());
+    mpms::AddBBoxLabelConfidence({-1}, input_sequence.get());
     mpms::AddBBoxClassString({"anything"}, input_sequence.get());
     mpms::AddBBoxClassIndex({-1}, input_sequence.get());
     mpms::AddBBoxTrackString({"anything"}, input_sequence.get());

mediapipe/calculators/tensorflow/tensor_squeeze_dimensions_calculator.cc

@@ -12,6 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+#include "absl/log/absl_log.h"
 #include "mediapipe/calculators/tensorflow/tensor_squeeze_dimensions_calculator.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/port/ret_check.h"
@@ -99,10 +100,11 @@ class TensorSqueezeDimensionsCalculator : public CalculatorBase {
       }
     }
     if (remove_dims_.empty()) {
-      LOG(ERROR) << "TensorSqueezeDimensionsCalculator is squeezing input with "
+      ABSL_LOG(ERROR)
-                    "no single-dimensions. Calculator will be a no-op.";
+          << "TensorSqueezeDimensionsCalculator is squeezing input with "
-      LOG(ERROR) << "Input to TensorSqueezeDimensionsCalculator has shape "
+             "no single-dimensions. Calculator will be a no-op.";
-                 << tensor_shape.DebugString();
+      ABSL_LOG(ERROR) << "Input to TensorSqueezeDimensionsCalculator has shape "
+                      << tensor_shape.DebugString();
     }
   }
 };

mediapipe/calculators/tensorflow/tensor_to_image_frame_calculator.cc

@@ -14,6 +14,7 @@
 
 #include <iostream>
 
+#include "absl/log/absl_check.h"
 #include "mediapipe/calculators/tensorflow/tensor_to_image_frame_calculator.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/formats/image_frame.h"
@@ -99,7 +100,7 @@ absl::Status TensorToImageFrameCalculator::Process(CalculatorContext* cc) {
   const tf::Tensor& input_tensor = cc->Inputs().Tag(kTensor).Get<tf::Tensor>();
   int32_t depth = 1;
   if (input_tensor.dims() != 2) {  // Depth is 1 for 2D tensors.
-    CHECK(3 == input_tensor.dims())
+    ABSL_CHECK(3 == input_tensor.dims())
         << "Only 2 or 3-D Tensors can be converted to frames. Instead got: "
         << input_tensor.dims();
     depth = input_tensor.dim_size(2);

mediapipe/calculators/tensorflow/tensor_to_matrix_calculator.cc

@@ -15,6 +15,7 @@
 // Calculator converts from one-dimensional Tensor of DT_FLOAT to Matrix
 // OR from (batched) two-dimensional Tensor of DT_FLOAT to Matrix.
 
+#include "absl/log/absl_check.h"
 #include "mediapipe/calculators/tensorflow/tensor_to_matrix_calculator.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/formats/matrix.h"
@@ -36,7 +37,7 @@ constexpr char kReference[] = "REFERENCE";
 
 absl::Status FillTimeSeriesHeaderIfValid(const Packet& header_packet,
                                          TimeSeriesHeader* header) {
-  CHECK(header);
+  ABSL_CHECK(header);
   if (header_packet.IsEmpty()) {
     return absl::UnknownError("No header found.");
   }
@@ -191,7 +192,7 @@ absl::Status TensorToMatrixCalculator::Process(CalculatorContext* cc) {
       << "Tensor stream packet does not contain a Tensor.";
 
   const tf::Tensor& input_tensor = cc->Inputs().Tag(kTensor).Get<tf::Tensor>();
-  CHECK(1 == input_tensor.dims() || 2 == input_tensor.dims())
+  ABSL_CHECK(1 == input_tensor.dims() || 2 == input_tensor.dims())
       << "Only 1-D or 2-D Tensors can be converted to matrices.";
   const int32_t length = input_tensor.dim_size(input_tensor.dims() - 1);
   const int32_t width =

mediapipe/calculators/tensorflow/tensor_to_vector_float_calculator.cc

@@ -15,12 +15,16 @@
 // Calculator converts from one-dimensional Tensor of DT_FLOAT to vector<float>
 // OR from (batched) two-dimensional Tensor of DT_FLOAT to vector<vector<float>.
 
+#include <memory>
+#include <vector>
+
 #include "mediapipe/calculators/tensorflow/tensor_to_vector_float_calculator_options.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/port/ret_check.h"
 #include "mediapipe/framework/port/status.h"
 #include "tensorflow/core/framework/tensor.h"
 #include "tensorflow/core/framework/types.h"
+#include "tensorflow/core/platform/bfloat16.h"
 
 namespace mediapipe {
 
@@ -76,21 +80,31 @@ absl::Status TensorToVectorFloatCalculator::Open(CalculatorContext* cc) {
 absl::Status TensorToVectorFloatCalculator::Process(CalculatorContext* cc) {
   const tf::Tensor& input_tensor =
       cc->Inputs().Index(0).Value().Get<tf::Tensor>();
-  RET_CHECK(tf::DT_FLOAT == input_tensor.dtype())
+  RET_CHECK(tf::DT_FLOAT == input_tensor.dtype() ||
-      << "expected DT_FLOAT input but got "
+            tf::DT_BFLOAT16 == input_tensor.dtype())
+      << "expected DT_FLOAT or DT_BFLOAT_16 input but got "
       << tensorflow::DataTypeString(input_tensor.dtype());
 
   if (options_.tensor_is_2d()) {
     RET_CHECK(2 == input_tensor.dims())
         << "Expected 2-dimensional Tensor, but the tensor shape is: "
         << input_tensor.shape().DebugString();
-    auto output = absl::make_unique<std::vector<std::vector<float>>>(
+    auto output = std::make_unique<std::vector<std::vector<float>>>(
         input_tensor.dim_size(0), std::vector<float>(input_tensor.dim_size(1)));
     for (int i = 0; i < input_tensor.dim_size(0); ++i) {
       auto& instance_output = output->at(i);
-      const auto& slice = input_tensor.Slice(i, i + 1).unaligned_flat<float>();
+      if (tf::DT_BFLOAT16 == input_tensor.dtype()) {
-      for (int j = 0; j < input_tensor.dim_size(1); ++j) {
+        const auto& slice =
-        instance_output.at(j) = slice(j);
+            input_tensor.Slice(i, i + 1).unaligned_flat<tf::bfloat16>();
+        for (int j = 0; j < input_tensor.dim_size(1); ++j) {
+          instance_output.at(j) = static_cast<float>(slice(j));
+        }
+      } else {
+        const auto& slice =
+            input_tensor.Slice(i, i + 1).unaligned_flat<float>();
+        for (int j = 0; j < input_tensor.dim_size(1); ++j) {
+          instance_output.at(j) = slice(j);
+        }
       }
     }
     cc->Outputs().Index(0).Add(output.release(), cc->InputTimestamp());
@@ -101,10 +115,17 @@ absl::Status TensorToVectorFloatCalculator::Process(CalculatorContext* cc) {
           << "tensor shape is: " << input_tensor.shape().DebugString();
     }
     auto output =
-        absl::make_unique<std::vector<float>>(input_tensor.NumElements());
+        std::make_unique<std::vector<float>>(input_tensor.NumElements());
-    const auto& tensor_values = input_tensor.unaligned_flat<float>();
+    if (tf::DT_BFLOAT16 == input_tensor.dtype()) {
-    for (int i = 0; i < input_tensor.NumElements(); ++i) {
+      const auto& tensor_values = input_tensor.unaligned_flat<tf::bfloat16>();
-      output->at(i) = tensor_values(i);
+      for (int i = 0; i < input_tensor.NumElements(); ++i) {
+        output->at(i) = static_cast<float>(tensor_values(i));
+      }
+    } else {
+      const auto& tensor_values = input_tensor.unaligned_flat<float>();
+      for (int i = 0; i < input_tensor.NumElements(); ++i) {
+        output->at(i) = tensor_values(i);
+      }
     }
     cc->Outputs().Index(0).Add(output.release(), cc->InputTimestamp());
   }

mediapipe/calculators/tensorflow/tensor_to_vector_float_calculator_test.cc

@@ -12,6 +12,8 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+#include <memory>
+
 #include "mediapipe/calculators/tensorflow/tensor_to_vector_float_calculator_options.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/calculator_runner.h"
@@ -19,6 +21,7 @@
 #include "mediapipe/util/packet_test_util.h"
 #include "tensorflow/core/framework/tensor.h"
 #include "tensorflow/core/framework/types.pb.h"
+#include "tensorflow/core/platform/bfloat16.h"
 
 namespace mediapipe {
 
@@ -72,6 +75,62 @@ TEST_F(TensorToVectorFloatCalculatorTest, ConvertsToVectorFloat) {
   }
 }
 
