Compare commits

..
Author SHA1 Message Date
dependabot[bot]andGitHub 4ce9284ca8 Bump word-wrap from 1.2.3 to 1.2.4
Bumps [word-wrap](https://github.com/jonschlinkert/word-wrap) from 1.2.3 to 1.2.4.
- [Release notes](https://github.com/jonschlinkert/word-wrap/releases)
- [Commits](https://github.com/jonschlinkert/word-wrap/compare/1.2.3...1.2.4)

---
updated-dependencies:
- dependency-name: word-wrap
  dependency-type: indirect
...

Signed-off-by: dependabot[bot] <[email protected]>
2023-07-20 01:06:45 +00:00
934 changed files with 9736 additions and 32610 deletions
@@ -40,16 +40,18 @@ body:
label: Programming Language and version (e.g. C++, Python, Java)
validations:
required: true
- type: input
- type: textarea
id: current_model
attributes:
label: Describe the actual behavior
render: shell
validations:
required: true
- type: input
- type: textarea
id: expected_model
attributes:
label: Describe the expected behaviour
render: shell
validations:
required: true
- type: textarea
@@ -41,16 +41,18 @@ body:
label: Task name (e.g. Image classification, Gesture recognition etc.)
validations:
required: true
- type: input
- type: textarea
id: current_model
attributes:
label: Describe the actual behavior
render: shell
validations:
required: true
- type: input
- type: textarea
id: expected_model
attributes:
label: Describe the expected behaviour
render: shell
validations:
required: true
- type: textarea
@@ -31,16 +31,18 @@ body:
label: URL that shows the problem
validations:
required: false
- type: input
- type: textarea
id: current_model
attributes:
label: Describe the actual behavior
render: shell
validations:
required: false
- type: input
- type: textarea
id: expected_model
attributes:
label: Describe the expected behaviour
render: shell
validations:
required: false
- type: textarea
@@ -28,33 +28,37 @@ body:
- 'No'
validations:
required: false
- type: input
- type: textarea
id: behaviour
attributes:
label: Describe the feature and the current behaviour/state
render: shell
validations:
required: true
- type: input
- type: textarea
id: api_change
attributes:
label: Will this change the current API? How?
render: shell
validations:
required: false
- type: input
- type: textarea
id: benifit
attributes:
label: Who will benefit with this feature?
validations:
required: false
- type: input
- type: textarea
id: use_case
attributes:
label: Please specify the use cases for this feature
render: shell
validations:
required: true
- type: input
- type: textarea
id: info_other
attributes:
label: Any Other info
render: shell
validations:
required: false
@@ -87,13 +87,14 @@ body:
placeholder:
validations:
required: false
- type: input
- type: textarea
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
@@ -80,16 +80,18 @@ body:
label: Xcode & Tulsi version (if issue is related to building for iOS)
validations:
required: false
- type: input
- type: textarea
id: current_model
attributes:
label: Describe the actual behavior
render: shell
validations:
required: true
- type: input
- type: textarea
id: expected_model
attributes:
label: Describe the expected behaviour
render: shell
validations:
required: true
- type: textarea
@@ -48,16 +48,18 @@ body:
placeholder: e.g. C++, Python, Java
validations:
required: false
- type: input
- type: textarea
id: current_model
attributes:
label: Describe the actual behavior
render: shell
validations:
required: false
- type: input
- type: textarea
id: expected_model
attributes:
label: Describe the expected behaviour
render: shell
validations:
required: false
- type: textarea
+1 -3
View File
@@ -39,9 +39,7 @@ jobs:
# Limit the No. of API calls in one run default value is 30.
operations-per-run: 500
# Prevent to remove stale label when PRs or issues are updated.
remove-stale-when-updated: true
# List of labels to remove when issues/PRs unstale.
labels-to-remove-when-unstale: 'stat:awaiting response'
remove-stale-when-updated: false
# comment on issue if not active for more then 7 days.
stale-issue-message: 'This issue has been marked stale because it has no recent activity since 7 days. It will be closed if no further activity occurs. Thank you.'
# comment on PR if not active for more then 14 days.
+9 -26
View File
@@ -73,9 +73,12 @@ http_archive(
http_archive(
name = "zlib",
build_file = "@//third_party:zlib.BUILD",
sha256 = "b3a24de97a8fdbc835b9833169501030b8977031bcb54b3b3ac13740f846ab30",
strip_prefix = "zlib-1.2.13",
url = "http://zlib.net/fossils/zlib-1.2.13.tar.gz",
sha256 = "c3e5e9fdd5004dcb542feda5ee4f0ff0744628baf8ed2dd5d66f8ca1197cb1a1",
strip_prefix = "zlib-1.2.11",
urls = [
"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",
],
@@ -176,25 +179,6 @@ http_archive(
],
)
# 2023-06-05
# This version of Glog is required for Windows support, but currently causes
# crashes on some Android devices.
http_archive(
name = "com_github_glog_glog_windows",
strip_prefix = "glog-3a0d4d22c5ae0b9a2216988411cfa6bf860cc372",
sha256 = "170d08f80210b82d95563f4723a15095eff1aad1863000e8eeb569c96a98fefb",
urls = [
"https://github.com/google/glog/archive/3a0d4d22c5ae0b9a2216988411cfa6bf860cc372.zip",
],
patches = [
"@//third_party:com_github_glog_glog.diff",
"@//third_party:com_github_glog_glog_windows_patch.diff",
],
patch_args = [
"-p1",
],
)
# easyexif
http_archive(
name = "easyexif",
@@ -501,10 +485,10 @@ http_archive(
)
# TensorFlow repo should always go after the other external dependencies.
# TF on 2023-07-26.
_TENSORFLOW_GIT_COMMIT = "e92261fd4cec0b726692081c4d2966b75abf31dd"
# TF on 2023-06-13.
_TENSORFLOW_GIT_COMMIT = "491681a5620e41bf079a582ac39c585cc86878b9"
# curl -L https://github.com/tensorflow/tensorflow/archive/<TENSORFLOW_GIT_COMMIT>.tar.gz | shasum -a 256
_TENSORFLOW_SHA256 = "478a229bd4ec70a5b568ac23b5ea013d9fca46a47d6c43e30365a0412b9febf4"
_TENSORFLOW_SHA256 = "9f76389af7a2835e68413322c1eaabfadc912f02a76d71dc16be507f9ca3d3ac"
http_archive(
name = "org_tensorflow",
urls = [
@@ -512,7 +496,6 @@ 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",
],
+2 -2
View File
@@ -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)) {
ABSL_LOG(INFO) << packet.Get<string>();
LOG(INFO) << packet.Get<string>();
}
```
+1 -1
View File
@@ -138,7 +138,7 @@ Create a `BUILD` file in the `$APPLICATION_PATH` and add the following build
rules:
```
MIN_IOS_VERSION = "12.0"
MIN_IOS_VERSION = "11.0"
load(
"@build_bazel_rules_apple//apple:ios.bzl",
+43 -54
View File
@@ -14,54 +14,57 @@
licenses(["notice"]) # Apache 2.0
load("@mediapipe//mediapipe:platforms.bzl", "config_setting_and_platform")
# Note: yes, these need to use "//external:android/crosstool", not
# @androidndk//:default_crosstool.
# Generic Android
config_setting(
name = "android",
constraint_values = [
"@platforms//os:android",
],
values = {"crosstool_top": "//external:android/crosstool"},
visibility = ["//visibility:public"],
)
# Android x86 32-bit.
config_setting_and_platform(
config_setting(
name = "android_x86",
constraint_values = [
"@platforms//os:android",
"@platforms//cpu:x86_32",
],
values = {
"crosstool_top": "//external:android/crosstool",
"cpu": "x86",
},
visibility = ["//visibility:public"],
)
# Android x86 64-bit.
config_setting_and_platform(
config_setting(
name = "android_x86_64",
constraint_values = [
"@platforms//os:android",
"@platforms//cpu:x86_64",
],
values = {
"crosstool_top": "//external:android/crosstool",
"cpu": "x86_64",
},
visibility = ["//visibility:public"],
)
# Android ARMv7.
config_setting_and_platform(
config_setting(
name = "android_armeabi",
values = {
"crosstool_top": "//external:android/crosstool",
"cpu": "armeabi",
},
visibility = ["//visibility:public"],
)
config_setting(
name = "android_arm",
constraint_values = [
"@platforms//os:android",
"@platforms//cpu:armv7",
],
values = {
"crosstool_top": "//external:android/crosstool",
"cpu": "armeabi-v7a",
},
visibility = ["//visibility:public"],
)
# Android ARM64.
config_setting_and_platform(
config_setting(
name = "android_arm64",
constraint_values = [
"@platforms//os:android",
"@platforms//cpu:arm64",
],
values = {
"crosstool_top": "//external:android/crosstool",
"cpu": "arm64-v8a",
},
visibility = ["//visibility:public"],
)
@@ -75,7 +78,7 @@ config_setting(
)
# MacOS x86 64-bit.
config_setting_and_platform(
config_setting(
name = "macos_x86_64",
constraint_values = [
"@platforms//os:macos",
@@ -85,7 +88,7 @@ config_setting_and_platform(
)
# MacOS ARM64.
config_setting_and_platform(
config_setting(
name = "macos_arm64",
constraint_values = [
"@platforms//os:macos",
@@ -104,7 +107,7 @@ config_setting(
)
# iOS device ARM32.
config_setting_and_platform(
config_setting(
name = "ios_armv7",
constraint_values = [
"@platforms//os:ios",
@@ -114,7 +117,7 @@ config_setting_and_platform(
)
# iOS device ARM64.
config_setting_and_platform(
config_setting(
name = "ios_arm64",
constraint_values = [
"@platforms//os:ios",
@@ -124,7 +127,7 @@ config_setting_and_platform(
)
# iOS device ARM64E.
config_setting_and_platform(
config_setting(
name = "ios_arm64e",
constraint_values = [
"@platforms//os:ios",
@@ -134,7 +137,7 @@ config_setting_and_platform(
)
# iOS simulator x86 32-bit.
config_setting_and_platform(
config_setting(
name = "ios_i386",
constraint_values = [
"@platforms//os:ios",
@@ -145,7 +148,7 @@ config_setting_and_platform(
)
# iOS simulator x86 64-bit.
config_setting_and_platform(
config_setting(
name = "ios_x86_64",
constraint_values = [
"@platforms//os:ios",
@@ -156,7 +159,7 @@ config_setting_and_platform(
)
# iOS simulator ARM64.
config_setting_and_platform(
config_setting(
name = "ios_sim_arm64",
constraint_values = [
"@platforms//os:ios",
@@ -166,6 +169,7 @@ config_setting_and_platform(
visibility = ["//visibility:public"],
)
# Generic Apple.
alias(
name = "apple",
actual = select({
@@ -176,24 +180,9 @@ alias(
visibility = ["//visibility:public"],
)
# Windows 64-bit.
config_setting_and_platform(
config_setting(
name = "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"],
values = {"cpu": "x64_windows"},
)
exports_files(
+1 -9
View File
@@ -12,7 +12,6 @@
# 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"])
@@ -146,7 +145,6 @@ 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",
@@ -165,9 +163,8 @@ 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",
@@ -188,7 +185,6 @@ 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,
)
@@ -228,7 +224,6 @@ 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",
],
@@ -299,7 +294,6 @@ 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",
],
@@ -333,7 +327,6 @@ 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",
],
)
@@ -352,7 +345,6 @@ 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",
],
@@ -23,7 +23,6 @@
#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"
@@ -139,7 +138,7 @@ absl::Status FramewiseTransformCalculatorBase::Process(CalculatorContext* cc) {
TransformFrame(input_frame, &output_frame);
// Copy output from vector<float> to Eigen::Vector.
ABSL_CHECK_EQ(output_frame.size(), num_output_channels_);
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>();
@@ -16,8 +16,6 @@
#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;
@@ -47,9 +45,9 @@ void CopyVectorToChannel(const std::vector<float>& vec, Matrix* matrix,
if (matrix->cols() == 0) {
matrix->resize(matrix->rows(), vec.size());
} else {
ABSL_CHECK_EQ(vec.size(), matrix->cols());
CHECK_EQ(vec.size(), matrix->cols());
}
ABSL_CHECK_LT(channel, matrix->rows());
CHECK_LT(channel, matrix->rows());
matrix->row(channel) =
Eigen::Map<const Eigen::ArrayXf>(vec.data(), vec.size());
}
@@ -79,7 +77,7 @@ absl::Status RationalFactorResampleCalculator::Open(CalculatorContext* cc) {
r = ResamplerFromOptions(source_sample_rate_, target_sample_rate_,
resample_options);
if (!r) {
ABSL_LOG(ERROR) << "Failed to initialize resampler.";
LOG(ERROR) << "Failed to initialize resampler.";
return absl::UnknownError("Failed to initialize resampler.");
}
}
@@ -27,6 +27,7 @@
#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 {
@@ -22,7 +22,6 @@
#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"
@@ -883,11 +882,11 @@ void BM_ProcessDC(benchmark::State& state) {
const CalculatorRunner::StreamContents& output = runner.Outputs().Index(0);
const Matrix& output_matrix = output.packets[0].Get<Matrix>();
ABSL_LOG(INFO) << "Output matrix=" << output_matrix.rows() << "x"
<< output_matrix.cols();
ABSL_LOG(INFO) << "First values=" << output_matrix(0, 0) << ", "
<< output_matrix(1, 0) << ", " << output_matrix(2, 0) << ", "
<< output_matrix(3, 0);
LOG(INFO) << "Output matrix=" << output_matrix.rows() << "x"
<< output_matrix.cols();
LOG(INFO) << "First values=" << output_matrix(0, 0) << ", "
<< output_matrix(1, 0) << ", " << output_matrix(2, 0) << ", "
<< output_matrix(3, 0);
}
BENCHMARK(BM_ProcessDC);
@@ -18,7 +18,6 @@
#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"
@@ -60,7 +59,7 @@ class StabilizedLogCalculator : public CalculatorBase {
output_scale_ = stabilized_log_calculator_options.output_scale();
check_nonnegativity_ =
stabilized_log_calculator_options.check_nonnegativity();
ABSL_CHECK_GE(stabilizer_, 0.0)
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.
@@ -18,7 +18,6 @@
#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"
@@ -105,7 +104,7 @@ 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_));
ABSL_CHECK_EQ(current_output_frame_start, cumulative_completed_samples_);
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;
@@ -17,7 +17,6 @@
#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"
@@ -71,7 +70,7 @@ void BM_TimeSeriesFramerCalculator(benchmark::State& state) {
}
// Initialize graph.
mediapipe::CalculatorGraph graph;
ABSL_CHECK_OK(graph.Initialize(config));
CHECK_OK(graph.Initialize(config));
// Prepare input header.
auto header = std::make_unique<mediapipe::TimeSeriesHeader>();
header->set_sample_rate(kSampleRate);
@@ -79,13 +78,13 @@ void BM_TimeSeriesFramerCalculator(benchmark::State& state) {
state.ResumeTiming(); // Resume benchmark timing.
ABSL_CHECK_OK(graph.StartRun({}, {{"input", Adopt(header.release())}}));
CHECK_OK(graph.StartRun({}, {{"input", Adopt(header.release())}}));
for (auto& packet : input_packets) {
ABSL_CHECK_OK(graph.AddPacketToInputStream("input", packet));
CHECK_OK(graph.AddPacketToInputStream("input", packet));
}
ABSL_CHECK(!graph.HasError());
ABSL_CHECK_OK(graph.CloseAllInputStreams());
ABSL_CHECK_OK(graph.WaitUntilIdle());
CHECK(!graph.HasError());
CHECK_OK(graph.CloseAllInputStreams());
CHECK_OK(graph.WaitUntilIdle());
}
}
BENCHMARK(BM_TimeSeriesFramerCalculator);
@@ -19,7 +19,6 @@
#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"
@@ -187,12 +186,11 @@ class TimeSeriesFramerCalculatorTest
const int num_unique_output_samples =
round((output().packets.size() - 1) * frame_step_samples) +
frame_duration_samples;
ABSL_LOG(INFO) << "packets.size()=" << output().packets.size()
<< " frame_duration_samples=" << frame_duration_samples
<< " frame_step_samples=" << frame_step_samples
<< " num_input_samples_=" << num_input_samples_
<< " num_unique_output_samples="
<< num_unique_output_samples;
LOG(INFO) << "packets.size()=" << output().packets.size()
<< " frame_duration_samples=" << frame_duration_samples
<< " frame_step_samples=" << frame_step_samples
<< " num_input_samples_=" << num_input_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()) {
+1 -34
View File
@@ -325,7 +325,6 @@ cc_library(
":concatenate_vector_calculator_cc_proto",
"//mediapipe/framework:calculator_framework",
"//mediapipe/framework/api2:node",
"//mediapipe/framework/formats:body_rig_cc_proto",
"//mediapipe/framework/formats:classification_cc_proto",
"//mediapipe/framework/formats:landmark_cc_proto",
"//mediapipe/framework/port:ret_check",
@@ -583,7 +582,6 @@ cc_library(
"//mediapipe/framework/port:logging",
"//mediapipe/framework/port:status",
"//mediapipe/framework/tool:options_util",
"@com_google_absl//absl/log:absl_check",
],
alwayslink = 1,
)
@@ -599,7 +597,6 @@ 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",
],
)
@@ -632,7 +629,6 @@ cc_library(
"//mediapipe/framework:calculator_framework",
"//mediapipe/framework/port:ret_check",
"//mediapipe/framework/port:status",
"@com_google_absl//absl/log:absl_log",
],
alwayslink = 1,
)
@@ -780,11 +776,10 @@ 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,
@@ -840,7 +835,6 @@ 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",
],
)
@@ -945,7 +939,6 @@ cc_library(
deps = [
":split_vector_calculator_cc_proto",
"//mediapipe/framework:calculator_framework",
"//mediapipe/framework/formats:body_rig_cc_proto",
"//mediapipe/framework/formats:landmark_cc_proto",
"//mediapipe/framework/port:ret_check",
"//mediapipe/framework/port:status",
@@ -1029,7 +1022,6 @@ cc_library(
"//mediapipe/framework:calculator_framework",
"//mediapipe/framework/api2:node",
"//mediapipe/framework/port:status",
"@com_google_absl//absl/log:absl_log",
],
alwayslink = 1,
)
@@ -1068,7 +1060,6 @@ cc_test(
"//mediapipe/framework:calculator_runner",
"//mediapipe/framework/port:gtest_main",
"//mediapipe/framework/port:parse_text_proto",
"@com_google_absl//absl/log:absl_log",
],
)
@@ -1115,7 +1106,6 @@ cc_library(
"//mediapipe/framework/api2:node",
"//mediapipe/framework/port:ret_check",
"//mediapipe/framework/port:status",
"@com_google_absl//absl/log:absl_log",
],
alwayslink = 1,
)
@@ -1391,26 +1381,3 @@ cc_test(
"@com_google_absl//absl/types:optional",
],
)
cc_library(
name = "value_or_default_calculator",
srcs = ["value_or_default_calculator.cc"],
visibility = ["//visibility:public"],
deps = [
"//mediapipe/framework:calculator_framework",
"//mediapipe/framework/port:status",
],
alwayslink = True,
)
cc_test(
name = "value_or_default_calculator_test",
srcs = ["value_or_default_calculator_test.cc"],
deps = [
":value_or_default_calculator",
"//mediapipe/framework:calculator_framework",
"//mediapipe/framework:calculator_runner",
"//mediapipe/framework:packet",
"//mediapipe/framework/port:gtest_main",
],
)
@@ -13,7 +13,6 @@
// limitations under the License.
#include <string>
#include <utility>
#include <vector>
#include "absl/memory/memory.h"
@@ -164,75 +163,6 @@ TEST_F(BeginEndLoopCalculatorGraphTest, MultipleVectors) {
PacketOfIntsEq(input_timestamp2, std::vector<int>{3, 4})));
}
TEST(BeginEndLoopCalculatorPossibleDataRaceTest,
EndLoopForIntegersDoesNotRace) {
auto graph_config = ParseTextProtoOrDie<CalculatorGraphConfig>(
R"pb(
num_threads: 4
input_stream: "ints"
node {
calculator: "BeginLoopIntegerCalculator"
input_stream: "ITERABLE:ints"
output_stream: "ITEM:int"
output_stream: "BATCH_END:timestamp"
}
node {
calculator: "IncrementCalculator"
input_stream: "int"
output_stream: "int_plus_one"
}
# BEGIN: Data race possibility
# EndLoop###Calculator and another calculator using the same input
# may introduce race due to EndLoop###Calculator possibly consuming
# packet.
node {
calculator: "EndLoopIntegersCalculator"
input_stream: "ITEM:int_plus_one"
input_stream: "BATCH_END:timestamp"
output_stream: "ITERABLE:ints_plus_one"
}
node {
calculator: "IncrementCalculator"
input_stream: "int_plus_one"
output_stream: "int_plus_two"
}
# END: Data race possibility
node {
calculator: "EndLoopIntegersCalculator"
input_stream: "ITEM:int_plus_two"
input_stream: "BATCH_END:timestamp"
output_stream: "ITERABLE:ints_plus_two"
}
)pb");
std::vector<Packet> int_plus_one_packets;
tool::AddVectorSink("ints_plus_one", &graph_config, &int_plus_one_packets);
std::vector<Packet> int_original_packets;
tool::AddVectorSink("ints_plus_two", &graph_config, &int_original_packets);
CalculatorGraph graph;
MP_ASSERT_OK(graph.Initialize(graph_config));
MP_ASSERT_OK(graph.StartRun({}));
for (int i = 0; i < 100; ++i) {
std::vector<int> ints = {i, i + 1, i + 2};
Timestamp ts = Timestamp(i);
MP_ASSERT_OK(graph.AddPacketToInputStream(
"ints", MakePacket<std::vector<int>>(std::move(ints)).At(ts)));
MP_ASSERT_OK(graph.WaitUntilIdle());
EXPECT_THAT(int_plus_one_packets,
testing::ElementsAre(
PacketOfIntsEq(ts, std::vector<int>{i + 1, i + 2, i + 3})));
EXPECT_THAT(int_original_packets,
testing::ElementsAre(
PacketOfIntsEq(ts, std::vector<int>{i + 2, i + 3, i + 4})));
int_plus_one_packets.clear();
int_original_packets.clear();
}
MP_ASSERT_OK(graph.CloseAllPacketSources());
MP_ASSERT_OK(graph.WaitUntilDone());
}
// Passes non empty vector through or outputs empty vector in case of timestamp
// bound update.
class PassThroughOrEmptyVectorCalculator : public CalculatorBase {
@@ -92,7 +92,7 @@ class BypassCalculator : public Node {
auto options = cc->Options<BypassCalculatorOptions>();
RET_CHECK_EQ(options.pass_input_stream().size(),
options.pass_output_stream().size());
MP_ASSIGN_OR_RETURN(
ASSIGN_OR_RETURN(
auto pass_streams,
GetPassMap(options, *cc->Inputs().TagMap(), *cc->Outputs().TagMap()));
std::set<CollectionItemId> pass_out;
@@ -121,9 +121,8 @@ class BypassCalculator : public Node {
// Saves the map of passthrough input and output stream ids.
absl::Status Open(CalculatorContext* cc) override {
auto options = cc->Options<BypassCalculatorOptions>();
MP_ASSIGN_OR_RETURN(
pass_streams_,
GetPassMap(options, *cc->Inputs().TagMap(), *cc->Outputs().TagMap()));
ASSIGN_OR_RETURN(pass_streams_, GetPassMap(options, *cc->Inputs().TagMap(),
*cc->Outputs().TagMap()));
return absl::OkStatus();
}
@@ -18,7 +18,6 @@
#include "mediapipe/calculators/core/concatenate_vector_calculator.pb.h"
#include "mediapipe/framework/api2/node.h"
#include "mediapipe/framework/calculator_framework.h"
#include "mediapipe/framework/formats/body_rig.pb.h"
#include "mediapipe/framework/formats/classification.pb.h"
#include "mediapipe/framework/formats/landmark.pb.h"
#include "mediapipe/framework/port/canonical_errors.h"
@@ -129,19 +128,6 @@ class ConcatenateClassificationListCalculator
};
MEDIAPIPE_REGISTER_NODE(ConcatenateClassificationListCalculator);
class ConcatenateJointListCalculator
: public ConcatenateListsCalculator<Joint, JointList> {
protected:
int ListSize(const JointList& list) const override {
return list.joint_size();
}
const Joint GetItem(const JointList& list, int idx) const override {
return list.joint(idx);
}
Joint* AddItem(JointList& list) const override { return list.add_joint(); }
};
MEDIAPIPE_REGISTER_NODE(ConcatenateJointListCalculator);
} // namespace api2
} // namespace mediapipe
@@ -14,8 +14,6 @@
#include "mediapipe/calculators/core/end_loop_calculator.h"
#include <array>
#include <utility>
#include <vector>
#include "mediapipe/framework/formats/classification.pb.h"
@@ -86,8 +84,4 @@ 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
@@ -55,16 +55,16 @@ class EndLoopCalculator : public CalculatorBase {
if (!input_stream_collection_) {
input_stream_collection_.reset(new IterableT);
}
if constexpr (std::is_copy_constructible_v<ItemT>) {
input_stream_collection_->push_back(
cc->Inputs().Tag("ITEM").Get<ItemT>());
// Try to consume the item and move it into the collection. If the items
// are not consumable, then try to copy them instead. If the items are
// not copyable, then an error will be returned.
auto item_ptr_or = cc->Inputs().Tag("ITEM").Value().Consume<ItemT>();
if (item_ptr_or.ok()) {
input_stream_collection_->push_back(std::move(*item_ptr_or.value()));
} else {
// Try to consume the item and move it into the collection. Return an
// error if the items are not consumable.
auto item_ptr_or = cc->Inputs().Tag("ITEM").Value().Consume<ItemT>();
if (item_ptr_or.ok()) {
input_stream_collection_->push_back(std::move(*item_ptr_or.value()));
if constexpr (std::is_copy_constructible_v<ItemT>) {
input_stream_collection_->push_back(
cc->Inputs().Tag("ITEM").template Get<ItemT>());
} else {
return absl::InternalError(
"The item type is not copiable. Consider making the "
@@ -12,7 +12,6 @@
// 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"
@@ -357,18 +356,18 @@ TEST_F(GateCalculatorTest, AllowWithStateChangeNoDataStreams) {
RunTimeStepWithoutDataStream(kTimestampValue2, "ALLOW", true);
constexpr int64_t kTimestampValue3 = 45;
RunTimeStepWithoutDataStream(kTimestampValue3, "ALLOW", false);
ABSL_LOG(INFO) << "a";
LOG(INFO) << "a";
const std::vector<Packet>& output =
runner()->Outputs().Get("STATE_CHANGE", 0).packets;
ABSL_LOG(INFO) << "s";
LOG(INFO) << "s";
ASSERT_EQ(2, output.size());
ABSL_LOG(INFO) << "d";
LOG(INFO) << "d";
EXPECT_EQ(kTimestampValue1, output[0].Timestamp().Value());
EXPECT_EQ(kTimestampValue3, output[1].Timestamp().Value());
ABSL_LOG(INFO) << "f";
LOG(INFO) << "f";
EXPECT_EQ(true, output[0].Get<bool>()); // Allow.