+TEST_F(TensorToVectorFloatCalculatorTest, CheckBFloat16Type) {
+  SetUpRunner(false, false);
+  const tf::TensorShape tensor_shape(std::vector<tf::int64>{5});
+  auto tensor = std::make_unique<tf::Tensor>(tf::DT_BFLOAT16, tensor_shape);
+  auto tensor_vec = tensor->vec<tf::bfloat16>();
+  for (int i = 0; i < 5; ++i) {
+    tensor_vec(i) = static_cast<tf::bfloat16>(1 << i);
+  }
+
+  const int64_t time = 1234;
+  runner_->MutableInputs()->Index(0).packets.push_back(
+      Adopt(tensor.release()).At(Timestamp(time)));
+
+  EXPECT_TRUE(runner_->Run().ok());
+  const std::vector<Packet>& output_packets =
+      runner_->Outputs().Index(0).packets;
+  EXPECT_EQ(1, output_packets.size());
+  EXPECT_EQ(time, output_packets[0].Timestamp().Value());
+  const std::vector<float>& output_vector =
+      output_packets[0].Get<std::vector<float>>();
+
+  EXPECT_EQ(5, output_vector.size());
+  for (int i = 0; i < 5; ++i) {
+    const float expected = static_cast<float>(1 << i);
+    EXPECT_EQ(expected, output_vector[i]);
+  }
+}
+
+TEST_F(TensorToVectorFloatCalculatorTest, CheckBFloat16TypeAllDim) {
+  SetUpRunner(false, true);
+  const tf::TensorShape tensor_shape(std::vector<tf::int64>{2, 2, 2});
+  auto tensor = std::make_unique<tf::Tensor>(tf::DT_BFLOAT16, tensor_shape);
+  auto slice = tensor->flat<tf::bfloat16>();
+  for (int i = 0; i < 2 * 2 * 2; ++i) {
+    // 2^i can be represented exactly in floating point numbers if 'i' is small.
+    slice(i) = static_cast<tf::bfloat16>(1 << i);
+  }
+
+  const int64_t time = 1234;
+  runner_->MutableInputs()->Index(0).packets.push_back(
+      Adopt(tensor.release()).At(Timestamp(time)));
+
+  EXPECT_TRUE(runner_->Run().ok());
+  const std::vector<Packet>& output_packets =
+      runner_->Outputs().Index(0).packets;
+  EXPECT_EQ(1, output_packets.size());
+  EXPECT_EQ(time, output_packets[0].Timestamp().Value());
+  const std::vector<float>& output_vector =
+      output_packets[0].Get<std::vector<float>>();
+  EXPECT_EQ(2 * 2 * 2, output_vector.size());
+  for (int i = 0; i < 2 * 2 * 2; ++i) {
+    const float expected = static_cast<float>(1 << i);
+    EXPECT_EQ(expected, output_vector[i]);
+  }
+}
+
 TEST_F(TensorToVectorFloatCalculatorTest, ConvertsBatchedToVectorVectorFloat) {
   SetUpRunner(true, false);
   const tf::TensorShape tensor_shape(std::vector<tf::int64>{1, 5});

mediapipe/calculators/tensorflow/tensorflow_inference_calculator.cc

@@ -20,6 +20,7 @@
 #include <vector>
 
 #include "absl/base/thread_annotations.h"
+#include "absl/log/absl_check.h"
 #include "absl/memory/memory.h"
 #include "absl/strings/str_split.h"
 #include "absl/synchronization/mutex.h"
@@ -515,7 +516,7 @@ class TensorFlowInferenceCalculator : public CalculatorBase {
         tf::Tensor concated;
         const tf::Status concat_status =
             tf::tensor::Concat(keyed_tensors.second, &concated);
-        CHECK(concat_status.ok()) << concat_status.ToString();
+        ABSL_CHECK(concat_status.ok()) << concat_status.ToString();
         input_tensors.emplace_back(tag_to_tensor_map_[keyed_tensors.first],
                                    concated);
       }
@@ -597,7 +598,7 @@ class TensorFlowInferenceCalculator : public CalculatorBase {
         std::vector<tf::Tensor> split_tensors;
         const tf::Status split_status =
             tf::tensor::Split(outputs[i], split_vector, &split_tensors);
-        CHECK(split_status.ok()) << split_status.ToString();
+        ABSL_CHECK(split_status.ok()) << split_status.ToString();
         // Loop over timestamps so that we don't copy the padding.
         for (int j = 0; j < inference_state->batch_timestamps_.size(); ++j) {
           tf::Tensor output_tensor(split_tensors[j]);

mediapipe/calculators/tensorflow/tensorflow_inference_calculator_test.cc

@@ -17,6 +17,8 @@
 #include <vector>
 
 #include "absl/flags/flag.h"
+#include "absl/log/absl_check.h"
+#include "absl/log/absl_log.h"
 #include "mediapipe/calculators/tensorflow/tensorflow_inference_calculator.pb.h"
 #include "mediapipe/calculators/tensorflow/tensorflow_session_from_frozen_graph_generator.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
@@ -118,7 +120,7 @@ class TensorflowInferenceCalculatorTest : public ::testing::Test {
   // Create tensor from Vector and add as a Packet to the provided tag as input.
   void AddVectorToInputsAsPacket(const std::vector<Packet>& packets,
                                  const std::string& tag) {
-    CHECK(!packets.empty())
+    ABSL_CHECK(!packets.empty())
         << "Please specify at least some data in the packet";
     auto packets_ptr = absl::make_unique<std::vector<Packet>>(packets);
     runner_->MutableInputs()->Tag(tag).packets.push_back(
@@ -586,12 +588,12 @@ TEST_F(TensorflowInferenceCalculatorTest, TestRecurrentStates) {
       runner_->Outputs().Tag(kMultipliedTag).packets;
   ASSERT_EQ(2, output_packets_mult.size());
   const tf::Tensor& tensor_mult = output_packets_mult[0].Get<tf::Tensor>();
-  LOG(INFO) << "timestamp: " << 0;
+  ABSL_LOG(INFO) << "timestamp: " << 0;
   auto expected_tensor = tf::test::AsTensor<int32_t>({3, 8, 15});
   tf::test::ExpectTensorEqual<int32_t>(tensor_mult, expected_tensor);
   const tf::Tensor& tensor_mult1 = output_packets_mult[1].Get<tf::Tensor>();
   auto expected_tensor1 = tf::test::AsTensor<int32_t>({9, 32, 75});
-  LOG(INFO) << "timestamp: " << 1;
+  ABSL_LOG(INFO) << "timestamp: " << 1;
   tf::test::ExpectTensorEqual<int32_t>(tensor_mult1, expected_tensor1);
 
   EXPECT_EQ(2, runner_
@@ -627,12 +629,12 @@ TEST_F(TensorflowInferenceCalculatorTest, TestRecurrentStateOverride) {
       runner_->Outputs().Tag(kMultipliedTag).packets;
   ASSERT_EQ(2, output_packets_mult.size());
   const tf::Tensor& tensor_mult = output_packets_mult[0].Get<tf::Tensor>();
-  LOG(INFO) << "timestamp: " << 0;
+  ABSL_LOG(INFO) << "timestamp: " << 0;
   auto expected_tensor = tf::test::AsTensor<int32_t>({3, 4, 5});
   tf::test::ExpectTensorEqual<int32_t>(tensor_mult, expected_tensor);
   const tf::Tensor& tensor_mult1 = output_packets_mult[1].Get<tf::Tensor>();
   auto expected_tensor1 = tf::test::AsTensor<int32_t>({3, 4, 5});
-  LOG(INFO) << "timestamp: " << 1;
+  ABSL_LOG(INFO) << "timestamp: " << 1;
   tf::test::ExpectTensorEqual<int32_t>(tensor_mult1, expected_tensor1);
 
   EXPECT_EQ(2, runner_

mediapipe/calculators/tensorflow/tensorflow_session_from_frozen_graph_calculator.cc

@@ -23,12 +23,12 @@
 
 #include <string>
 
+#include "absl/log/absl_log.h"
 #include "mediapipe/calculators/tensorflow/tensorflow_session.h"
 #include "mediapipe/calculators/tensorflow/tensorflow_session_from_frozen_graph_calculator.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/deps/clock.h"
 #include "mediapipe/framework/deps/monotonic_clock.h"
-#include "mediapipe/framework/port/logging.h"
 #include "mediapipe/framework/port/ret_check.h"
 #include "mediapipe/framework/port/status.h"
 #include "mediapipe/framework/tool/status_util.h"
@@ -156,8 +156,8 @@ class TensorFlowSessionFromFrozenGraphCalculator : public CalculatorBase {
 
     cc->OutputSidePackets().Tag(kSessionTag).Set(Adopt(session.release()));
     const uint64_t end_time = absl::ToUnixMicros(clock->TimeNow());
-    LOG(INFO) << "Loaded frozen model in: " << end_time - start_time
+    ABSL_LOG(INFO) << "Loaded frozen model in: " << end_time - start_time
-              << " microseconds.";
+                   << " microseconds.";
     return absl::OkStatus();
   }
 

mediapipe/calculators/tensorflow/tensorflow_session_from_frozen_graph_generator.cc

@@ -24,13 +24,13 @@
 
 #include <string>
 
+#include "absl/log/absl_log.h"
 #include "mediapipe/calculators/tensorflow/tensorflow_session.h"
 #include "mediapipe/calculators/tensorflow/tensorflow_session_from_frozen_graph_generator.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/deps/clock.h"
 #include "mediapipe/framework/deps/monotonic_clock.h"
 #include "mediapipe/framework/port/file_helpers.h"
-#include "mediapipe/framework/port/logging.h"
 #include "mediapipe/framework/port/ret_check.h"
 #include "mediapipe/framework/port/status.h"
 #include "mediapipe/framework/tool/status_util.h"
@@ -155,8 +155,8 @@ class TensorFlowSessionFromFrozenGraphGenerator : public PacketGenerator {
 
     output_side_packets->Tag(kSessionTag) = Adopt(session.release());
     const uint64_t end_time = absl::ToUnixMicros(clock->TimeNow());
-    LOG(INFO) << "Loaded frozen model in: " << end_time - start_time
+    ABSL_LOG(INFO) << "Loaded frozen model in: " << end_time - start_time
-              << " microseconds.";
+                   << " microseconds.";
     return absl::OkStatus();
   }
 };

mediapipe/calculators/tensorflow/tensorflow_session_from_saved_model_calculator.cc

@@ -17,6 +17,7 @@
 #if !defined(__ANDROID__)
 #include "mediapipe/framework/port/file_helpers.h"
 #endif
+#include "absl/log/absl_log.h"
 #include "absl/strings/str_replace.h"
 #include "mediapipe/calculators/tensorflow/tensorflow_session.h"
 #include "mediapipe/calculators/tensorflow/tensorflow_session_from_saved_model_calculator.pb.h"
@@ -69,7 +70,7 @@ const std::string MaybeConvertSignatureToTag(
                    [](unsigned char c) { return std::toupper(c); });
     output = absl::StrReplaceAll(
         output, {{"/", "_"}, {"-", "_"}, {".", "_"}, {":", "_"}});
-    LOG(INFO) << "Renamed TAG from: " << name << " to " << output;
+    ABSL_LOG(INFO) << "Renamed TAG from: " << name << " to " << output;
     return output;
   } else {
     return name;

mediapipe/calculators/tensorflow/tensorflow_session_from_saved_model_generator.cc

@@ -19,6 +19,7 @@
 #if !defined(__ANDROID__)
 #include "mediapipe/framework/port/file_helpers.h"
 #endif
+#include "absl/log/absl_log.h"
 #include "absl/strings/str_replace.h"
 #include "mediapipe/calculators/tensorflow/tensorflow_session.h"
 #include "mediapipe/calculators/tensorflow/tensorflow_session_from_saved_model_generator.pb.h"
@@ -75,7 +76,7 @@ const std::string MaybeConvertSignatureToTag(
                    [](unsigned char c) { return std::toupper(c); });
     output = absl::StrReplaceAll(
         output, {{"/", "_"}, {"-", "_"}, {".", "_"}, {":", "_"}});
-    LOG(INFO) << "Renamed TAG from: " << name << " to " << output;
+    ABSL_LOG(INFO) << "Renamed TAG from: " << name << " to " << output;
     return output;
   } else {
     return name;

mediapipe/calculators/tensorflow/unpack_media_sequence_calculator.cc

@@ -13,6 +13,7 @@
 // limitations under the License.
 
 #include "absl/container/flat_hash_map.h"
+#include "absl/log/absl_log.h"
 #include "absl/strings/match.h"
 #include "mediapipe/calculators/core/packet_resampler_calculator.pb.h"
 #include "mediapipe/calculators/tensorflow/unpack_media_sequence_calculator.pb.h"
@@ -201,8 +202,8 @@ class UnpackMediaSequenceCalculator : public CalculatorBase {
     first_timestamp_seen_ = Timestamp::OneOverPostStream().Value();
     for (const auto& map_kv : sequence_->feature_lists().feature_list()) {
       if (absl::StrContains(map_kv.first, "/timestamp")) {
-        LOG(INFO) << "Found feature timestamps: " << map_kv.first
+        ABSL_LOG(INFO) << "Found feature timestamps: " << map_kv.first
-                  << " with size: " << map_kv.second.feature_size();
+                       << " with size: " << map_kv.second.feature_size();
         int64_t recent_timestamp = Timestamp::PreStream().Value();
         for (int i = 0; i < map_kv.second.feature_size(); ++i) {
           int64_t next_timestamp =
@@ -309,8 +310,8 @@ class UnpackMediaSequenceCalculator : public CalculatorBase {
         audio_decoder_options->set_end_time(
             end_time + options.extra_padding_from_media_decoder());
       }
-      LOG(INFO) << "Created AudioDecoderOptions:\n"
+      ABSL_LOG(INFO) << "Created AudioDecoderOptions:\n"
-                << audio_decoder_options->DebugString();
+                     << audio_decoder_options->DebugString();
       cc->OutputSidePackets()
           .Tag(kAudioDecoderOptions)
           .Set(Adopt(audio_decoder_options.release()));
@@ -331,8 +332,8 @@ class UnpackMediaSequenceCalculator : public CalculatorBase {
             ->set_end_time(Timestamp::FromSeconds(end_time).Value());
       }
 
-      LOG(INFO) << "Created PacketResamplerOptions:\n"
+      ABSL_LOG(INFO) << "Created PacketResamplerOptions:\n"
-                << resampler_options->DebugString();
+                     << resampler_options->DebugString();
       cc->OutputSidePackets()
           .Tag(kPacketResamplerOptions)
           .Set(Adopt(resampler_options.release()));
@@ -351,7 +352,8 @@ class UnpackMediaSequenceCalculator : public CalculatorBase {
   absl::Status Process(CalculatorContext* cc) override {
     if (timestamps_.empty()) {
       // This occurs when we only have metadata to unpack.
-      LOG(INFO) << "only unpacking metadata because there are no timestamps.";
+      ABSL_LOG(INFO)
+          << "only unpacking metadata because there are no timestamps.";
       return tool::StatusStop();
     }
     // In Process(), we loop through timestamps on a reference stream and emit

mediapipe/calculators/tensorflow/unpack_media_sequence_calculator_test.cc

@@ -12,6 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+#include "absl/log/absl_log.h"
 #include "absl/memory/memory.h"
 #include "absl/strings/numbers.h"
 #include "mediapipe/calculators/core/packet_resampler_calculator.pb.h"
@@ -81,7 +82,7 @@ class UnpackMediaSequenceCalculatorTest : public ::testing::Test {
     if (options != nullptr) {
       *config.mutable_options() = *options;
     }
-    LOG(INFO) << config.DebugString();
+    ABSL_LOG(INFO) << config.DebugString();
     runner_ = absl::make_unique<CalculatorRunner>(config);
   }
 

mediapipe/calculators/tensorflow/unpack_yt8m_sequence_example_calculator.cc

@@ -14,6 +14,8 @@
 
 #include <iterator>
 
+#include "absl/log/absl_check.h"
+#include "absl/log/absl_log.h"
 #include "mediapipe/calculators/tensorflow/lapped_tensor_buffer_calculator.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/packet.h"
@@ -46,7 +48,7 @@ std::string GetQuantizedFeature(
                                .Get(index)
                                .bytes_list()
                                .value();
-  CHECK_EQ(1, bytes_list.size());
+  ABSL_CHECK_EQ(1, bytes_list.size());
   return bytes_list.Get(0);
 }
 }  // namespace
@@ -149,8 +151,9 @@ class UnpackYt8mSequenceExampleCalculator : public CalculatorBase {
             .Set(MakePacket<int>(segment_size));
       }
     }
-    LOG(INFO) << "Reading the sequence example that contains yt8m id: "
+    ABSL_LOG(INFO) << "Reading the sequence example that contains yt8m id: "
-              << yt8m_id << ". Feature list length: " << feature_list_length_;
+                   << yt8m_id
+                   << ". Feature list length: " << feature_list_length_;
     return absl::OkStatus();
   }
 

mediapipe/calculators/tensorflow/vector_float_to_tensor_calculator.cc

@@ -14,6 +14,7 @@
 //
 // Converts vector<float> (or vector<vector<float>>) to 1D (or 2D) tf::Tensor.
 
+#include "absl/log/absl_log.h"
 #include "mediapipe/calculators/tensorflow/vector_float_to_tensor_calculator_options.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/port/ret_check.h"
@@ -68,7 +69,7 @@ absl::Status VectorFloatToTensorCalculator::GetContract(
         // Output vector<float>.
     );
   } else {
-    LOG(FATAL) << "input size not supported";
+    ABSL_LOG(FATAL) << "input size not supported";
   }
   RET_CHECK_EQ(cc->Outputs().NumEntries(), 1)
       << "Only one output stream is supported.";
@@ -125,7 +126,7 @@ absl::Status VectorFloatToTensorCalculator::Process(CalculatorContext* cc) {
     }
     cc->Outputs().Index(0).Add(output.release(), cc->InputTimestamp());
   } else {
-    LOG(FATAL) << "input size not supported";
+    ABSL_LOG(FATAL) << "input size not supported";
   }
   return absl::OkStatus();
 }

mediapipe/calculators/tensorflow/vector_int_to_tensor_calculator.cc

@@ -15,6 +15,8 @@
 // Converts a single int or vector<int> or vector<vector<int>> to 1D (or 2D)
 // tf::Tensor.
 