EXPECT_EQ(false, output[1].Get<bool>()); // Disallow.
ABSL_LOG(INFO) << "g";
LOG(INFO) << "g";
}
TEST_F(GateCalculatorTest, DisallowWithStateChange) {
@@ -12,7 +12,6 @@
// 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"
@@ -79,7 +78,7 @@ absl::Status ImmediateMuxCalculator::Process(CalculatorContext* cc) {
if (packet.Timestamp() >= cc->Outputs().Index(0).NextTimestampBound()) {
cc->Outputs().Index(0).AddPacket(packet);
} else {
ABSL_LOG_FIRST_N(WARNING, 5)
LOG_FIRST_N(WARNING, 5)
<< "Dropping a packet with timestamp " << packet.Timestamp();
}
if (cc->Outputs().NumEntries() >= 2) {
@@ -16,7 +16,6 @@
#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"
@@ -210,7 +209,7 @@ TEST(MatrixMultiplyCalculatorTest, Multiply) {
MatrixFromTextProto(kSamplesText, &samples);
Matrix expected;
MatrixFromTextProto(kExpectedText, &expected);
ABSL_CHECK_EQ(samples.cols(), expected.cols());
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
@@ -12,7 +12,6 @@
// 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"
@@ -54,7 +53,7 @@ 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) {
ABSL_LOG(WARNING)
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 "
@@ -73,8 +72,8 @@ class MergeCalculator : public Node {
}
}
ABSL_LOG(WARNING) << "Empty input packets at timestamp "
<< cc->InputTimestamp().Value();
LOG(WARNING) << "Empty input packets at timestamp "
<< cc->InputTimestamp().Value();
return absl::OkStatus();
}
@@ -16,9 +16,6 @@
#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) {
@@ -180,7 +177,7 @@ PacketResamplerCalculator::GetSamplingStrategy(
const PacketResamplerCalculatorOptions& options) {
if (options.reproducible_sampling()) {
if (!options.jitter_with_reflection()) {
ABSL_LOG(WARNING)
LOG(WARNING)
<< "reproducible_sampling enabled w/ jitter_with_reflection "
"disabled. "
<< "reproducible_sampling always uses jitter with reflection, "
@@ -203,15 +200,15 @@ PacketResamplerCalculator::GetSamplingStrategy(
Timestamp PacketResamplerCalculator::PeriodIndexToTimestamp(
int64_t index) const {
ABSL_CHECK_EQ(jitter_, 0.0);
ABSL_CHECK_NE(first_timestamp_, Timestamp::Unset());
CHECK_EQ(jitter_, 0.0);
CHECK_NE(first_timestamp_, Timestamp::Unset());
return first_timestamp_ + TimestampDiffFromSeconds(index / frame_rate_);
}
int64_t PacketResamplerCalculator::TimestampToPeriodIndex(
Timestamp timestamp) const {
ABSL_CHECK_EQ(jitter_, 0.0);
ABSL_CHECK_NE(first_timestamp_, Timestamp::Unset());
CHECK_EQ(jitter_, 0.0);
CHECK_NE(first_timestamp_, Timestamp::Unset());
return MathUtil::SafeRound<int64_t, double>(
(timestamp - first_timestamp_).Seconds() * frame_rate_);
}
@@ -232,15 +229,13 @@ absl::Status LegacyJitterWithReflectionStrategy::Open(CalculatorContext* cc) {
if (resampler_options.output_header() !=
PacketResamplerCalculatorOptions::NONE) {
ABSL_LOG(WARNING)
<< "VideoHeader::frame_rate holds the target value and not "
"the actual value.";
LOG(WARNING) << "VideoHeader::frame_rate holds the target value and not "
"the actual value.";
}
if (calculator_->flush_last_packet_) {
ABSL_LOG(WARNING)
<< "PacketResamplerCalculatorOptions.flush_last_packet is "
"ignored, because we are adding jitter.";
LOG(WARNING) << "PacketResamplerCalculatorOptions.flush_last_packet is "
"ignored, because we are adding jitter.";
}
const auto& seed = cc->InputSidePackets().Tag(kSeedTag).Get<std::string>();
@@ -259,7 +254,7 @@ absl::Status LegacyJitterWithReflectionStrategy::Open(CalculatorContext* cc) {
}
absl::Status LegacyJitterWithReflectionStrategy::Close(CalculatorContext* cc) {
if (!packet_reservoir_->IsEmpty()) {
ABSL_LOG(INFO) << "Emitting pack from reservoir.";
LOG(INFO) << "Emitting pack from reservoir.";
calculator_->OutputWithinLimits(cc, packet_reservoir_->GetSample());
}
return absl::OkStatus();
@@ -290,7 +285,7 @@ absl::Status LegacyJitterWithReflectionStrategy::Process(
if (calculator_->frame_time_usec_ <
(cc->InputTimestamp() - calculator_->last_packet_.Timestamp()).Value()) {
ABSL_LOG_FIRST_N(WARNING, 2)
LOG_FIRST_N(WARNING, 2)
<< "Adding jitter is not very useful when upsampling.";
}
@@ -345,8 +340,8 @@ void LegacyJitterWithReflectionStrategy::UpdateNextOutputTimestampWithJitter() {
next_output_timestamp_ = Timestamp(ReflectBetween(
next_output_timestamp_.Value(), next_output_timestamp_min_.Value(),
next_output_timestamp_max_.Value()));
ABSL_CHECK_GE(next_output_timestamp_, next_output_timestamp_min_);
ABSL_CHECK_LT(next_output_timestamp_, next_output_timestamp_max_);
CHECK_GE(next_output_timestamp_, next_output_timestamp_min_);
CHECK_LT(next_output_timestamp_, next_output_timestamp_max_);
}
absl::Status ReproducibleJitterWithReflectionStrategy::Open(
@@ -357,15 +352,13 @@ absl::Status ReproducibleJitterWithReflectionStrategy::Open(
if (resampler_options.output_header() !=
PacketResamplerCalculatorOptions::NONE) {
ABSL_LOG(WARNING)
<< "VideoHeader::frame_rate holds the target value and not "
"the actual value.";
LOG(WARNING) << "VideoHeader::frame_rate holds the target value and not "
"the actual value.";
}
if (calculator_->flush_last_packet_) {
ABSL_LOG(WARNING)
<< "PacketResamplerCalculatorOptions.flush_last_packet is "
"ignored, because we are adding jitter.";
LOG(WARNING) << "PacketResamplerCalculatorOptions.flush_last_packet is "
"ignored, because we are adding jitter.";
}
const auto& seed = cc->InputSidePackets().Tag(kSeedTag).Get<std::string>();
@@ -418,7 +411,7 @@ 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.
ABSL_LOG_FIRST_N(WARNING, 2)
LOG_FIRST_N(WARNING, 2)
<< "Adding jitter is not very useful when upsampling.";
}
@@ -506,15 +499,13 @@ absl::Status JitterWithoutReflectionStrategy::Open(CalculatorContext* cc) {
if (resampler_options.output_header() !=
PacketResamplerCalculatorOptions::NONE) {
ABSL_LOG(WARNING)
<< "VideoHeader::frame_rate holds the target value and not "
"the actual value.";
LOG(WARNING) << "VideoHeader::frame_rate holds the target value and not "
"the actual value.";
}
if (calculator_->flush_last_packet_) {
ABSL_LOG(WARNING)
<< "PacketResamplerCalculatorOptions.flush_last_packet is "
"ignored, because we are adding jitter.";
LOG(WARNING) << "PacketResamplerCalculatorOptions.flush_last_packet is "
"ignored, because we are adding jitter.";
}
const auto& seed = cc->InputSidePackets().Tag(kSeedTag).Get<std::string>();
@@ -564,7 +555,7 @@ absl::Status JitterWithoutReflectionStrategy::Process(CalculatorContext* cc) {
if (calculator_->frame_time_usec_ <
(cc->InputTimestamp() - calculator_->last_packet_.Timestamp()).Value()) {
ABSL_LOG_FIRST_N(WARNING, 2)
LOG_FIRST_N(WARNING, 2)
<< "Adding jitter is not very useful when upsampling.";
}
@@ -13,6 +13,7 @@
#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"
@@ -17,7 +17,6 @@
#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"
@@ -161,8 +160,8 @@ absl::Status PacketThinnerCalculator::Open(CalculatorContext* cc) {
thinner_type_ = options.thinner_type();
// This check enables us to assume only two thinner types exist in Process()
ABSL_CHECK(thinner_type_ == PacketThinnerCalculatorOptions::ASYNC ||
thinner_type_ == PacketThinnerCalculatorOptions::SYNC)
CHECK(thinner_type_ == PacketThinnerCalculatorOptions::ASYNC ||
thinner_type_ == PacketThinnerCalculatorOptions::SYNC)
<< "Unsupported thinner type.";
if (thinner_type_ == PacketThinnerCalculatorOptions::ASYNC) {
@@ -178,8 +177,7 @@ absl::Status PacketThinnerCalculator::Open(CalculatorContext* cc) {
} else {
period_ = TimestampDiff(options.period());
}
ABSL_CHECK_LT(TimestampDiff(0), period_)
<< "Specified period must be positive.";
CHECK_LT(TimestampDiff(0), period_) << "Specified period must be positive.";
if (options.has_start_time()) {
start_time_ = Timestamp(options.start_time());
@@ -191,7 +189,7 @@ absl::Status PacketThinnerCalculator::Open(CalculatorContext* cc) {
end_time_ =
options.has_end_time() ? Timestamp(options.end_time()) : Timestamp::Max();
ABSL_CHECK_LT(start_time_, end_time_)
CHECK_LT(start_time_, end_time_)
<< "Invalid PacketThinner: start_time must be earlier than end_time";
sync_output_timestamps_ = options.sync_output_timestamps();
@@ -234,7 +232,7 @@ absl::Status PacketThinnerCalculator::Close(CalculatorContext* cc) {
// Emit any saved packets before quitting.
if (!saved_packet_.IsEmpty()) {
// Only sync thinner should have saved packets.
ABSL_CHECK_EQ(PacketThinnerCalculatorOptions::SYNC, thinner_type_);
CHECK_EQ(PacketThinnerCalculatorOptions::SYNC, thinner_type_);
if (sync_output_timestamps_) {
cc->Outputs().Index(0).AddPacket(
saved_packet_.At(NearestSyncTimestamp(saved_packet_.Timestamp())));
@@ -271,7 +269,7 @@ absl::Status PacketThinnerCalculator::SyncThinnerProcess(
const Timestamp saved_sync = NearestSyncTimestamp(saved);
const Timestamp now = cc->InputTimestamp();
const Timestamp now_sync = NearestSyncTimestamp(now);
ABSL_CHECK_LE(saved_sync, now_sync);
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
@@ -297,7 +295,7 @@ absl::Status PacketThinnerCalculator::SyncThinnerProcess(
}
Timestamp PacketThinnerCalculator::NearestSyncTimestamp(Timestamp now) const {
ABSL_CHECK_NE(start_time_, Timestamp::Unset())
CHECK_NE(start_time_, Timestamp::Unset())
<< "Method only valid for sync thinner calculator.";
// Computation is done using int64 arithmetic. No easy way to avoid
@@ -305,12 +303,12 @@ 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();
ABSL_CHECK_LE(0, period64);
CHECK_LE(0, period64);
// Round now64 to its closest interval (units of period64).
int64_t sync64 =
(now64 - start64 + period64 / 2) / period64 * period64 + start64;
ABSL_CHECK_LE(abs(now64 - sync64), period64 / 2)
CHECK_LE(abs(now64 - sync64), period64 / 2)
<< "start64: " << start64 << "; now64: " << now64
<< "; sync64: " << sync64;
@@ -16,7 +16,6 @@
#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"
@@ -71,7 +70,7 @@ class SimpleRunner : public CalculatorRunner {
}
double GetFrameRate() const {
ABSL_CHECK(!Outputs().Index(0).header.IsEmpty());
CHECK(!Outputs().Index(0).header.IsEmpty());
return Outputs().Index(0).header.Get<VideoHeader>().frame_rate;
}
};
@@ -14,7 +14,6 @@
#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"
@@ -102,7 +101,7 @@ 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_) {
ABSL_LOG(FATAL) << "Not supported yet";
LOG(FATAL) << "Not supported yet";
}
// Store current packet for later output.
packet_cache_.push_back(kIn(cc).packet());
@@ -17,7 +17,6 @@
#include "mediapipe/calculators/core/split_vector_calculator.pb.h"
#include "mediapipe/framework/calculator_framework.h"
#include "mediapipe/framework/formats/body_rig.pb.h"
#include "mediapipe/framework/formats/landmark.pb.h"
#include "mediapipe/framework/port/canonical_errors.h"
#include "mediapipe/framework/port/ret_check.h"
@@ -197,18 +196,6 @@ class SplitLandmarkListCalculator
};
REGISTER_CALCULATOR(SplitLandmarkListCalculator);
class SplitJointListCalculator : public SplitListsCalculator<Joint, JointList> {
protected:
int ListSize(const JointList& list) const override {
return list.joint_size();
}
const Joint GetItem(const JointList& list, int idx) const override {
return list.joint(idx);
}
Joint* AddItem(JointList& list) const override { return list.add_joint(); }
};
REGISTER_CALCULATOR(SplitJointListCalculator);
} // namespace mediapipe
// NOLINTNEXTLINE
@@ -1,90 +0,0 @@
#include "mediapipe/framework/calculator_framework.h"
#include "mediapipe/framework/port/status.h"
namespace mediapipe {
namespace {
constexpr char kInputValueTag[] = "IN";
constexpr char kTickerTag[] = "TICK";
constexpr char kOutputTag[] = "OUT";
constexpr char kIndicationTag[] = "FLAG";
} // namespace
// For every packet received on the TICK stream, if the IN stream is not
// empty - emit its value as is as OUT. Otherwise output a default packet.
// FLAG outputs true every time the default value has been used. It does not
// output anything when IN has a value.
//
// Example config:
// node {
// calculator: "ValueOrDefaultCalculator"
// input_stream: "IN:sometimes_missing_value"
// input_stream: "TICK:clock"
// output_stream: "OUT:value_or_default"
// output_stream: "FLAG:used_default"
// input_side_packet: "default"
// }
//
// TODO: Consider adding an option for a default value as a input-stream
// instead of a side-packet, so it will enable using standard calculators
// instead of creating a new packet-generators. It will also allow a dynamic
// default value.
class ValueOrDefaultCalculator : public mediapipe::CalculatorBase {
public:
ValueOrDefaultCalculator() {}
ValueOrDefaultCalculator(const ValueOrDefaultCalculator&) = delete;
ValueOrDefaultCalculator& operator=(const ValueOrDefaultCalculator&) = delete;
static mediapipe::Status GetContract(mediapipe::CalculatorContract* cc) {
cc->Inputs().Tag(kInputValueTag).SetAny();
cc->Inputs().Tag(kTickerTag).SetAny();
cc->Outputs().Tag(kOutputTag).SetSameAs(&cc->Inputs().Tag(kInputValueTag));
cc->Outputs().Tag(kIndicationTag).Set<bool>();
cc->InputSidePackets().Index(0).SetSameAs(
&cc->Inputs().Tag(kInputValueTag));
return mediapipe::OkStatus();
}
mediapipe::Status Open(mediapipe::CalculatorContext* cc) override {
if (!cc->Inputs().Tag(kInputValueTag).Header().IsEmpty()) {
cc->Outputs()
.Tag(kOutputTag)
.SetHeader(cc->Inputs().Tag(kInputValueTag).Header());
}
default_ = cc->InputSidePackets().Index(0);
cc->SetOffset(mediapipe::TimestampDiff(0));
return mediapipe::OkStatus();
}
mediapipe::Status Process(mediapipe::CalculatorContext* cc) override {
// Output according to the TICK signal.
if (cc->Inputs().Tag(kTickerTag).IsEmpty()) {
return mediapipe::OkStatus();
}
if (!cc->Inputs().Tag(kInputValueTag).IsEmpty()) {
// Output the input as is:
cc->Outputs()
.Tag(kOutputTag)
.AddPacket(cc->Inputs().Tag(kInputValueTag).Value());
} else {
// Output default:
cc->Outputs()
.Tag(kOutputTag)
.AddPacket(default_.At(cc->InputTimestamp()));
cc->Outputs()
.Tag(kIndicationTag)
.Add(new bool(true), cc->InputTimestamp());
}
return mediapipe::OkStatus();
}
private:
// The default value to replicate every time there is no new value.
mediapipe::Packet default_;
};
REGISTER_CALCULATOR(ValueOrDefaultCalculator);
} // namespace mediapipe
@@ -1,240 +0,0 @@
#include <algorithm>
#include <cstdint>
#include <vector>
#include "mediapipe/framework/calculator_framework.h"
#include "mediapipe/framework/calculator_runner.h"
#include "mediapipe/framework/packet.h"
#include "mediapipe/framework/port/gmock.h"
#include "mediapipe/framework/port/gtest.h"
#include "mediapipe/framework/port/status_matchers.h"
namespace mediapipe {
namespace {
using ::testing::AllOf;
using ::testing::ContainerEq;
using ::testing::Each;
using ::testing::ElementsAre;
using ::testing::IsEmpty;
using ::testing::SizeIs;
using ::testing::Test;
const int kDefaultValue = 0;
// Utility to a create a mediapipe graph runner with the tested calculator and a
// default value, for all the tests.
class ValueOrDefaultRunner : public mediapipe::CalculatorRunner {
public:
ValueOrDefaultRunner()
: mediapipe::CalculatorRunner(R"pb(
calculator: "ValueOrDefaultCalculator"
input_stream: "IN:in"
input_stream: "TICK:tick"
input_side_packet: "default"
output_stream: "OUT:out"
output_stream: "FLAG:used_default"
)pb") {
MutableSidePackets()->Index(0) = mediapipe::MakePacket<int>(kDefaultValue);
}
// Utility to push inputs to the runner to the TICK stream, so we could easily
// tick.
void TickAt(int64_t time) {
// The type or value of the stream isn't relevant, we use just a bool.
MutableInputs()->Tag("TICK").packets.push_back(
mediapipe::Adopt(new bool(false)).At(mediapipe::Timestamp(time)));
}
// Utility to push the real inputs to the runner (IN stream).
void ProvideInput(int64_t time, int value) {
MutableInputs()->Tag("IN").packets.push_back(
mediapipe::Adopt(new int(value)).At(mediapipe::Timestamp(time)));
}
// Extracts the timestamps (as int64) of the output stream of the calculator.
std::vector<int64_t> GetOutputTimestamps() const {
std::vector<int64_t> timestamps;
for (const mediapipe::Packet& packet : Outputs().Tag("OUT").packets) {
timestamps.emplace_back(packet.Timestamp().Value());
}
return timestamps;
}
// Extracts the values from the output stream of the calculator.
std::vector<int> GetOutputValues() const {
std::vector<int> values;
for (const mediapipe::Packet& packet : Outputs().Tag("OUT").packets) {
values.emplace_back(packet.Get<int>());
}
return values;
}
// Extracts the timestamps (as int64) of the flag stream, which indicates on
// times without an input value (i.e. using the default value).
std::vector<int64_t> GetFlagTimestamps() const {
std::vector<int64_t> timestamps;
for (const mediapipe::Packet& packet : Outputs().Tag("FLAG").packets) {
timestamps.emplace_back(packet.Timestamp().Value());
}
return timestamps;
}
// Extracts the output from the flags stream (which should always be true).
std::vector<bool> GetFlagValues() const {
std::vector<bool> flags;
for (const mediapipe::Packet& packet : Outputs().Tag("FLAG").packets) {
flags.emplace_back(packet.Get<bool>());
}
return flags;
}
};
// To be used as input values:
std::vector<int> GetIntegersRange(int size) {
std::vector<int> result;
for (int i = 0; i < size; ++i) {
// We start with default-value+1 so it won't contain the default value.
result.push_back(kDefaultValue + 1 + i);
}
return result;
}
TEST(ValueOrDefaultCalculatorTest, NoInputs) {
// Check that when no real inputs are provided - we get the default value over
// and over, with the correct timestamps.
ValueOrDefaultRunner runner;
const std::vector<int64_t> ticks = {0, 1, 2, 5, 8, 12, 33, 231};
for (int tick : ticks) {
runner.TickAt(tick);
}
MP_EXPECT_OK(runner.Run());
// Make sure we get the right timestamps:
EXPECT_THAT(runner.GetOutputTimestamps(), ContainerEq(ticks));
// All should be default value:
EXPECT_THAT(runner.GetOutputValues(),
AllOf(Each(kDefaultValue), SizeIs(ticks.size())));
// We should get the default indication all the time:
EXPECT_THAT(runner.GetFlagTimestamps(), ContainerEq(ticks));
}
TEST(ValueOrDefaultCalculatorTest, NeverDefault) {
// Check that when we provide the inputs on time - we get them as outputs.
ValueOrDefaultRunner runner;
const std::vector<int64_t> ticks = {0, 1, 2, 5, 8, 12, 33, 231};
const std::vector<int> values = GetIntegersRange(ticks.size());
for (int i = 0; i < ticks.size(); ++i) {
runner.TickAt(ticks[i]);
runner.ProvideInput(ticks[i], values[i]);
}
MP_EXPECT_OK(runner.Run());
// Make sure we get the right timestamps:
EXPECT_THAT(runner.GetOutputTimestamps(), ContainerEq(ticks));
// Should get the inputs values:
EXPECT_THAT(runner.GetOutputValues(), ContainerEq(values));
// We should never get the default indication:
EXPECT_THAT(runner.GetFlagTimestamps(), IsEmpty());
}
TEST(ValueOrDefaultCalculatorTest, DefaultAndValues) {
// Check that when we provide inputs only part of the time - we get them, but
// defaults at the missing times.
// That's the usual use case for this calculator.
ValueOrDefaultRunner runner;
const std::vector<int64_t> ticks = {0, 1, 5, 8, 12, 231};
// Provide inputs only part of the ticks.
// Chosen so there will be defaults before the first input, between the
// inputs and after the last input.
const std::vector<int64_t> in_ticks = {/*0,*/ 1, 5, /*8,*/ 12, /*, 231*/};
const std::vector<int> in_values = GetIntegersRange(in_ticks.size());
for (int tick : ticks) {
runner.TickAt(tick);
}
for (int i = 0; i < in_ticks.size(); ++i) {
runner.ProvideInput(in_ticks[i], in_values[i]);
}
MP_EXPECT_OK(runner.Run());
// Make sure we get all the timestamps:
EXPECT_THAT(runner.GetOutputTimestamps(), ContainerEq(ticks));
// The timestamps of the flag should be exactly the ones not in in_ticks.
EXPECT_THAT(runner.GetFlagTimestamps(), ElementsAre(0, 8, 231));
// And the values are default in these times, and the input values for
// in_ticks.
EXPECT_THAT(
runner.GetOutputValues(),
ElementsAre(kDefaultValue, 1, 2, kDefaultValue, 3, kDefaultValue));
}
TEST(ValueOrDefaultCalculatorTest, TimestampsMissmatch) {
// Check that when we provide the inputs not on time - we don't get them.
ValueOrDefaultRunner runner;
const std::vector<int64_t> ticks = {1, 2, 5, 8, 12, 33, 231};
// The timestamps chosen so it will be before the first tick, in between ticks
// and after the last one. Also - more inputs than ticks.
const std::vector<int64_t> in_ticks = {0, 3, 4, 6, 7, 9, 10,
11, 13, 14, 15, 16, 232};
const std::vector<int> in_values = GetIntegersRange(in_ticks.size());
for (int tick : ticks) {
runner.TickAt(tick);
}
for (int i = 0; i < in_ticks.size(); ++i) {
runner.ProvideInput(in_ticks[i], in_values[i]);
}
MP_EXPECT_OK(runner.Run());
// Non of the in_ticks should be inserted:
EXPECT_THAT(runner.GetOutputTimestamps(), ContainerEq(ticks));
EXPECT_THAT(runner.GetOutputValues(),
AllOf(Each(kDefaultValue), SizeIs(ticks.size())));
// All (and only) ticks should get the default.
EXPECT_THAT(runner.GetFlagTimestamps(), ContainerEq(ticks));
}
TEST(ValueOrDefaultCalculatorTest, FlagValue) {
// Since we anyway suppose that the Flag is a bool - there is nothing
// interesting to check, but we should check once that the value is the right
// (true) one.
ValueOrDefaultRunner runner;
runner.TickAt(0);
MP_EXPECT_OK(runner.Run());
EXPECT_THAT(runner.GetFlagValues(), ElementsAre(true));
}
TEST(ValueOrDefaultCalculatorTest, FullTest) {
// Make sure that nothing gets wrong with an input that have both right and
// wrong timestamps, some defaults etc.