+#include "absl/log/absl_check.h"
+#include "absl/log/absl_log.h"
 #include "mediapipe/calculators/tensorflow/vector_int_to_tensor_calculator_options.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/port/ret_check.h"
@@ -86,7 +88,7 @@ absl::Status VectorIntToTensorCalculator::GetContract(CalculatorContract* cc) {
       cc->Inputs().Tag(kVectorInt).Set<std::vector<int>>();
     }
   } else {
-    LOG(FATAL) << "input size not supported";
+    ABSL_LOG(FATAL) << "input size not supported";
   }
   RET_CHECK_EQ(cc->Outputs().NumEntries(), 1)
       << "Only one output stream is supported.";
@@ -113,11 +115,11 @@ absl::Status VectorIntToTensorCalculator::Process(CalculatorContext* cc) {
             .Get<std::vector<std::vector<int>>>();
 
     const int32_t rows = input.size();
-    CHECK_GE(rows, 1);
+    ABSL_CHECK_GE(rows, 1);
     const int32_t cols = input[0].size();
-    CHECK_GE(cols, 1);
+    ABSL_CHECK_GE(cols, 1);
     for (int i = 1; i < rows; ++i) {
-      CHECK_EQ(input[i].size(), cols);
+      ABSL_CHECK_EQ(input[i].size(), cols);
     }
     if (options_.transpose()) {
       tensor_shape = tf::TensorShape({cols, rows});
@@ -140,7 +142,7 @@ absl::Status VectorIntToTensorCalculator::Process(CalculatorContext* cc) {
               AssignMatrixValue<int>(c, r, input[r][c], output.get());
               break;
             default:
-              LOG(FATAL) << "tensor data type is not supported.";
+              ABSL_LOG(FATAL) << "tensor data type is not supported.";
           }
         }
       }
@@ -158,7 +160,7 @@ absl::Status VectorIntToTensorCalculator::Process(CalculatorContext* cc) {
               AssignMatrixValue<int>(r, c, input[r][c], output.get());
               break;
             default:
-              LOG(FATAL) << "tensor data type is not supported.";
+              ABSL_LOG(FATAL) << "tensor data type is not supported.";
           }
         }
       }
@@ -171,7 +173,7 @@ absl::Status VectorIntToTensorCalculator::Process(CalculatorContext* cc) {
     } else {
       input = cc->Inputs().Tag(kVectorInt).Value().Get<std::vector<int>>();
     }
-    CHECK_GE(input.size(), 1);
+    ABSL_CHECK_GE(input.size(), 1);
     const int32_t length = input.size();
     tensor_shape = tf::TensorShape({length});
     auto output = ::absl::make_unique<tf::Tensor>(options_.tensor_data_type(),
@@ -188,12 +190,12 @@ absl::Status VectorIntToTensorCalculator::Process(CalculatorContext* cc) {
           output->tensor<int, 1>()(i) = input.at(i);
           break;
         default:
-          LOG(FATAL) << "tensor data type is not supported.";
+          ABSL_LOG(FATAL) << "tensor data type is not supported.";
       }
     }
     cc->Outputs().Tag(kTensorOut).Add(output.release(), cc->InputTimestamp());
   } else {
-    LOG(FATAL) << "input size not supported";
+    ABSL_LOG(FATAL) << "input size not supported";
   }
   return absl::OkStatus();
 }

mediapipe/calculators/tensorflow/vector_string_to_tensor_calculator.cc

@@ -15,6 +15,7 @@
 // Converts vector<std::string> (or vector<vector<std::string>>) to 1D (or 2D)
 // tf::Tensor.
 
+#include "absl/log/absl_log.h"
 #include "mediapipe/calculators/tensorflow/vector_string_to_tensor_calculator_options.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/port/ret_check.h"
@@ -69,7 +70,7 @@ absl::Status VectorStringToTensorCalculator::GetContract(
         // Input vector<std::string>.
     );
   } else {
-    LOG(FATAL) << "input size not supported";
+    ABSL_LOG(FATAL) << "input size not supported";
   }
   RET_CHECK_EQ(cc->Outputs().NumEntries(), 1)
       << "Only one output stream is supported.";
@@ -129,7 +130,7 @@ absl::Status VectorStringToTensorCalculator::Process(CalculatorContext* cc) {
     }
     cc->Outputs().Index(0).Add(output.release(), cc->InputTimestamp());
   } else {
-    LOG(FATAL) << "input size not supported";
+    ABSL_LOG(FATAL) << "input size not supported";
   }
   return absl::OkStatus();
 }

mediapipe/calculators/tflite/BUILD

@@ -103,6 +103,8 @@ cc_library(
         "//mediapipe/framework/formats/object_detection:anchor_cc_proto",
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/framework/port:status",
+        "@com_google_absl//absl/log:absl_check",
+        "@com_google_absl//absl/log:absl_log",
     ],
     alwayslink = 1,
 )
@@ -196,10 +198,13 @@ cc_library(
     deps = [
         ":tflite_inference_calculator_cc_proto",
         "//mediapipe/framework:calculator_framework",
+        "//mediapipe/framework/port:logging",
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/framework/stream_handler:fixed_size_input_stream_handler",
         "//mediapipe/util/tflite:config",
         "//mediapipe/util/tflite:tflite_model_loader",
+        "@com_google_absl//absl/log:absl_check",
+        "@com_google_absl//absl/log:absl_log",
         "@com_google_absl//absl/memory",
         "@org_tensorflow//tensorflow/lite:framework",
         "@org_tensorflow//tensorflow/lite/delegates/xnnpack:xnnpack_delegate",
@@ -275,6 +280,7 @@ cc_library(
         "//mediapipe/framework/stream_handler:fixed_size_input_stream_handler",
         "//mediapipe/util:resource_util",
         "//mediapipe/util/tflite:config",
+        "@com_google_absl//absl/log:absl_check",
         "@org_tensorflow//tensorflow/lite:framework",
         "@org_tensorflow//tensorflow/lite/kernels:builtin_ops",
     ] + selects.with_or({
@@ -392,6 +398,8 @@ cc_library(
         "//mediapipe/framework/formats/object_detection:anchor_cc_proto",
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/util/tflite:config",
+        "@com_google_absl//absl/log:absl_check",
+        "@com_google_absl//absl/log:absl_log",
         "@com_google_absl//absl/strings:str_format",
         "@com_google_absl//absl/types:span",
         "@org_tensorflow//tensorflow/lite:framework",
@@ -428,6 +436,7 @@ cc_library(
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/util:resource_util",
         "@com_google_absl//absl/container:node_hash_map",
+        "@com_google_absl//absl/log:absl_check",
         "@com_google_absl//absl/strings:str_format",
         "@com_google_absl//absl/types:span",
         "@org_tensorflow//tensorflow/lite:framework",
@@ -456,6 +465,7 @@ cc_library(
         "//mediapipe/framework:calculator_framework",
         "//mediapipe/framework/formats:landmark_cc_proto",
         "//mediapipe/framework/port:ret_check",
+        "@com_google_absl//absl/log:absl_check",
         "@org_tensorflow//tensorflow/lite:framework",
     ],
     alwayslink = 1,

mediapipe/calculators/tflite/ssd_anchors_calculator.cc

@@ -16,6 +16,8 @@
 #include <utility>
 #include <vector>
 
+#include "absl/log/absl_check.h"
+#include "absl/log/absl_log.h"
 #include "mediapipe/calculators/tflite/ssd_anchors_calculator.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/formats/object_detection/anchor.pb.h"
@@ -272,13 +274,13 @@ absl::Status SsdAnchorsCalculator::GenerateAnchors(
 
   if (options.feature_map_height_size()) {
     if (options.strides_size()) {
-      LOG(ERROR) << "Found feature map shapes. Strides will be ignored.";
+      ABSL_LOG(ERROR) << "Found feature map shapes. Strides will be ignored.";
     }
-    CHECK_EQ(options.feature_map_height_size(), kNumLayers);
+    ABSL_CHECK_EQ(options.feature_map_height_size(), kNumLayers);
-    CHECK_EQ(options.feature_map_height_size(),
+    ABSL_CHECK_EQ(options.feature_map_height_size(),
-             options.feature_map_width_size());
+                  options.feature_map_width_size());
   } else {
-    CHECK_EQ(options.strides_size(), kNumLayers);
+    ABSL_CHECK_EQ(options.strides_size(), kNumLayers);
   }
 
   if (options.multiscale_anchor_generation()) {

mediapipe/calculators/tflite/tflite_converter_calculator.cc

@@ -15,6 +15,7 @@
 #include <string>
 #include <vector>
 
+#include "absl/log/absl_check.h"
 #include "mediapipe/calculators/tflite/tflite_converter_calculator.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/formats/image_frame.h"
@@ -643,7 +644,7 @@ absl::Status TfLiteConverterCalculator::LoadOptions(CalculatorContext* cc) {
   if (options.has_output_tensor_float_range()) {
     output_range_.emplace(options.output_tensor_float_range().min(),
                           options.output_tensor_float_range().max());
-    CHECK_GT(output_range_->second, output_range_->first);
+    ABSL_CHECK_GT(output_range_->second, output_range_->first);
   }
 
   // Custom div and sub values.
@@ -661,9 +662,9 @@ absl::Status TfLiteConverterCalculator::LoadOptions(CalculatorContext* cc) {
 
   // Get desired way to handle input channels.
   max_num_channels_ = options.max_num_channels();
-  CHECK_GE(max_num_channels_, 1);
+  ABSL_CHECK_GE(max_num_channels_, 1);
-  CHECK_LE(max_num_channels_, 4);
+  ABSL_CHECK_LE(max_num_channels_, 4);
-  CHECK_NE(max_num_channels_, 2);
+  ABSL_CHECK_NE(max_num_channels_, 2);
 #if defined(MEDIAPIPE_IOS)
   if (cc->Inputs().HasTag(kGpuBufferTag))
     // Currently on iOS, tflite gpu input tensor must be 4 channels,

mediapipe/calculators/tflite/tflite_inference_calculator.cc

@@ -17,9 +17,12 @@
 #include <string>
 #include <vector>
 
+#include "absl/log/absl_check.h"
+#include "absl/log/absl_log.h"
 #include "absl/memory/memory.h"
 #include "mediapipe/calculators/tflite/tflite_inference_calculator.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
+#include "mediapipe/framework/port/logging.h"
 #include "mediapipe/framework/port/ret_check.h"
 #include "mediapipe/util/tflite/config.h"
 