ValueOrDefaultRunner runner;
const std::vector<int64_t> ticks = {1, 2, 5, 8, 12, 33, 231};
const std::vector<int64_t> in_ticks = {0, 2, 4, 6, 8, 9, 12, 33, 54, 232};
const std::vector<int> in_values = GetIntegersRange(in_ticks.size());
for (int tick : ticks) {
runner.TickAt(tick);
}
for (int i = 0; i < in_ticks.size(); ++i) {
runner.ProvideInput(in_ticks[i], in_values[i]);
}
MP_EXPECT_OK(runner.Run());
EXPECT_THAT(runner.GetOutputTimestamps(), ContainerEq(ticks));
// Calculated by hand:
EXPECT_THAT(
runner.GetOutputValues(),
ElementsAre(kDefaultValue, 2, kDefaultValue, 5, 7, 8, kDefaultValue));
EXPECT_THAT(runner.GetFlagTimestamps(), ElementsAre(1, 5, 231));
EXPECT_THAT(runner.GetFlagValues(), AllOf(Each(true), SizeIs(3)));
}
} // namespace
} // namespace mediapipe
+2 -44
View File
@@ -97,7 +97,6 @@ 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,
)
@@ -126,7 +125,6 @@ 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,
)
@@ -153,11 +151,11 @@ 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": [
@@ -204,7 +202,6 @@ 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": [],
@@ -264,12 +261,9 @@ 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",
@@ -279,36 +273,6 @@ 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"],
@@ -336,7 +300,6 @@ 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": [
@@ -433,7 +396,6 @@ 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",
],
)
@@ -458,8 +420,6 @@ 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",
],
@@ -665,9 +625,9 @@ 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": [
@@ -705,7 +665,6 @@ 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",
],
)
@@ -728,7 +687,6 @@ 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",
@@ -20,7 +20,6 @@
#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"
@@ -54,10 +53,6 @@ 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;
}
}
@@ -223,7 +218,7 @@ class GlTextureWarpAffineRunner
absl::StrCat(mediapipe::kMediaPipeFragmentShaderPreamble,
interpolation_def, kFragShader);
MP_ASSIGN_OR_RETURN(program_, create_fn(vert_src, frag_src));
ASSIGN_OR_RETURN(program_, create_fn(vert_src, frag_src));
auto create_custom_zero_fn = [&]() -> absl::StatusOr<Program> {
std::string custom_zero_border_mode_def = R"(
@@ -236,10 +231,10 @@ class GlTextureWarpAffineRunner
};
#if GL_CLAMP_TO_BORDER_MAY_BE_SUPPORTED
if (!IsGlClampToBorderSupported(gl_helper_->GetGlContext())) {
MP_ASSIGN_OR_RETURN(program_custom_zero_, create_custom_zero_fn());
ASSIGN_OR_RETURN(program_custom_zero_, create_custom_zero_fn());
}
#else
MP_ASSIGN_OR_RETURN(program_custom_zero_, create_custom_zero_fn());
ASSIGN_OR_RETURN(program_custom_zero_, create_custom_zero_fn());
#endif // GL_CLAMP_TO_BORDER_MAY_BE_SUPPORTED
glGenFramebuffers(1, &framebuffer_);
@@ -389,8 +384,6 @@ class GlTextureWarpAffineRunner
glActiveTexture(GL_TEXTURE0);
glBindTexture(GL_TEXTURE_2D, 0);
glFlush();
return absl::OkStatus();
}
@@ -15,7 +15,6 @@
#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"
@@ -184,8 +183,8 @@ absl::Status BilateralFilterCalculator::Open(CalculatorContext* cc) {
sigma_color_ = options_.sigma_color();
sigma_space_ = options_.sigma_space();
ABSL_CHECK_GE(sigma_color_, 0.0);
ABSL_CHECK_GE(sigma_space_, 0.0);
CHECK_GE(sigma_color_, 0.0);
CHECK_GE(sigma_space_, 0.0);
if (!use_gpu_) sigma_color_ *= 255.0;
if (use_gpu_) {
@@ -12,7 +12,6 @@
// 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"
@@ -26,8 +25,8 @@
namespace mediapipe {
namespace {
void SetColorChannel(int channel, uint8 value, cv::Mat* mat) {
ABSL_CHECK(mat->depth() == CV_8U);
ABSL_CHECK(channel < mat->channels());
CHECK(mat->depth() == CV_8U);
CHECK(channel < mat->channels());
const int step = mat->channels();
for (int r = 0; r < mat->rows; ++r) {
uint8* row_ptr = mat->ptr<uint8>(r);
@@ -16,7 +16,6 @@
#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"
@@ -203,9 +202,8 @@ absl::Status ImageCroppingCalculator::ValidateBorderModeForGPU(
switch (options.border_mode()) {
case mediapipe::ImageCroppingCalculatorOptions::BORDER_ZERO:
ABSL_LOG(WARNING)
<< "BORDER_ZERO mode is not supported by GPU "
<< "implementation and will fall back into BORDER_REPLICATE";
LOG(WARNING) << "BORDER_ZERO mode is not supported by GPU "
<< "implementation and will fall back into BORDER_REPLICATE";
break;
case mediapipe::ImageCroppingCalculatorOptions::BORDER_REPLICATE:
break;
@@ -92,11 +92,11 @@ absl::StatusOr<ImageFileProperties> GetImageFileProperites(
properties.set_focal_length_mm(result.FocalLength);
properties.set_focal_length_35mm(result.FocalLengthIn35mm);
MP_ASSIGN_OR_RETURN(auto focal_length_pixels,
ComputeFocalLengthInPixels(properties.image_width(),
properties.image_height(),
properties.focal_length_35mm(),
properties.focal_length_mm()));
ASSIGN_OR_RETURN(auto focal_length_pixels,
ComputeFocalLengthInPixels(properties.image_width(),
properties.image_height(),
properties.focal_length_35mm(),
properties.focal_length_mm()));
properties.set_focal_length_pixels(focal_length_pixels);
return properties;
@@ -151,7 +151,7 @@ class ImageFilePropertiesCalculator : public CalculatorBase {
if (cc->InputSidePackets().NumEntries() == 1) {
const std::string& image_bytes =
cc->InputSidePackets().Index(0).Get<std::string>();
MP_ASSIGN_OR_RETURN(properties_, GetImageFileProperites(image_bytes));
ASSIGN_OR_RETURN(properties_, GetImageFileProperites(image_bytes));
read_properties_ = true;
}
@@ -169,7 +169,7 @@ class ImageFilePropertiesCalculator : public CalculatorBase {
return absl::OkStatus();
}
const std::string& image_bytes = cc->Inputs().Index(0).Get<std::string>();
MP_ASSIGN_OR_RETURN(properties_, GetImageFileProperites(image_bytes));
ASSIGN_OR_RETURN(properties_, GetImageFileProperites(image_bytes));
read_properties_ = true;
}
if (read_properties_) {
@@ -12,7 +12,6 @@
// 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"
@@ -28,7 +27,6 @@
#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"
@@ -62,42 +60,42 @@ constexpr char kVideoPrestreamTag[] = "VIDEO_PRESTREAM";
int RotationModeToDegrees(mediapipe::RotationMode_Mode rotation) {
switch (rotation) {
case mediapipe::RotationMode::UNKNOWN:
case mediapipe::RotationMode::ROTATION_0:
case mediapipe::RotationMode_Mode_UNKNOWN:
case mediapipe::RotationMode_Mode_ROTATION_0:
return 0;
case mediapipe::RotationMode::ROTATION_90:
case mediapipe::RotationMode_Mode_ROTATION_90:
return 90;
case mediapipe::RotationMode::ROTATION_180:
case mediapipe::RotationMode_Mode_ROTATION_180:
return 180;
case mediapipe::RotationMode::ROTATION_270:
case mediapipe::RotationMode_Mode_ROTATION_270:
return 270;
}
}
mediapipe::RotationMode_Mode DegreesToRotationMode(int degrees) {
switch (degrees) {
case 0:
return mediapipe::RotationMode::ROTATION_0;
return mediapipe::RotationMode_Mode_ROTATION_0;
case 90:
return mediapipe::RotationMode::ROTATION_90;
return mediapipe::RotationMode_Mode_ROTATION_90;
case 180:
return mediapipe::RotationMode::ROTATION_180;
return mediapipe::RotationMode_Mode_ROTATION_180;
case 270:
return mediapipe::RotationMode::ROTATION_270;
return mediapipe::RotationMode_Mode_ROTATION_270;
default:
return mediapipe::RotationMode::UNKNOWN;
return mediapipe::RotationMode_Mode_UNKNOWN;
}
}
mediapipe::ScaleMode_Mode ParseScaleMode(
mediapipe::ScaleMode_Mode scale_mode,
mediapipe::ScaleMode_Mode default_mode) {
switch (scale_mode) {
case mediapipe::ScaleMode::DEFAULT:
case mediapipe::ScaleMode_Mode_DEFAULT:
return default_mode;
case mediapipe::ScaleMode::STRETCH:
case mediapipe::ScaleMode_Mode_STRETCH:
return scale_mode;
case mediapipe::ScaleMode::FIT:
case mediapipe::ScaleMode_Mode_FIT:
return scale_mode;
case mediapipe::ScaleMode::FILL_AND_CROP:
case mediapipe::ScaleMode_Mode_FILL_AND_CROP:
return scale_mode;
default:
return default_mode;
@@ -210,8 +208,6 @@ 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_;
@@ -347,11 +343,6 @@ 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.
@@ -466,48 +457,26 @@ 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::STRETCH) {
if (interpolation_mode_ == ImageTransformationCalculatorOptions::LINEAR) {
// Use INTER_AREA for downscaling if interpolation mode is set to
// 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;
}
if (scale_mode_ == mediapipe::ScaleMode_Mode_STRETCH) {
int scale_flag =
input_mat.cols > output_width_ && input_mat.rows > output_height_
? cv::INTER_AREA
: cv::INTER_LINEAR;
cv::resize(input_mat, scaled_mat, cv::Size(output_width_, output_height_),
0, 0, opencv_interpolation_mode);
0, 0, scale_flag);
} 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);
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) {
int scale_flag = scale < 1.0f ? cv::INTER_AREA : cv::INTER_LINEAR;
if (scale_mode_ == mediapipe::ScaleMode_Mode_FIT) {
cv::Mat intermediate_mat;
cv::resize(input_mat, intermediate_mat,
cv::Size(target_width, target_height), 0, 0,
opencv_interpolation_mode);
cv::Size(target_width, target_height), 0, 0, scale_flag);
const int top = (output_height_ - target_height) / 2;
const int bottom = output_height_ - target_height - top;
const int left = (output_width_ - target_width) / 2;
@@ -519,7 +488,7 @@ absl::Status ImageTransformationCalculator::RenderCpu(CalculatorContext* cc) {
padding_color_);
} else {
cv::resize(input_mat, scaled_mat, cv::Size(target_width, target_height),
0, 0, opencv_interpolation_mode);
0, 0, scale_flag);
output_width = target_width;
output_height = target_height;
}
@@ -545,17 +514,17 @@ absl::Status ImageTransformationCalculator::RenderCpu(CalculatorContext* cc) {
cv::warpAffine(input_mat, rotated_mat, rotation_mat, rotated_size);
} else {
switch (rotation_) {
case mediapipe::RotationMode::UNKNOWN:
case mediapipe::RotationMode::ROTATION_0:
case mediapipe::RotationMode_Mode_UNKNOWN:
case mediapipe::RotationMode_Mode_ROTATION_0:
rotated_mat = input_mat;
break;
case mediapipe::RotationMode::ROTATION_90:
case mediapipe::RotationMode_Mode_ROTATION_90:
cv::rotate(input_mat, rotated_mat, cv::ROTATE_90_COUNTERCLOCKWISE);
break;
case mediapipe::RotationMode::ROTATION_180:
case mediapipe::RotationMode_Mode_ROTATION_180:
cv::rotate(input_mat, rotated_mat, cv::ROTATE_180);
break;
case mediapipe::RotationMode::ROTATION_270:
case mediapipe::RotationMode_Mode_ROTATION_270:
cv::rotate(input_mat, rotated_mat, cv::ROTATE_90_CLOCKWISE);
break;
}
@@ -592,7 +561,7 @@ absl::Status ImageTransformationCalculator::RenderGpu(CalculatorContext* cc) {
ComputeOutputDimensions(input_width, input_height, &output_width,
&output_height);
if (scale_mode_ == mediapipe::ScaleMode::FILL_AND_CROP) {
if (scale_mode_ == mediapipe::ScaleMode_Mode_FILL_AND_CROP) {
const float scale =
std::min(static_cast<float>(output_width_) / input_width,
static_cast<float>(output_height_) / input_height);
@@ -659,12 +628,6 @@ 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_,
@@ -689,8 +652,8 @@ void ImageTransformationCalculator::ComputeOutputDimensions(
if (output_width_ > 0 && output_height_ > 0) {
*output_width = output_width_;
*output_height = output_height_;
} else if (rotation_ == mediapipe::RotationMode::ROTATION_90 ||
rotation_ == mediapipe::RotationMode::ROTATION_270) {
} else if (rotation_ == mediapipe::RotationMode_Mode_ROTATION_90 ||
rotation_ == mediapipe::RotationMode_Mode_ROTATION_270) {
*output_width = input_height;
*output_height = input_width;
} else {
@@ -703,9 +666,9 @@ 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::FIT) {
if (rotation_ == mediapipe::RotationMode::ROTATION_90 ||
rotation_ == mediapipe::RotationMode::ROTATION_270) {
if (scale_mode_ == mediapipe::ScaleMode_Mode_FIT) {
if (rotation_ == mediapipe::RotationMode_Mode_ROTATION_90 ||
rotation_ == mediapipe::RotationMode_Mode_ROTATION_270) {
std::swap(input_width, input_height);
}
const float input_aspect_ratio =
@@ -54,15 +54,4 @@ 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;
}
@@ -1,315 +0,0 @@
#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
@@ -12,7 +12,6 @@
// 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"
@@ -62,7 +61,7 @@ absl::Status OpenCvImageEncoderCalculator::Open(CalculatorContext* cc) {
absl::Status OpenCvImageEncoderCalculator::Process(CalculatorContext* cc) {
const ImageFrame& image_frame = cc->Inputs().Index(0).Get<ImageFrame>();
ABSL_CHECK_EQ(1, image_frame.ByteDepth());
CHECK_EQ(1, image_frame.ByteDepth());
std::unique_ptr<OpenCvImageEncoderCalculatorResults> encoded_result =
absl::make_unique<OpenCvImageEncoderCalculatorResults>();
@@ -18,8 +18,6 @@
#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"
@@ -295,7 +293,7 @@ absl::Status ScaleImageCalculator::InitializeFrameInfo(CalculatorContext* cc) {
header->width = output_width_;
header->height = output_height_;
header->format = output_format_;
ABSL_LOG(INFO) << "OUTPUTTING HEADER on stream";
LOG(INFO) << "OUTPUTTING HEADER on stream";
cc->Outputs()
.Tag("VIDEO_HEADER")
.Add(header.release(), Timestamp::PreStream());
@@ -395,11 +393,10 @@ absl::Status ScaleImageCalculator::Open(CalculatorContext* cc) {
.SetHeader(Adopt(output_header.release()));
has_header_ = true;
} else {
ABSL_LOG(WARNING)
<< "Stream had a VideoHeader which didn't have sufficient "
"information. "
"Dropping VideoHeader and trying to deduce needed "
"information.";
LOG(WARNING) << "Stream had a VideoHeader which didn't have sufficient "
"information. "
"Dropping VideoHeader and trying to deduce needed "
"information.";
input_width_ = 0;
input_height_ = 0;
if (!options_.has_input_format()) {
@@ -510,7 +507,7 @@ absl::Status ScaleImageCalculator::ValidateImageFrame(
absl::Status ScaleImageCalculator::ValidateYUVImage(CalculatorContext* cc,
const YUVImage& yuv_image) {
ABSL_CHECK_EQ(input_format_, ImageFormat::YCBCR420P);
CHECK_EQ(input_format_, ImageFormat::YCBCR420P);
if (!has_header_) {
if (input_width_ != yuv_image.width() ||
input_height_ != yuv_image.height()) {
@@ -18,7 +18,6 @@
#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"
@@ -41,10 +40,10 @@ 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) {
ABSL_CHECK(crop_width);
ABSL_CHECK(crop_height);
ABSL_CHECK(col_start);
ABSL_CHECK(row_start);
CHECK(crop_width);
CHECK(crop_height);
CHECK(col_start);
CHECK(row_start);
double min_aspect_ratio_q = 0.0;
double max_aspect_ratio_q = 0.0;
@@ -84,8 +83,8 @@ absl::Status FindCropDimensions(int input_width, int input_height, //
}
}
ABSL_CHECK_LE(*crop_width, input_width);
ABSL_CHECK_LE(*crop_height, input_height);
CHECK_LE(*crop_width, input_width);
CHECK_LE(*crop_height, input_height);
return absl::OkStatus();
}
@@ -97,8 +96,8 @@ absl::Status FindOutputDimensions(int input_width, //
bool preserve_aspect_ratio, //
int scale_to_multiple_of, //
int* output_width, int* output_height) {
ABSL_CHECK(output_width);
ABSL_CHECK(output_height);
CHECK(output_width);
CHECK(output_height);
if (target_max_area > 0 && input_width * input_height > target_max_area) {
preserve_aspect_ratio = true;
@@ -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"
@@ -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()) {
ABSL_LOG(ERROR) << "Warning: mixing input format types. ";
LOG(ERROR) << "Warning: mixing input format types. ";
}
auto previous_texture = gpu_helper_.CreateSourceTexture(previous_frame);
@@ -14,7 +14,6 @@
#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"
@@ -170,7 +169,7 @@ void RunTest(bool use_gpu, float mix_ratio, cv::Mat& test_result) {
}
}
} else {
ABSL_LOG(ERROR) << "invalid ratio";
LOG(ERROR) << "invalid ratio";
}
}
@@ -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"
@@ -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)) {
ABSL_LOG(ERROR) << "Only 3 or 4 channel 8-bit input image supported";
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)) {
ABSL_LOG(ERROR) << "Only RGB or RGBA input image supported";
LOG(ERROR) << "Only RGB or RGBA input image supported";
}
auto input_texture = gpu_helper_.CreateSourceTexture(input_frame);
-1
View File
@@ -18,7 +18,6 @@ licenses(["notice"])
filegroup(
name = "test_images",
srcs = [
"binary_mask.png",
"dino.jpg",
"dino_quality_50.jpg",
"dino_quality_80.jpg",
Binary file not shown.

Before

Width:  |  Height:  |  Size: 771 B

@@ -79,8 +79,8 @@ class WarpAffineRunnerHolder<ImageFrame> {
}
absl::StatusOr<RunnerType*> GetRunner() {
if (!runner_) {
MP_ASSIGN_OR_RETURN(
runner_, CreateAffineTransformationOpenCvRunner(interpolation_));
ASSIGN_OR_RETURN(runner_,
CreateAffineTransformationOpenCvRunner(interpolation_));
}
return runner_.get();
}
@@ -108,9 +108,8 @@ class WarpAffineRunnerHolder<mediapipe::GpuBuffer> {
}
absl::StatusOr<RunnerType*> GetRunner() {
if (!runner_) {
MP_ASSIGN_OR_RETURN(
runner_, CreateAffineTransformationGlRunner(gl_helper_, gpu_origin_,
interpolation_));
ASSIGN_OR_RETURN(runner_, CreateAffineTransformationGlRunner(
gl_helper_, gpu_origin_, interpolation_));
}
return runner_.get();
}
@@ -152,25 +151,24 @@ class WarpAffineRunnerHolder<mediapipe::Image> {
AffineTransformation::BorderMode border_mode) override {
if (input.UsesGpu()) {
#if !MEDIAPIPE_DISABLE_GPU
MP_ASSIGN_OR_RETURN(auto* runner, gpu_holder_.GetRunner());
MP_ASSIGN_OR_RETURN(
auto result,
runner->Run(input.GetGpuBuffer(), matrix, size, border_mode));
ASSIGN_OR_RETURN(auto* runner, gpu_holder_.GetRunner());
ASSIGN_OR_RETURN(auto result, runner->Run(input.GetGpuBuffer(), matrix,
size, border_mode));
return mediapipe::Image(*result);
#else
return absl::UnavailableError("GPU support is disabled");
#endif // !MEDIAPIPE_DISABLE_GPU
}
#if !MEDIAPIPE_DISABLE_OPENCV
MP_ASSIGN_OR_RETURN(auto* runner, cpu_holder_.GetRunner());
ASSIGN_OR_RETURN(auto* runner, cpu_holder_.GetRunner());
const auto& frame_ptr = input.GetImageFrameSharedPtr();
// Wrap image into image frame.
const ImageFrame image_frame(frame_ptr->Format(), frame_ptr->Width(),
frame_ptr->Height(), frame_ptr->WidthStep(),
const_cast<uint8_t*>(frame_ptr->PixelData()),
[](uint8_t* data){});
MP_ASSIGN_OR_RETURN(auto result,
runner->Run(image_frame, matrix, size, border_mode));
ASSIGN_OR_RETURN(auto result,
runner->Run(image_frame, matrix, size, border_mode));
return mediapipe::Image(std::make_shared<ImageFrame>(std::move(result)));
#else
return absl::UnavailableError("OpenCV support is disabled");
@@ -215,8 +213,8 @@ class WarpAffineCalculatorImpl : public mediapipe::api2::NodeImpl<InterfaceT> {
AffineTransformation::Size output_size;
output_size.width = out_width;
output_size.height = out_height;
MP_ASSIGN_OR_RETURN(auto* runner, holder_.GetRunner());
MP_ASSIGN_OR_RETURN(
ASSIGN_OR_RETURN(auto* runner, holder_.GetRunner());
ASSIGN_OR_RETURN(
auto result,
runner->Run(
*InterfaceT::kInImage(cc), transform, output_size,
-2
View File
@@ -31,14 +31,12 @@ 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,
)
@@ -11,12 +11,10 @@
// 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"
@@ -41,55 +39,64 @@ void DumpPostStreamPacket(Packet* post_stream_packet, const Packet& packet) {
*post_stream_packet = packet;
}
}
} // namespace
absl::Status CallbackPacketCalculator::GetContract(CalculatorContract* cc) {
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.";
// 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) {
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();
}
return absl::OkStatus();
}
absl::Status CallbackPacketCalculator::Open(CalculatorContext* cc) {
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:
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)
<< "Invalid type to dump into.";
<< "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();
}
return absl::OkStatus();
}
absl::Status CallbackPacketCalculator::Process(CalculatorContext* cc) {
return absl::OkStatus();
}
absl::Status Process(CalculatorContext* cc) override {
return absl::OkStatus();
}
};
REGISTER_CALCULATOR(CallbackPacketCalculator);
@@ -1,39 +0,0 @@
// 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_
-68
View File
@@ -87,7 +87,6 @@ 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",
@@ -182,7 +181,6 @@ 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,
@@ -200,7 +198,6 @@ 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",
],
)
@@ -448,7 +445,6 @@ 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",
@@ -478,7 +474,6 @@ 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",
@@ -625,7 +620,6 @@ mediapipe_proto_library(
deps = [
"//mediapipe/framework:calculator_options_proto",
"//mediapipe/framework:calculator_proto",
"//mediapipe/gpu:gpu_origin_proto",
],
)
@@ -655,18 +649,7 @@ 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"],
@@ -716,11 +699,9 @@ 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",
],
)
@@ -756,8 +737,6 @@ 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({
@@ -814,7 +793,6 @@ 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,
)
@@ -980,48 +958,6 @@ cc_test(
],
)
cc_library(
name = "tensor_to_joints_calculator",
srcs = ["tensor_to_joints_calculator.cc"],
hdrs = ["tensor_to_joints_calculator.h"],
deps = [
":tensor_to_joints_calculator_cc_proto",
"//mediapipe/framework:calculator_framework",
"//mediapipe/framework/api2:node",
"//mediapipe/framework/formats:body_rig_cc_proto",
"//mediapipe/framework/formats:tensor",
"//mediapipe/framework/port:ret_check",
"//mediapipe/framework/port:status",
],
alwayslink = 1,
)
mediapipe_proto_library(
name = "tensor_to_joints_calculator_proto",
srcs = ["tensor_to_joints_calculator.proto"],
deps = [
"//mediapipe/framework:calculator_options_proto",
"//mediapipe/framework:calculator_proto",
],
)
cc_test(
name = "tensor_to_joints_calculator_test",
srcs = ["tensor_to_joints_calculator_test.cc"],
deps = [
":tensor_to_joints_calculator",
":tensor_to_joints_calculator_cc_proto",
"//mediapipe/framework:calculator_framework",
"//mediapipe/framework:calculator_runner",
"//mediapipe/framework:timestamp",
"//mediapipe/framework/formats:body_rig_cc_proto",
"//mediapipe/framework/formats:tensor",
"//mediapipe/framework/port:gtest_main",
"//mediapipe/framework/port:parse_text_proto",
"@com_google_absl//absl/strings",
],
)
cc_library(
name = "image_to_tensor_calculator",
srcs = ["image_to_tensor_calculator.cc"],
@@ -1049,8 +985,6 @@ 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"],
@@ -1143,7 +1077,6 @@ 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",
@@ -1271,7 +1204,6 @@ 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",
],
}),
@@ -20,7 +20,6 @@
#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"
@@ -349,7 +348,7 @@ absl::Status AudioToTensorCalculator::Process(CalculatorContext* cc) {
return absl::InvalidArgumentError(
"The audio data should be stored in column-major.");
}
ABSL_CHECK(channels_match || mono_output);
CHECK(channels_match || mono_output);
const Matrix& input = channels_match ? input_frame
// Mono mixdown.
: input_frame.colwise().mean();
@@ -458,7 +457,7 @@ absl::Status AudioToTensorCalculator::SetupStreamingResampler(
}
void AudioToTensorCalculator::AppendZerosToSampleBuffer(int num_samples) {
ABSL_CHECK_GE(num_samples, 0); // Ensured by `UpdateContract`.