@@ -109,8 +112,8 @@ std::unique_ptr<tflite::Interpreter> BuildEdgeTpuInterpreter(
     edgetpu::EdgeTpuContext* edgetpu_context) {
   resolver->AddCustom(edgetpu::kCustomOp, edgetpu::RegisterCustomOp());
   std::unique_ptr<tflite::Interpreter> interpreter;
-  CHECK_EQ(tflite::InterpreterBuilder(model, *resolver)(&interpreter),
+  ABSL_CHECK_EQ(tflite::InterpreterBuilder(model, *resolver)(&interpreter),
-           kTfLiteOk);
+                kTfLiteOk);
   interpreter->SetExternalContext(kTfLiteEdgeTpuContext, edgetpu_context);
   return interpreter;
 }
@@ -406,11 +409,12 @@ absl::Status TfLiteInferenceCalculator::Open(CalculatorContext* cc) {
   }
 
   if (use_advanced_gpu_api_ && !gpu_input_) {
-    LOG(WARNING) << "Cannot use advanced GPU APIs, input must be GPU buffers."
+    ABSL_LOG(WARNING)
-                    "Falling back to the default TFLite API.";
+        << "Cannot use advanced GPU APIs, input must be GPU buffers."
+           "Falling back to the default TFLite API.";
     use_advanced_gpu_api_ = false;
   }
-  CHECK(!use_advanced_gpu_api_ || gpu_inference_);
+  ABSL_CHECK(!use_advanced_gpu_api_ || gpu_inference_);
 
   MP_RETURN_IF_ERROR(LoadModel(cc));
 
@@ -802,9 +806,10 @@ absl::Status TfLiteInferenceCalculator::InitTFLiteGPURunner(
       const int tensor_idx = interpreter_->inputs()[i];
       interpreter_->SetTensorParametersReadWrite(tensor_idx, kTfLiteFloat32, "",
                                                  shape, quant);
-      CHECK(interpreter_->ResizeInputTensor(tensor_idx, shape) == kTfLiteOk);
+      ABSL_CHECK(interpreter_->ResizeInputTensor(tensor_idx, shape) ==
+                 kTfLiteOk);
     }
-    CHECK(interpreter_->AllocateTensors() == kTfLiteOk);
+    ABSL_CHECK(interpreter_->AllocateTensors() == kTfLiteOk);
   }
 
   // Create and bind OpenGL buffers for outputs.
@@ -1053,7 +1058,7 @@ absl::Status TfLiteInferenceCalculator::LoadDelegate(CalculatorContext* cc) {
           gpu_data_in_[i]->shape.w * gpu_data_in_[i]->shape.c;
       // Input to model can be RGBA only.
       if (tensor->dims->data[3] != 4) {
-        LOG(WARNING) << "Please ensure input GPU tensor is 4 channels.";
+        ABSL_LOG(WARNING) << "Please ensure input GPU tensor is 4 channels.";
       }
       const std::string shader_source =
           absl::Substitute(R"(#include <metal_stdlib>

mediapipe/calculators/tflite/tflite_tensors_to_classification_calculator.cc

@@ -17,6 +17,7 @@
 #include <vector>
 
 #include "absl/container/node_hash_map.h"
+#include "absl/log/absl_check.h"
 #include "absl/strings/str_format.h"
 #include "absl/types/span.h"
 #include "mediapipe/calculators/tflite/tflite_tensors_to_classification_calculator.pb.h"
@@ -172,7 +173,7 @@ absl::Status TfLiteTensorsToClassificationCalculator::Process(
 
   // Note that partial_sort will raise error when top_k_ >
   // classification_list->classification_size().
-  CHECK_GE(classification_list->classification_size(), top_k_);
+  ABSL_CHECK_GE(classification_list->classification_size(), top_k_);
   auto raw_classification_list = classification_list->mutable_classification();
   if (top_k_ > 0 && classification_list->classification_size() >= top_k_) {
     std::partial_sort(raw_classification_list->begin(),

mediapipe/calculators/tflite/tflite_tensors_to_detections_calculator.cc

@@ -15,6 +15,8 @@
 #include <unordered_map>
 #include <vector>
 
+#include "absl/log/absl_check.h"
+#include "absl/log/absl_log.h"
 #include "absl/strings/str_format.h"
 #include "absl/types/span.h"
 #include "mediapipe/calculators/tflite/tflite_tensors_to_detections_calculator.pb.h"
@@ -93,7 +95,7 @@ void ConvertRawValuesToAnchors(const float* raw_anchors, int num_boxes,
 
 void ConvertAnchorsToRawValues(const std::vector<Anchor>& anchors,
                                int num_boxes, float* raw_anchors) {
-  CHECK_EQ(anchors.size(), num_boxes);
+  ABSL_CHECK_EQ(anchors.size(), num_boxes);
   int box = 0;
   for (const auto& anchor : anchors) {
     raw_anchors[box * kNumCoordsPerBox + 0] = anchor.y_center();
@@ -288,14 +290,14 @@ absl::Status TfLiteTensorsToDetectionsCalculator::ProcessCPU(
     const TfLiteTensor* raw_score_tensor = &input_tensors[1];
 
     // TODO: Add flexible input tensor size handling.
-    CHECK_EQ(raw_box_tensor->dims->size, 3);
+    ABSL_CHECK_EQ(raw_box_tensor->dims->size, 3);
-    CHECK_EQ(raw_box_tensor->dims->data[0], 1);
+    ABSL_CHECK_EQ(raw_box_tensor->dims->data[0], 1);
-    CHECK_EQ(raw_box_tensor->dims->data[1], num_boxes_);
+    ABSL_CHECK_EQ(raw_box_tensor->dims->data[1], num_boxes_);
-    CHECK_EQ(raw_box_tensor->dims->data[2], num_coords_);
+    ABSL_CHECK_EQ(raw_box_tensor->dims->data[2], num_coords_);
-    CHECK_EQ(raw_score_tensor->dims->size, 3);
+    ABSL_CHECK_EQ(raw_score_tensor->dims->size, 3);
-    CHECK_EQ(raw_score_tensor->dims->data[0], 1);
+    ABSL_CHECK_EQ(raw_score_tensor->dims->data[0], 1);
-    CHECK_EQ(raw_score_tensor->dims->data[1], num_boxes_);
+    ABSL_CHECK_EQ(raw_score_tensor->dims->data[1], num_boxes_);
-    CHECK_EQ(raw_score_tensor->dims->data[2], num_classes_);
+    ABSL_CHECK_EQ(raw_score_tensor->dims->data[2], num_classes_);
     const float* raw_boxes = raw_box_tensor->data.f;
     const float* raw_scores = raw_score_tensor->data.f;
 
@@ -303,13 +305,13 @@ absl::Status TfLiteTensorsToDetectionsCalculator::ProcessCPU(
     if (!anchors_init_) {
       if (input_tensors.size() == kNumInputTensorsWithAnchors) {
         const TfLiteTensor* anchor_tensor = &input_tensors[2];
-        CHECK_EQ(anchor_tensor->dims->size, 2);
+        ABSL_CHECK_EQ(anchor_tensor->dims->size, 2);
-        CHECK_EQ(anchor_tensor->dims->data[0], num_boxes_);
+        ABSL_CHECK_EQ(anchor_tensor->dims->data[0], num_boxes_);
-        CHECK_EQ(anchor_tensor->dims->data[1], kNumCoordsPerBox);
+        ABSL_CHECK_EQ(anchor_tensor->dims->data[1], kNumCoordsPerBox);
         const float* raw_anchors = anchor_tensor->data.f;
         ConvertRawValuesToAnchors(raw_anchors, num_boxes_, &anchors_);
       } else if (side_packet_anchors_) {
-        CHECK(!cc->InputSidePackets().Tag("ANCHORS").IsEmpty());
+        ABSL_CHECK(!cc->InputSidePackets().Tag("ANCHORS").IsEmpty());
         anchors_ =
             cc->InputSidePackets().Tag("ANCHORS").Get<std::vector<Anchor>>();
       } else {
@@ -409,7 +411,7 @@ absl::Status TfLiteTensorsToDetectionsCalculator::ProcessGPU(
         CopyBuffer(input_tensors[1], gpu_data_->raw_scores_buffer));
     if (!anchors_init_) {
       if (side_packet_anchors_) {
-        CHECK(!cc->InputSidePackets().Tag("ANCHORS").IsEmpty());
+        ABSL_CHECK(!cc->InputSidePackets().Tag("ANCHORS").IsEmpty());
         const auto& anchors =
             cc->InputSidePackets().Tag("ANCHORS").Get<std::vector<Anchor>>();
         std::vector<float> raw_anchors(num_boxes_ * kNumCoordsPerBox);
@@ -417,7 +419,7 @@ absl::Status TfLiteTensorsToDetectionsCalculator::ProcessGPU(
         MP_RETURN_IF_ERROR(gpu_data_->raw_anchors_buffer.Write<float>(
             absl::MakeSpan(raw_anchors)));
       } else {
-        CHECK_EQ(input_tensors.size(), kNumInputTensorsWithAnchors);
+        ABSL_CHECK_EQ(input_tensors.size(), kNumInputTensorsWithAnchors);
         MP_RETURN_IF_ERROR(
             CopyBuffer(input_tensors[2], gpu_data_->raw_anchors_buffer));
       }
@@ -477,7 +479,7 @@ absl::Status TfLiteTensorsToDetectionsCalculator::ProcessGPU(
                     commandBuffer:[gpu_helper_ commandBuffer]];
   if (!anchors_init_) {
     if (side_packet_anchors_) {
-      CHECK(!cc->InputSidePackets().Tag("ANCHORS").IsEmpty());
+      ABSL_CHECK(!cc->InputSidePackets().Tag("ANCHORS").IsEmpty());
       const auto& anchors =
           cc->InputSidePackets().Tag("ANCHORS").Get<std::vector<Anchor>>();
       std::vector<float> raw_anchors(num_boxes_ * kNumCoordsPerBox);
@@ -541,7 +543,7 @@ absl::Status TfLiteTensorsToDetectionsCalculator::ProcessGPU(
                                          output_detections));
 
 #else
-  LOG(ERROR) << "GPU input on non-Android not supported yet.";
+  ABSL_LOG(ERROR) << "GPU input on non-Android not supported yet.";
 #endif  // MEDIAPIPE_TFLITE_GL_INFERENCE
   return absl::OkStatus();
 }
@@ -567,12 +569,12 @@ absl::Status TfLiteTensorsToDetectionsCalculator::LoadOptions(
   num_coords_ = options_.num_coords();
 
   // Currently only support 2D when num_values_per_keypoint equals to 2.
-  CHECK_EQ(options_.num_values_per_keypoint(), 2);
+  ABSL_CHECK_EQ(options_.num_values_per_keypoint(), 2);
 
   // Check if the output size is equal to the requested boxes and keypoints.
-  CHECK_EQ(options_.num_keypoints() * options_.num_values_per_keypoint() +
+  ABSL_CHECK_EQ(options_.num_keypoints() * options_.num_values_per_keypoint() +
-               kNumCoordsPerBox,
+                    kNumCoordsPerBox,
-           num_coords_);
+                num_coords_);
 
   for (int i = 0; i < options_.ignore_classes_size(); ++i) {
     ignore_classes_.insert(options_.ignore_classes(i));
@@ -897,10 +899,11 @@ void main() {
     int max_wg_size;  //  typically <= 1024
     glGetIntegeri_v(GL_MAX_COMPUTE_WORK_GROUP_SIZE, 1,
                     &max_wg_size);  // y-dim
-    CHECK_LT(num_classes_, max_wg_size)
+    ABSL_CHECK_LT(num_classes_, max_wg_size)
         << "# classes must be < " << max_wg_size;
     // TODO support better filtering.
-    CHECK_LE(ignore_classes_.size(), 1) << "Only ignore class 0 is allowed";
+    ABSL_CHECK_LE(ignore_classes_.size(), 1)
+        << "Only ignore class 0 is allowed";
 
     // Shader program
     GlShader score_shader;
@@ -1115,7 +1118,7 @@ kernel void scoreKernel(
       ignore_classes_.size() ? 1 : 0);
 
   // TODO support better filtering.
-  CHECK_LE(ignore_classes_.size(), 1) << "Only ignore class 0 is allowed";
+  ABSL_CHECK_LE(ignore_classes_.size(), 1) << "Only ignore class 0 is allowed";
 
   {
     // Shader program
@@ -1147,7 +1150,8 @@ kernel void scoreKernel(
                             options:MTLResourceStorageModeShared];
     // # filter classes supported is hardware dependent.
     int max_wg_size = gpu_data_->score_program.maxTotalThreadsPerThreadgroup;
-    CHECK_LT(num_classes_, max_wg_size) << "# classes must be <" << max_wg_size;
+    ABSL_CHECK_LT(num_classes_, max_wg_size)
+        << "# classes must be <" << max_wg_size;
   }
 
 #endif  // MEDIAPIPE_TFLITE_GL_INFERENCE

mediapipe/calculators/tflite/tflite_tensors_to_landmarks_calculator.cc

@@ -12,6 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+#include "absl/log/absl_check.h"
 #include "mediapipe/calculators/tflite/tflite_tensors_to_landmarks_calculator.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/formats/landmark.pb.h"
@@ -199,7 +200,7 @@ absl::Status TfLiteTensorsToLandmarksCalculator::Process(
     num_values *= raw_tensor->dims->data[i];
   }
   const int num_dimensions = num_values / num_landmarks_;
-  CHECK_GT(num_dimensions, 0);
+  ABSL_CHECK_GT(num_dimensions, 0);
 
   const float* raw_landmarks = raw_tensor->data.f;
 

mediapipe/calculators/util/BUILD

@@ -183,9 +183,9 @@ cc_library(
         "//mediapipe/framework:calculator_options_cc_proto",
         "//mediapipe/framework:timestamp",
         "//mediapipe/framework/deps:clock",
-        "//mediapipe/framework/port:logging",
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/framework/port:status",
+        "@com_google_absl//absl/log:absl_log",
         "@com_google_absl//absl/strings",
         "@com_google_absl//absl/time",
     ],
@@ -248,11 +248,12 @@ cc_library(
         ":annotation_overlay_calculator_cc_proto",
         "//mediapipe/framework:calculator_framework",
         "//mediapipe/framework:calculator_options_cc_proto",
+        "//mediapipe/framework/formats:image",
         "//mediapipe/framework/formats:image_format_cc_proto",
         "//mediapipe/framework/formats:image_frame",
         "//mediapipe/framework/formats:image_frame_opencv",
+        "//mediapipe/framework/formats:image_opencv",
         "//mediapipe/framework/formats:video_stream_header",
-        "//mediapipe/framework/port:logging",
         "//mediapipe/framework/port:opencv_core",
         "//mediapipe/framework/port:opencv_imgproc",
         "//mediapipe/framework/port:status",
@@ -260,6 +261,7 @@ cc_library(
         "//mediapipe/util:annotation_renderer",
         "//mediapipe/util:color_cc_proto",
         "//mediapipe/util:render_data_cc_proto",
+        "@com_google_absl//absl/log:absl_log",
         "@com_google_absl//absl/strings",
     ] + select({
         "//mediapipe/gpu:disable_gpu": [],
@@ -267,6 +269,7 @@ cc_library(
             "//mediapipe/gpu:gl_calculator_helper",
             "//mediapipe/gpu:gl_simple_shaders",
             "//mediapipe/gpu:gpu_buffer",
+            "//mediapipe/gpu:gpu_buffer_format",
             "//mediapipe/gpu:shader_util",
         ],
     }),
@@ -374,9 +377,10 @@ cc_library(
         "//mediapipe/framework/formats:detection_cc_proto",
         "//mediapipe/framework/formats:image_frame",
         "//mediapipe/framework/formats:location",
-        "//mediapipe/framework/port:logging",
         "//mediapipe/framework/port:rectangle",
         "//mediapipe/framework/port:status",
+        "@com_google_absl//absl/log:absl_check",
+        "@com_google_absl//absl/log:absl_log",
     ],
     alwayslink = 1,
 )
@@ -675,6 +679,7 @@ cc_library(
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/util:color_cc_proto",
         "//mediapipe/util:render_data_cc_proto",
+        "@com_google_absl//absl/log:absl_check",
         "@com_google_absl//absl/memory",
         "@com_google_absl//absl/strings",
     ],
@@ -731,6 +736,7 @@ cc_library(
         "//mediapipe/framework/port:statusor",
         "//mediapipe/util:color_cc_proto",
         "//mediapipe/util:render_data_cc_proto",
+        "@com_google_absl//absl/log:absl_check",
         "@com_google_absl//absl/strings",
     ],
     alwayslink = 1,
@@ -746,6 +752,7 @@ cc_library(
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/util:color_cc_proto",
         "//mediapipe/util:render_data_cc_proto",
+        "@com_google_absl//absl/log:absl_check",
     ],
     alwayslink = 1,
 )
@@ -1149,6 +1156,7 @@ cc_library(
         "//mediapipe/framework/port:file_helpers",
         "//mediapipe/framework/port:ret_check",
         "//mediapipe/framework/port:status",
+        "@com_google_absl//absl/log:absl_log",
     ],
     alwayslink = 1,
 )
@@ -1209,6 +1217,7 @@ cc_library(
         "//mediapipe/framework/port:rectangle",
         "//mediapipe/framework/port:status",
         "//mediapipe/util:rectangle_util",
+        "@com_google_absl//absl/log:absl_check",
         "@com_google_absl//absl/memory",
     ],
     alwayslink = 1,
@@ -1480,6 +1489,7 @@ cc_library(
         "//mediapipe/framework/formats:landmark_cc_proto",
         "//mediapipe/framework/port:core_proto",
         "//mediapipe/framework/port:ret_check",
+        "@com_google_absl//absl/log:absl_check",
         "@com_google_absl//absl/memory",
     ],
     alwayslink = 1,

mediapipe/calculators/util/annotation_overlay_calculator.cc

@@ -14,15 +14,17 @@
 
 #include <memory>
 
+#include "absl/log/absl_log.h"
 #include "absl/strings/str_cat.h"
 #include "mediapipe/calculators/util/annotation_overlay_calculator.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
 #include "mediapipe/framework/calculator_options.pb.h"
+#include "mediapipe/framework/formats/image.h"
 #include "mediapipe/framework/formats/image_format.pb.h"
 #include "mediapipe/framework/formats/image_frame.h"
 #include "mediapipe/framework/formats/image_frame_opencv.h"
+#include "mediapipe/framework/formats/image_opencv.h"
 #include "mediapipe/framework/formats/video_stream_header.h"
-#include "mediapipe/framework/port/logging.h"
 #include "mediapipe/framework/port/opencv_core_inc.h"
 #include "mediapipe/framework/port/opencv_imgproc_inc.h"
 #include "mediapipe/framework/port/status.h"
@@ -35,6 +37,7 @@
 #include "mediapipe/gpu/gl_calculator_helper.h"
 #include "mediapipe/gpu/gl_simple_shaders.h"
 #include "mediapipe/gpu/gpu_buffer.h"
+#include "mediapipe/gpu/gpu_buffer_format.h"
 #include "mediapipe/gpu/shader_util.h"
 #endif  // !MEDIAPIPE_DISABLE_GPU
 
@@ -45,6 +48,7 @@ namespace {
 constexpr char kVectorTag[] = "VECTOR";
 constexpr char kGpuBufferTag[] = "IMAGE_GPU";
 constexpr char kImageFrameTag[] = "IMAGE";
+constexpr char kImageTag[] = "UIMAGE";  // Universal Image
 
 enum { ATTRIB_VERTEX, ATTRIB_TEXTURE_POSITION, NUM_ATTRIBUTES };
 
@@ -57,13 +61,16 @@ size_t RoundUp(size_t n, size_t m) { return ((n + m - 1) / m) * m; }  // NOLINT
 constexpr uchar kAnnotationBackgroundColor = 2;  // Grayscale value.
 