CHECK_GE(num_samples, 0); // Ensured by `UpdateContract`.
if (num_samples == 0) {
return;
}
@@ -517,8 +516,8 @@ absl::Status AudioToTensorCalculator::OutputTensor(const Matrix& block,
// The last two elements are Nyquist component.
fft_output_matrix(fft_size_ - 2) = fft_output_[1]; // Nyquist real part
fft_output_matrix(fft_size_ - 1) = 0.0f; // Nyquist imagery part
MP_ASSIGN_OR_RETURN(output_tensor, ConvertToTensor(fft_output_matrix,
{2, fft_size_ / 2}));
ASSIGN_OR_RETURN(output_tensor, ConvertToTensor(fft_output_matrix,
{2, fft_size_ / 2}));
break;
}
case Options::WITH_DC_AND_NYQUIST: {
@@ -529,7 +528,7 @@ absl::Status AudioToTensorCalculator::OutputTensor(const Matrix& block,
// The last two elements are Nyquist component.
fft_output_matrix(fft_size_) = fft_output_[1]; // Nyquist real part
fft_output_matrix(fft_size_ + 1) = 0.0f; // Nyquist imagery part
MP_ASSIGN_OR_RETURN(
ASSIGN_OR_RETURN(
output_tensor,
ConvertToTensor(fft_output_matrix, {2, (fft_size_ + 2) / 2}));
break;
@@ -537,7 +536,7 @@ absl::Status AudioToTensorCalculator::OutputTensor(const Matrix& block,
case Options::WITHOUT_DC_AND_NYQUIST: {
Matrix fft_output_matrix =
Eigen::Map<const Matrix>(fft_output_.data() + 2, 1, fft_size_ - 2);
MP_ASSIGN_OR_RETURN(
ASSIGN_OR_RETURN(
output_tensor,
ConvertToTensor(fft_output_matrix, {2, (fft_size_ - 2) / 2}));
break;
@@ -547,8 +546,8 @@ absl::Status AudioToTensorCalculator::OutputTensor(const Matrix& block,
}
} else {
MP_ASSIGN_OR_RETURN(output_tensor,
ConvertToTensor(block, {num_channels_, num_samples_}));
ASSIGN_OR_RETURN(output_tensor,
ConvertToTensor(block, {num_channels_, num_samples_}));
}
kTensorsOut(cc).Send(std::move(output_tensor), timestamp);
return absl::OkStatus();
@@ -161,9 +161,9 @@ absl::Status BertPreprocessorCalculator::Open(CalculatorContext* cc) {
&kMetadataExtractorSideIn(cc).Get();
const tflite::ProcessUnit* tokenizer_metadata =
metadata_extractor->GetInputProcessUnit(kTokenizerProcessUnitIndex);
MP_ASSIGN_OR_RETURN(tokenizer_,
tasks::text::tokenizers::CreateTokenizerFromProcessUnit(
tokenizer_metadata, metadata_extractor));
ASSIGN_OR_RETURN(tokenizer_,
tasks::text::tokenizers::CreateTokenizerFromProcessUnit(
tokenizer_metadata, metadata_extractor));
auto* input_tensors_metadata = metadata_extractor->GetInputTensorMetadata();
input_ids_tensor_index_ = FindTensorIndexByMetadataName(
@@ -67,10 +67,9 @@ absl::StatusOr<std::vector<std::vector<int>>> RunBertPreprocessorCalculator(
tool::AddVectorSink("tensors", &graph_config, &output_packets);
std::string model_buffer = tasks::core::LoadBinaryContent(model_path.data());
MP_ASSIGN_OR_RETURN(
std::unique_ptr<ModelMetadataExtractor> metadata_extractor,
ModelMetadataExtractor::CreateFromModelBuffer(model_buffer.data(),
model_buffer.size()));
ASSIGN_OR_RETURN(std::unique_ptr<ModelMetadataExtractor> metadata_extractor,
ModelMetadataExtractor::CreateFromModelBuffer(
model_buffer.data(), model_buffer.size()));
// Run the graph.
CalculatorGraph graph;
MP_RETURN_IF_ERROR(graph.Initialize(
@@ -18,7 +18,6 @@
#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"
@@ -66,7 +65,7 @@ template <typename T>
Tensor MakeTensor(std::initializer_list<int> shape,
std::initializer_list<T> values) {
Tensor tensor(TensorElementType<T>::value, shape);
ABSL_CHECK_EQ(values.size(), tensor.shape().num_elements())
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;
@@ -16,7 +16,6 @@
#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"
@@ -192,19 +191,18 @@ class ImageToTensorCalculator : public Node {
}
#if MEDIAPIPE_DISABLE_GPU
MP_ASSIGN_OR_RETURN(auto image, GetInputImage(kIn(cc)));
ASSIGN_OR_RETURN(auto image, GetInputImage(kIn(cc)));
#else
const bool is_input_gpu = kInGpu(cc).IsConnected();
MP_ASSIGN_OR_RETURN(auto image, is_input_gpu ? GetInputImage(kInGpu(cc))
: GetInputImage(kIn(cc)));
ASSIGN_OR_RETURN(auto image, is_input_gpu ? GetInputImage(kInGpu(cc))
: GetInputImage(kIn(cc)));
#endif // MEDIAPIPE_DISABLE_GPU
RotatedRect roi = GetRoi(image->width(), image->height(), norm_rect);
const int tensor_width = params_.output_width.value_or(image->width());
const int tensor_height = params_.output_height.value_or(image->height());
MP_ASSIGN_OR_RETURN(auto padding,
PadRoi(tensor_width, tensor_height,
options_.keep_aspect_ratio(), &roi));
ASSIGN_OR_RETURN(auto padding, PadRoi(tensor_width, tensor_height,
options_.keep_aspect_ratio(), &roi));
if (kOutLetterboxPadding(cc).IsConnected()) {
kOutLetterboxPadding(cc).Send(padding);
}
@@ -248,20 +246,20 @@ class ImageToTensorCalculator : public Node {
if (!gpu_converter_) {
#if !MEDIAPIPE_DISABLE_GPU
#if MEDIAPIPE_METAL_ENABLED
MP_ASSIGN_OR_RETURN(
ASSIGN_OR_RETURN(
gpu_converter_,
CreateMetalConverter(cc, GetBorderMode(options_.border_mode())));
#elif MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_31
MP_ASSIGN_OR_RETURN(gpu_converter_,
CreateImageToGlBufferTensorConverter(
cc, DoesGpuInputStartAtBottom(options_),
GetBorderMode(options_.border_mode())));
ASSIGN_OR_RETURN(gpu_converter_,
CreateImageToGlBufferTensorConverter(
cc, DoesGpuInputStartAtBottom(options_),
GetBorderMode(options_.border_mode())));
#else
if (!gpu_converter_) {
MP_ASSIGN_OR_RETURN(gpu_converter_,
CreateImageToGlTextureTensorConverter(
cc, DoesGpuInputStartAtBottom(options_),
GetBorderMode(options_.border_mode())));
ASSIGN_OR_RETURN(gpu_converter_,
CreateImageToGlTextureTensorConverter(
cc, DoesGpuInputStartAtBottom(options_),
GetBorderMode(options_.border_mode())));
}
if (!gpu_converter_) {
return absl::UnimplementedError(
@@ -273,24 +271,22 @@ class ImageToTensorCalculator : public Node {
} else {
if (!cpu_converter_) {
#if !MEDIAPIPE_DISABLE_OPENCV
MP_ASSIGN_OR_RETURN(
cpu_converter_,
CreateOpenCvConverter(
cc, GetBorderMode(options_.border_mode()),
GetOutputTensorType(/*uses_gpu=*/false, params_)));
ASSIGN_OR_RETURN(cpu_converter_,
CreateOpenCvConverter(
cc, GetBorderMode(options_.border_mode()),
GetOutputTensorType(/*uses_gpu=*/false, params_)));
// TODO: FrameBuffer-based converter needs to call GetGpuBuffer()
// to get access to a FrameBuffer view. Investigate if GetGpuBuffer() can be
// made available even with MEDIAPIPE_DISABLE_GPU set.
#elif MEDIAPIPE_ENABLE_HALIDE
MP_ASSIGN_OR_RETURN(
cpu_converter_,
CreateFrameBufferConverter(
cc, GetBorderMode(options_.border_mode()),
GetOutputTensorType(/*uses_gpu=*/false, params_)));
ASSIGN_OR_RETURN(cpu_converter_,
CreateFrameBufferConverter(
cc, GetBorderMode(options_.border_mode()),
GetOutputTensorType(/*uses_gpu=*/false, params_)));
#else
ABSL_LOG(FATAL) << "Cannot create image to tensor CPU converter since "
"MEDIAPIPE_DISABLE_OPENCV is defined and "
"MEDIAPIPE_ENABLE_HALIDE is not defined.";
LOG(FATAL) << "Cannot create image to tensor CPU converter since "
"MEDIAPIPE_DISABLE_OPENCV is defined and "
"MEDIAPIPE_ENABLE_HALIDE is not defined.";
#endif // !MEDIAPIPE_DISABLE_HALIDE
}
}
@@ -18,7 +18,6 @@
#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"
@@ -206,7 +205,7 @@ mediapipe::ImageFormat::Format GetImageFormat(int image_channels) {
} else if (image_channels == 1) {
return ImageFormat::GRAY8;
}
ABSL_CHECK(false) << "Unsupported input image channles: " << image_channels;
CHECK(false) << "Unsupported input image channles: " << image_channels;
}
Packet MakeImageFramePacket(cv::Mat input) {
@@ -175,9 +175,9 @@ absl::Status FrameBufferProcessor::CropRotateResize90Degrees(
cropped_buffer_ = std::make_unique<uint8_t[]>(cropped_buffer_size);
cropped_buffer_size_ = cropped_buffer_size;
}
MP_ASSIGN_OR_RETURN(
cropped, frame_buffer::CreateFromRawBuffer(
cropped_buffer_.get(), cropped_dims, input->format()));
ASSIGN_OR_RETURN(cropped,
frame_buffer::CreateFromRawBuffer(
cropped_buffer_.get(), cropped_dims, input->format()));
}
MP_RETURN_IF_ERROR(
frame_buffer::Crop(*input, left, top, right, bottom, cropped.get()));
@@ -194,9 +194,9 @@ absl::Status FrameBufferProcessor::CropRotateResize90Degrees(
rotated_buffer_ = std::make_unique<uint8_t[]>(rotated_buffer_size);
rotated_buffer_size_ = rotated_buffer_size;
}
MP_ASSIGN_OR_RETURN(auto rotated, frame_buffer::CreateFromRawBuffer(
rotated_buffer_.get(), rotated_dims,
cropped->format()));
ASSIGN_OR_RETURN(auto rotated, frame_buffer::CreateFromRawBuffer(
rotated_buffer_.get(), rotated_dims,
cropped->format()));
}
MP_RETURN_IF_ERROR(
frame_buffer::Rotate(*cropped, rotation_degrees, rotated.get()));
@@ -217,10 +217,9 @@ absl::Status FrameBufferProcessor::ConvertToFloatTensor(
RET_CHECK(output_tensor.element_type() == Tensor::ElementType::kFloat32);
constexpr float kInputImageRangeMin = 0.0f;
constexpr float kInputImageRangeMax = 255.0f;
MP_ASSIGN_OR_RETURN(
auto transform,
GetValueRangeTransformation(kInputImageRangeMin, kInputImageRangeMax,
range_min, range_max));
ASSIGN_OR_RETURN(auto transform, GetValueRangeTransformation(
kInputImageRangeMin, kInputImageRangeMax,
range_min, range_max));
return frame_buffer::ToFloatTensor(*input_frame, transform.scale,
transform.offset, output_tensor);
}
@@ -255,7 +255,7 @@ class GlProcessor : public ImageToTensorConverter {
<< "OpenGL ES 3.1 is required.";
command_queue_ = tflite::gpu::gl::NewCommandQueue(gpu_info);
MP_ASSIGN_OR_RETURN(
ASSIGN_OR_RETURN(
auto extractor,
SubRectExtractorGl::Create(gl_helper_.GetGlContext(),
input_starts_at_bottom, border_mode));
@@ -293,10 +293,10 @@ class GlProcessor : public ImageToTensorConverter {
constexpr float kInputImageRangeMin = 0.0f;
constexpr float kInputImageRangeMax = 1.0f;
MP_ASSIGN_OR_RETURN(auto transform,
GetValueRangeTransformation(
kInputImageRangeMin, kInputImageRangeMax,
range_min, range_max));
ASSIGN_OR_RETURN(auto transform,
GetValueRangeTransformation(kInputImageRangeMin,
kInputImageRangeMax,
range_min, range_max));
const int output_size = output_tensor.bytes() / output_shape.dims[0];
auto buffer_view = output_tensor.GetOpenGlBufferWriteView();
@@ -22,7 +22,6 @@
#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"
@@ -193,10 +192,10 @@ class GlProcessor : public ImageToTensorConverter {
constexpr float kInputImageRangeMin = 0.0f;
constexpr float kInputImageRangeMax = 1.0f;
MP_ASSIGN_OR_RETURN(auto transform,
GetValueRangeTransformation(
kInputImageRangeMin, kInputImageRangeMax,
range_min, range_max));
ASSIGN_OR_RETURN(auto transform,
GetValueRangeTransformation(kInputImageRangeMin,
kInputImageRangeMax,
range_min, range_max));
auto tensor_view = output_tensor.GetOpenGlTexture2dWriteView();
MP_RETURN_IF_ERROR(ExtractSubRect(input_texture, roi,
/*flip_horizontaly=*/false,
@@ -260,7 +259,7 @@ 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();
ABSL_LOG_IF(FATAL, !gl_context) << "GlContext is not bound to the thread.";
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,
@@ -304,7 +303,6 @@ class GlProcessor : public ImageToTensorConverter {
glBindTexture(GL_TEXTURE_2D, 0);
glActiveTexture(GL_TEXTURE0);
glBindTexture(GL_TEXTURE_2D, 0);
glFlush();
return absl::OkStatus();
}
@@ -345,9 +345,9 @@ class MetalProcessor : public ImageToTensorConverter {
absl::Status Init(CalculatorContext* cc, BorderMode border_mode) {
metal_helper_ = [[MPPMetalHelper alloc] initWithCalculatorContext:cc];
RET_CHECK(metal_helper_);
MP_ASSIGN_OR_RETURN(extractor_, SubRectExtractorMetal::Make(
metal_helper_.mtlDevice,
OutputFormat::kF32C4, border_mode));
ASSIGN_OR_RETURN(extractor_, SubRectExtractorMetal::Make(
metal_helper_.mtlDevice,
OutputFormat::kF32C4, border_mode));
return absl::OkStatus();
}
@@ -373,7 +373,7 @@ class MetalProcessor : public ImageToTensorConverter {
constexpr float kInputImageRangeMin = 0.0f;
constexpr float kInputImageRangeMax = 1.0f;
MP_ASSIGN_OR_RETURN(
ASSIGN_OR_RETURN(
auto transform,
GetValueRangeTransformation(kInputImageRangeMin, kInputImageRangeMax,
range_min, range_max));
@@ -159,7 +159,7 @@ class OpenCvProcessor : public ImageToTensorConverter {
constexpr float kInputImageRangeMin = 0.0f;
constexpr float kInputImageRangeMax = 255.0f;
MP_ASSIGN_OR_RETURN(
ASSIGN_OR_RETURN(
auto transform,
GetValueRangeTransformation(kInputImageRangeMin, kInputImageRangeMax,
range_min, range_max));
@@ -88,20 +88,6 @@ 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.