 // Future Image type.
-inline bool HasImageTag(mediapipe::CalculatorContext* cc) { return false; }
+inline bool HasImageTag(mediapipe::CalculatorContext* cc) {
+  return cc->Inputs().HasTag(kImageTag);
+}
 }  // namespace
 
 // A calculator for rendering data on images.
 //
 // Inputs:
 //  1. IMAGE or IMAGE_GPU (optional): An ImageFrame (or GpuBuffer),
+//     or UIMAGE (an Image).
 //     containing the input image.
 //     If output is CPU, and input isn't provided, the renderer creates a
 //     blank canvas with the width, height and color provided in the options.
@@ -76,6 +83,7 @@ inline bool HasImageTag(mediapipe::CalculatorContext* cc) { return false; }
 //
 // Output:
 //  1. IMAGE or IMAGE_GPU: A rendered ImageFrame (or GpuBuffer),
+//     or UIMAGE (an Image).
 //  Note: Output types should match their corresponding input stream type.
 //
 // For CPU input frames, only SRGBA, SRGB and GRAY8 format are supported. The
@@ -135,6 +143,9 @@ class AnnotationOverlayCalculator : public CalculatorBase {
   absl::Status CreateRenderTargetCpu(CalculatorContext* cc,
                                      std::unique_ptr<cv::Mat>& image_mat,
                                      ImageFormat::Format* target_format);
+  absl::Status CreateRenderTargetCpuImage(CalculatorContext* cc,
+                                          std::unique_ptr<cv::Mat>& image_mat,
+                                          ImageFormat::Format* target_format);
   template <typename Type, const char* Tag>
   absl::Status CreateRenderTargetGpu(CalculatorContext* cc,
                                      std::unique_ptr<cv::Mat>& image_mat);
@@ -172,30 +183,38 @@ class AnnotationOverlayCalculator : public CalculatorBase {
 REGISTER_CALCULATOR(AnnotationOverlayCalculator);
 
 absl::Status AnnotationOverlayCalculator::GetContract(CalculatorContract* cc) {
-  CHECK_GE(cc->Inputs().NumEntries(), 1);
+  RET_CHECK_GE(cc->Inputs().NumEntries(), 1);
 
   bool use_gpu = false;
 
-  if (cc->Inputs().HasTag(kImageFrameTag) &&
+  RET_CHECK(cc->Inputs().HasTag(kImageFrameTag) +
-      cc->Inputs().HasTag(kGpuBufferTag)) {
+                cc->Inputs().HasTag(kGpuBufferTag) +
-    return absl::InternalError("Cannot have multiple input images.");
+                cc->Inputs().HasTag(kImageTag) <=
-  }
+            1);
-  if (cc->Inputs().HasTag(kGpuBufferTag) !=
+  RET_CHECK(cc->Outputs().HasTag(kImageFrameTag) +
-      cc->Outputs().HasTag(kGpuBufferTag)) {
+                cc->Outputs().HasTag(kGpuBufferTag) +
-    return absl::InternalError("GPU output must have GPU input.");
+                cc->Outputs().HasTag(kImageTag) ==
-  }
+            1);
 
   // Input image to render onto copy of. Should be same type as output.
 #if !MEDIAPIPE_DISABLE_GPU
   if (cc->Inputs().HasTag(kGpuBufferTag)) {
     cc->Inputs().Tag(kGpuBufferTag).Set<mediapipe::GpuBuffer>();
-    CHECK(cc->Outputs().HasTag(kGpuBufferTag));
+    RET_CHECK(cc->Outputs().HasTag(kGpuBufferTag));
     use_gpu = true;
   }
 #endif  // !MEDIAPIPE_DISABLE_GPU
   if (cc->Inputs().HasTag(kImageFrameTag)) {
     cc->Inputs().Tag(kImageFrameTag).Set<ImageFrame>();
-    CHECK(cc->Outputs().HasTag(kImageFrameTag));
+    RET_CHECK(cc->Outputs().HasTag(kImageFrameTag));
+  }
+
+  if (cc->Inputs().HasTag(kImageTag)) {
+    cc->Inputs().Tag(kImageTag).Set<mediapipe::Image>();
+    RET_CHECK(cc->Outputs().HasTag(kImageTag));
+#if !MEDIAPIPE_DISABLE_GPU
+    use_gpu = true;  // Prepare GPU resources because images can come in on GPU.
+#endif
   }
 
   // Data streams to render.
@@ -220,6 +239,9 @@ absl::Status AnnotationOverlayCalculator::GetContract(CalculatorContract* cc) {
   if (cc->Outputs().HasTag(kImageFrameTag)) {
     cc->Outputs().Tag(kImageFrameTag).Set<ImageFrame>();
   }
+  if (cc->Outputs().HasTag(kImageTag)) {
+    cc->Outputs().Tag(kImageTag).Set<mediapipe::Image>();
+  }
 
   if (use_gpu) {
 #if !MEDIAPIPE_DISABLE_GPU
@@ -252,9 +274,14 @@ absl::Status AnnotationOverlayCalculator::Open(CalculatorContext* cc) {
   renderer_ = absl::make_unique<AnnotationRenderer>();
   renderer_->SetFlipTextVertically(options_.flip_text_vertically());
   if (use_gpu_) renderer_->SetScaleFactor(options_.gpu_scale_factor());
+  if (renderer_->GetScaleFactor() < 1.0 && HasImageTag(cc))
+    ABSL_LOG(WARNING)
+        << "Annotation scale factor only supports GPU backed Image.";
 
   // Set the output header based on the input header (if present).
-  const char* tag = use_gpu_ ? kGpuBufferTag : kImageFrameTag;
+  const char* tag = HasImageTag(cc) ? kImageTag
+                    : use_gpu_      ? kGpuBufferTag
+                                    : kImageFrameTag;
   if (image_frame_available_ && !cc->Inputs().Tag(tag).Header().IsEmpty()) {
     const auto& input_header =
         cc->Inputs().Tag(tag).Header().Get<VideoHeader>();
@@ -280,6 +307,12 @@ absl::Status AnnotationOverlayCalculator::Process(CalculatorContext* cc) {
       cc->Inputs().Tag(kImageFrameTag).IsEmpty()) {
     return absl::OkStatus();
   }
+  if (cc->Inputs().HasTag(kImageTag) && cc->Inputs().Tag(kImageTag).IsEmpty()) {
+    return absl::OkStatus();
+  }
+  if (HasImageTag(cc)) {
+    use_gpu_ = cc->Inputs().Tag(kImageTag).Get<mediapipe::Image>().UsesGpu();
+  }
 
   // Initialize render target, drawn with OpenCV.
   std::unique_ptr<cv::Mat> image_mat;
@@ -289,10 +322,17 @@ absl::Status AnnotationOverlayCalculator::Process(CalculatorContext* cc) {
     if (!gpu_initialized_) {
       MP_RETURN_IF_ERROR(
           gpu_helper_.RunInGlContext([this, cc]() -> absl::Status {
+            if (HasImageTag(cc)) {
+              return GlSetup<mediapipe::Image, kImageTag>(cc);
+            }
             return GlSetup<mediapipe::GpuBuffer, kGpuBufferTag>(cc);
           }));
       gpu_initialized_ = true;
     }
+    if (HasImageTag(cc)) {
+      MP_RETURN_IF_ERROR(
+          (CreateRenderTargetGpu<mediapipe::Image, kImageTag>(cc, image_mat)));
+    }
     if (cc->Inputs().HasTag(kGpuBufferTag)) {
       MP_RETURN_IF_ERROR(
           (CreateRenderTargetGpu<mediapipe::GpuBuffer, kGpuBufferTag>(
@@ -300,6 +340,10 @@ absl::Status AnnotationOverlayCalculator::Process(CalculatorContext* cc) {
     }
 #endif  // !MEDIAPIPE_DISABLE_GPU
   } else {
+    if (cc->Outputs().HasTag(kImageTag)) {
+      MP_RETURN_IF_ERROR(
+          CreateRenderTargetCpuImage(cc, image_mat, &target_format));
+    }
     if (cc->Outputs().HasTag(kImageFrameTag)) {
       MP_RETURN_IF_ERROR(CreateRenderTargetCpu(cc, image_mat, &target_format));
     }
@@ -339,6 +383,9 @@ absl::Status AnnotationOverlayCalculator::Process(CalculatorContext* cc) {
     uchar* image_mat_ptr = image_mat->data;
     MP_RETURN_IF_ERROR(
         gpu_helper_.RunInGlContext([this, cc, image_mat_ptr]() -> absl::Status {
+          if (HasImageTag(cc)) {
+            return RenderToGpu<mediapipe::Image, kImageTag>(cc, image_mat_ptr);
+          }
           return RenderToGpu<mediapipe::GpuBuffer, kGpuBufferTag>(
               cc, image_mat_ptr);
         }));
@@ -381,6 +428,10 @@ absl::Status AnnotationOverlayCalculator::RenderToCpu(
                               ImageFrame::kDefaultAlignmentBoundary);
 #endif  // !MEDIAPIPE_DISABLE_GPU
 
+  if (HasImageTag(cc)) {
+    auto out = std::make_unique<mediapipe::Image>(std::move(output_frame));
+    cc->Outputs().Tag(kImageTag).Add(out.release(), cc->InputTimestamp());
+  }
   if (cc->Outputs().HasTag(kImageFrameTag)) {
     cc->Outputs()
         .Tag(kImageFrameTag)
@@ -399,7 +450,8 @@ absl::Status AnnotationOverlayCalculator::RenderToGpu(CalculatorContext* cc,
   auto input_texture = gpu_helper_.CreateSourceTexture(input_frame);
 
   auto output_texture = gpu_helper_.CreateDestinationTexture(
-      width_, height_, mediapipe::GpuBufferFormat::kBGRA32);
+      input_texture.width(), input_texture.height(),
+      mediapipe::GpuBufferFormat::kBGRA32);
 
   // Upload render target to GPU.
   {
@@ -428,7 +480,7 @@ absl::Status AnnotationOverlayCalculator::RenderToGpu(CalculatorContext* cc,
   }
 
   // Send out blended image as GPU packet.
-  auto output_frame = output_texture.GetFrame<Type>();
+  auto output_frame = output_texture.template GetFrame<Type>();
   cc->Outputs().Tag(Tag).Add(output_frame.release(), cc->InputTimestamp());
 
   // Cleanup
@@ -487,6 +539,54 @@ absl::Status AnnotationOverlayCalculator::CreateRenderTargetCpu(
   return absl::OkStatus();
 }
 
+absl::Status AnnotationOverlayCalculator::CreateRenderTargetCpuImage(
+    CalculatorContext* cc, std::unique_ptr<cv::Mat>& image_mat,
+    ImageFormat::Format* target_format) {
+  if (image_frame_available_) {
+    const auto& input_frame =
+        cc->Inputs().Tag(kImageTag).Get<mediapipe::Image>();
+
+    int target_mat_type;
+    switch (input_frame.image_format()) {
+      case ImageFormat::SRGBA:
+        *target_format = ImageFormat::SRGBA;
+        target_mat_type = CV_8UC4;
+        break;
+      case ImageFormat::SRGB:
+        *target_format = ImageFormat::SRGB;
+        target_mat_type = CV_8UC3;
+        break;
+      case ImageFormat::GRAY8:
+        *target_format = ImageFormat::SRGB;
+        target_mat_type = CV_8UC3;
+        break;
+      default:
+        return absl::UnknownError("Unexpected image frame format.");
+        break;
+    }
+
+    image_mat = absl::make_unique<cv::Mat>(
+        input_frame.height(), input_frame.width(), target_mat_type);
+
+    auto input_mat = formats::MatView(&input_frame);
+    if (input_frame.image_format() == ImageFormat::GRAY8) {
+      cv::Mat rgb_mat;
+      cv::cvtColor(*input_mat, rgb_mat, cv::COLOR_GRAY2RGB);
+      rgb_mat.copyTo(*image_mat);
+    } else {
+      input_mat->copyTo(*image_mat);
+    }
+  } else {
+    image_mat = absl::make_unique<cv::Mat>(
+        options_.canvas_height_px(), options_.canvas_width_px(), CV_8UC3,
+        cv::Scalar(options_.canvas_color().r(), options_.canvas_color().g(),
+                   options_.canvas_color().b()));
+    *target_format = ImageFormat::SRGB;
+  }
+
+  return absl::OkStatus();
+}
+
 template <typename Type, const char* Tag>
 absl::Status AnnotationOverlayCalculator::CreateRenderTargetGpu(
     CalculatorContext* cc, std::unique_ptr<cv::Mat>& image_mat) {

mediapipe/calculators/util/association_calculator.h

@@ -18,6 +18,7 @@
 #include <memory>
 #include <vector>
 
+#include "absl/log/absl_check.h"
 #include "absl/memory/memory.h"
 #include "mediapipe/calculators/util/association_calculator.pb.h"
 #include "mediapipe/framework/calculator_context.h"
@@ -72,7 +73,7 @@ class AssociationCalculator : public CalculatorBase {
       prev_input_stream_id_ = cc->Inputs().GetId("PREV", 0);
     }
     options_ = cc->Options<::mediapipe::AssociationCalculatorOptions>();
-    CHECK_GE(options_.min_similarity_threshold(), 0);
+    ABSL_CHECK_GE(options_.min_similarity_threshold(), 0);
 
     return absl::OkStatus();
   }

mediapipe/calculators/util/detection_label_id_to_text_calculator.cc

@@ -19,6 +19,7 @@
 #include "mediapipe/framework/port/integral_types.h"
 #include "mediapipe/framework/port/proto_ns.h"
 #include "mediapipe/framework/port/status.h"
+#include "mediapipe/framework/port/status_macros.h"
 #include "mediapipe/util/label_map.pb.h"
 #include "mediapipe/util/resource_util.h"
 
@@ -85,7 +86,8 @@ absl::Status DetectionLabelIdToTextCalculator::Open(CalculatorContext* cc) {
     ASSIGN_OR_RETURN(string_path,
                      PathToResourceAsFile(options.label_map_path()));
     std::string label_map_string;
-    MP_RETURN_IF_ERROR(file::GetContents(string_path, &label_map_string));
+    MP_RETURN_IF_ERROR(
+        mediapipe::GetResourceContents(string_path, &label_map_string));
 
     std::istringstream stream(label_map_string);
     std::string line;

mediapipe/calculators/util/detections_to_render_data_calculator.cc

@@ -12,6 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
+#include "absl/log/absl_check.h"
 #include "absl/memory/memory.h"
 #include "absl/strings/str_cat.h"
 #include "absl/strings/str_join.h"
@@ -233,13 +234,13 @@ void DetectionsToRenderDataCalculator::AddLabels(
     const Detection& detection,
     const DetectionsToRenderDataCalculatorOptions& options,
     float text_line_height, RenderData* render_data) {
-  CHECK(detection.label().empty() || detection.label_id().empty() ||
+  ABSL_CHECK(detection.label().empty() || detection.label_id().empty() ||
-        detection.label_size() == detection.label_id_size())
+             detection.label_size() == detection.label_id_size())
       << "String or integer labels should be of same size. Or only one of them "
          "is present.";
   const auto num_labels =
       std::max(detection.label_size(), detection.label_id_size());
-  CHECK_EQ(detection.score_size(), num_labels)
+  ABSL_CHECK_EQ(detection.score_size(), num_labels)
       << "Number of scores and labels should match for detection.";
 
   // Extracts all "label(_id),score" for the detection.
@@ -361,9 +362,9 @@ void DetectionsToRenderDataCalculator::AddDetectionToRenderData(
     const Detection& detection,
     const DetectionsToRenderDataCalculatorOptions& options,
     RenderData* render_data) {
-  CHECK(detection.location_data().format() == LocationData::BOUNDING_BOX ||
+  ABSL_CHECK(detection.location_data().format() == LocationData::BOUNDING_BOX ||
-        detection.location_data().format() ==
+             detection.location_data().format() ==
-            LocationData::RELATIVE_BOUNDING_BOX)
+                 LocationData::RELATIVE_BOUNDING_BOX)
       << "Only Detection with formats of BOUNDING_BOX or RELATIVE_BOUNDING_BOX "
          "are supported.";
   double text_line_height;

mediapipe/calculators/util/labels_to_render_data_calculator.cc

@@ -19,6 +19,7 @@
 #include <string>
 #include <vector>
 
+#include "absl/log/absl_check.h"
 #include "absl/strings/str_cat.h"
 #include "mediapipe/calculators/util/labels_to_render_data_calculator.pb.h"
 #include "mediapipe/framework/calculator_framework.h"
@@ -114,7 +115,8 @@ absl::Status LabelsToRenderDataCalculator::Process(CalculatorContext* cc) {
     video_height_ = video_header.height;
     return absl::OkStatus();
   } else {
-    CHECK_EQ(options_.location(), LabelsToRenderDataCalculatorOptions::TOP_LEFT)
+    ABSL_CHECK_EQ(options_.location(),
+                  LabelsToRenderDataCalculatorOptions::TOP_LEFT)
         << "Only TOP_LEFT is supported without VIDEO_PRESTREAM.";
   }
 
@@ -144,7 +146,7 @@ absl::Status LabelsToRenderDataCalculator::Process(CalculatorContext* cc) {
     if (cc->Inputs().HasTag(kScoresTag)) {
       std::vector<float> score_vector =
           cc->Inputs().Tag(kScoresTag).Get<std::vector<float>>();
-      CHECK_EQ(label_vector.size(), score_vector.size());
+      ABSL_CHECK_EQ(label_vector.size(), score_vector.size());
       scores.resize(label_vector.size());
       for (int i = 0; i < label_vector.size(); ++i) {
         scores[i] = score_vector[i];

mediapipe/calculators/util/landmarks_refinement_calculator.cc

@@ -18,6 +18,7 @@
 #include <set>
 #include <utility>
 
+#include "absl/log/absl_check.h"
 #include "absl/memory/memory.h"
 #include "mediapipe/calculators/util/landmarks_refinement_calculator.pb.h"
 #include "mediapipe/framework/api2/node.h"
@@ -102,7 +103,8 @@ void RefineZ(
           ->set_z(z_average);
     }
   } else {
-    CHECK(false) << "Z refinement is either not specified or not supported";
+    ABSL_CHECK(false)
+        << "Z refinement is either not specified or not supported";
   }
 }