@@ -60,7 +60,7 @@ absl::Status InferenceCalculatorCpuImpl::UpdateContract(
}
absl::Status InferenceCalculatorCpuImpl::Open(CalculatorContext* cc) {
MP_ASSIGN_OR_RETURN(inference_runner_, CreateInferenceRunner(cc));
ASSIGN_OR_RETURN(inference_runner_, CreateInferenceRunner(cc));
return absl::OkStatus();
}
@@ -71,8 +71,8 @@ absl::Status InferenceCalculatorCpuImpl::Process(CalculatorContext* cc) {
const auto& input_tensors = *kInTensors(cc);
RET_CHECK(!input_tensors.empty());
MP_ASSIGN_OR_RETURN(std::vector<Tensor> output_tensors,
inference_runner_->Run(cc, input_tensors));
ASSIGN_OR_RETURN(std::vector<Tensor> output_tensors,
inference_runner_->Run(cc, input_tensors));
kOutTensors(cc).Send(std::move(output_tensors));
return absl::OkStatus();
}
@@ -84,11 +84,11 @@ absl::Status InferenceCalculatorCpuImpl::Close(CalculatorContext* cc) {
absl::StatusOr<std::unique_ptr<InferenceRunner>>
InferenceCalculatorCpuImpl::CreateInferenceRunner(CalculatorContext* cc) {
MP_ASSIGN_OR_RETURN(auto model_packet, GetModelAsPacket(cc));
MP_ASSIGN_OR_RETURN(auto op_resolver_packet, GetOpResolverAsPacket(cc));
ASSIGN_OR_RETURN(auto model_packet, GetModelAsPacket(cc));
ASSIGN_OR_RETURN(auto op_resolver_packet, GetOpResolverAsPacket(cc));
const int interpreter_num_threads =
cc->Options<mediapipe::InferenceCalculatorOptions>().cpu_num_thread();
MP_ASSIGN_OR_RETURN(TfLiteDelegatePtr delegate, MaybeCreateDelegate(cc));
ASSIGN_OR_RETURN(TfLiteDelegatePtr delegate, MaybeCreateDelegate(cc));
return CreateInferenceInterpreterDelegateRunner(
std::move(model_packet), std::move(op_resolver_packet),
std::move(delegate), interpreter_num_threads);
@@ -100,7 +100,7 @@ absl::Status InferenceCalculatorGlImpl::GpuInferenceRunner::Init(
absl::Status InferenceCalculatorGlImpl::GpuInferenceRunner::LoadModel(
CalculatorContext* cc) {
MP_ASSIGN_OR_RETURN(model_packet_, GetModelAsPacket(cc));
ASSIGN_OR_RETURN(model_packet_, GetModelAsPacket(cc));
const auto& model = *model_packet_.Get();
if (kSideInOpResolver(cc).IsConnected()) {
const tflite::OpResolver& op_resolver = kSideInOpResolver(cc).Get();
@@ -12,7 +12,6 @@
// See the License for the specific language governing permissions and
// limitations under the License.
#include <cstdint>
#include <cstring>
#include <memory>
#include <string>
@@ -27,7 +26,6 @@
#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"
@@ -70,21 +68,14 @@ class InferenceCalculatorGlAdvancedImpl
const mediapipe::InferenceCalculatorOptions::Delegate::Gpu&
gpu_delegate_options);
absl::Status ReadGpuCaches(tflite::gpu::TFLiteGPURunner* gpu_runner) const;
// Writes caches to disk based on |cache_writing_behavior_|.
absl::Status SaveGpuCachesBasedOnBehavior(
tflite::gpu::TFLiteGPURunner* gpu_runner) const;
absl::Status SaveGpuCaches(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.
@@ -170,7 +161,7 @@ absl::Status
InferenceCalculatorGlAdvancedImpl::GpuInferenceRunner::InitTFLiteGPURunner(
CalculatorContext* cc,
const mediapipe::InferenceCalculatorOptions::Delegate& delegate) {
MP_ASSIGN_OR_RETURN(model_packet_, GetModelAsPacket(cc));
ASSIGN_OR_RETURN(model_packet_, GetModelAsPacket(cc));
const auto& model = *model_packet_.Get();
bool allow_precision_loss = delegate.gpu().allow_precision_loss();
@@ -241,8 +232,7 @@ InferenceCalculatorGlAdvancedImpl::GpuInferenceRunner::InitTFLiteGPURunner(
MP_RETURN_IF_ERROR(
on_disk_cache_helper_.ReadGpuCaches(tflite_gpu_runner_.get()));
MP_RETURN_IF_ERROR(tflite_gpu_runner_->Build());
return on_disk_cache_helper_.SaveGpuCachesBasedOnBehavior(
tflite_gpu_runner_.get());
return on_disk_cache_helper_.SaveGpuCaches(tflite_gpu_runner_.get());
}
#if defined(MEDIAPIPE_ANDROID) || defined(MEDIAPIPE_CHROMIUMOS)
@@ -271,51 +261,24 @@ 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 {
if (use_kernel_caching_) {
// Save kernel file.
MP_ASSIGN_OR_RETURN(std::vector<uint8_t> kernel_cache,
gpu_runner->GetSerializedBinaryCache());
ASSIGN_OR_RETURN(std::vector<uint8_t> kernel_cache,
gpu_runner->GetSerializedBinaryCache());
std::string cache_str(kernel_cache.begin(), kernel_cache.end());
MP_RETURN_IF_ERROR(
mediapipe::file::SetContents(cached_kernel_filename_, cache_str));
}
if (use_serialized_model_) {
// Save serialized model file.
MP_ASSIGN_OR_RETURN(std::vector<uint8_t> serialized_model_vec,
gpu_runner->GetSerializedModel());
ASSIGN_OR_RETURN(std::vector<uint8_t> serialized_model_vec,
gpu_runner->GetSerializedModel());
absl::string_view serialized_model(
reinterpret_cast<char*>(serialized_model_vec.data()),
serialized_model_vec.size());
@@ -355,12 +318,6 @@ 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 {
@@ -412,8 +369,8 @@ absl::Status InferenceCalculatorGlAdvancedImpl::Process(CalculatorContext* cc) {
RET_CHECK(!input_tensors.empty());
auto output_tensors = absl::make_unique<std::vector<Tensor>>();
MP_ASSIGN_OR_RETURN(*output_tensors,
gpu_inference_runner_->Process(cc, input_tensors));
ASSIGN_OR_RETURN(*output_tensors,
gpu_inference_runner_->Process(cc, input_tensors));
kOutTensors(cc).Send(std::move(output_tensors));
return absl::OkStatus();
@@ -21,7 +21,6 @@
#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"
@@ -75,7 +74,7 @@ tflite::gpu::BHWC BhwcFromTensorShape(const Tensor::Shape& shape) {
break;
default:
// Handles 0 and >4.
ABSL_LOG(FATAL)
LOG(FATAL)
<< "Dimensions size must be in range [1,4] for GPU inference, but "
<< shape.dims.size() << " is provided";
}
@@ -208,9 +207,9 @@ absl::Status InferenceCalculatorMetalImpl::Close(CalculatorContext* cc) {
absl::Status InferenceCalculatorMetalImpl::InitInterpreter(
CalculatorContext* cc) {
MP_ASSIGN_OR_RETURN(model_packet_, GetModelAsPacket(cc));
ASSIGN_OR_RETURN(model_packet_, GetModelAsPacket(cc));
const auto& model = *model_packet_.Get();
MP_ASSIGN_OR_RETURN(auto op_resolver_packet, GetOpResolverAsPacket(cc));
ASSIGN_OR_RETURN(auto op_resolver_packet, GetOpResolverAsPacket(cc));
const auto& op_resolver = op_resolver_packet.Get();
tflite::InterpreterBuilder interpreter_builder(model, op_resolver);
AddDelegate(cc, &interpreter_builder);
@@ -16,7 +16,7 @@
#include <string>
#include <vector>
#include "absl/log/absl_check.h"
#include "absl/log/check.h"
#include "absl/strings/str_cat.h"
#include "absl/strings/str_replace.h"
#include "absl/strings/string_view.h"
@@ -58,7 +58,7 @@ absl::Status InferenceCalculatorXnnpackImpl::UpdateContract(
}
absl::Status InferenceCalculatorXnnpackImpl::Open(CalculatorContext* cc) {
MP_ASSIGN_OR_RETURN(inference_runner_, CreateInferenceRunner(cc));
ASSIGN_OR_RETURN(inference_runner_, CreateInferenceRunner(cc));
return absl::OkStatus();
}
@@ -69,8 +69,8 @@ absl::Status InferenceCalculatorXnnpackImpl::Process(CalculatorContext* cc) {
const auto& input_tensors = *kInTensors(cc);
RET_CHECK(!input_tensors.empty());
MP_ASSIGN_OR_RETURN(std::vector<Tensor> output_tensors,
inference_runner_->Run(cc, input_tensors));
ASSIGN_OR_RETURN(std::vector<Tensor> output_tensors,
inference_runner_->Run(cc, input_tensors));
kOutTensors(cc).Send(std::move(output_tensors));
return absl::OkStatus();
}
@@ -82,11 +82,11 @@ absl::Status InferenceCalculatorXnnpackImpl::Close(CalculatorContext* cc) {
absl::StatusOr<std::unique_ptr<InferenceRunner>>
InferenceCalculatorXnnpackImpl::CreateInferenceRunner(CalculatorContext* cc) {
MP_ASSIGN_OR_RETURN(auto model_packet, GetModelAsPacket(cc));
MP_ASSIGN_OR_RETURN(auto op_resolver_packet, GetOpResolverAsPacket(cc));
ASSIGN_OR_RETURN(auto model_packet, GetModelAsPacket(cc));
ASSIGN_OR_RETURN(auto op_resolver_packet, GetOpResolverAsPacket(cc));
const int interpreter_num_threads =
cc->Options<mediapipe::InferenceCalculatorOptions>().cpu_num_thread();
MP_ASSIGN_OR_RETURN(TfLiteDelegatePtr delegate, CreateDelegate(cc));
ASSIGN_OR_RETURN(TfLiteDelegatePtr delegate, CreateDelegate(cc));
return CreateInferenceInterpreterDelegateRunner(
std::move(model_packet), std::move(op_resolver_packet),
std::move(delegate), interpreter_num_threads);
@@ -106,7 +106,7 @@ absl::Status RegexPreprocessorCalculator::Open(CalculatorContext* cc) {
return absl::InvalidArgumentError("No tensor metadata found");
}
MP_ASSIGN_OR_RETURN(
ASSIGN_OR_RETURN(
const auto* tokenizer_metadata,
metadata_extractor->FindFirstProcessUnit(
*tensor_metadata, tflite::ProcessUnitOptions_RegexTokenizerOptions));
@@ -115,9 +115,9 @@ absl::Status RegexPreprocessorCalculator::Open(CalculatorContext* cc) {
}
const tflite::RegexTokenizerOptions* regex_tokenizer_options =
tokenizer_metadata->options_as<tflite::RegexTokenizerOptions>();
MP_ASSIGN_OR_RETURN(tokenizer_,
tasks::text::tokenizers::CreateRegexTokenizerFromOptions(
regex_tokenizer_options, metadata_extractor));
ASSIGN_OR_RETURN(tokenizer_,
tasks::text::tokenizers::CreateRegexTokenizerFromOptions(
regex_tokenizer_options, metadata_extractor));
const auto& options =
cc->Options<mediapipe::RegexPreprocessorCalculatorOptions>();
@@ -67,10 +67,9 @@ absl::StatusOr<std::vector<int>> RunRegexPreprocessorCalculator(
tool::AddVectorSink("tensors", &graph_config, &output_packets);
std::string model_buffer = tasks::core::LoadBinaryContent(kTestModelPath);
MP_ASSIGN_OR_RETURN(
std::unique_ptr<ModelMetadataExtractor> metadata_extractor,
ModelMetadataExtractor::CreateFromModelBuffer(model_buffer.data(),
model_buffer.size()));
ASSIGN_OR_RETURN(std::unique_ptr<ModelMetadataExtractor> metadata_extractor,
ModelMetadataExtractor::CreateFromModelBuffer(
model_buffer.data(), model_buffer.size()));
// Run the graph.
CalculatorGraph graph;
MP_RETURN_IF_ERROR(graph.Initialize(
@@ -12,15 +12,9 @@
// 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"
@@ -28,8 +22,7 @@
#include "mediapipe/framework/formats/tensor.h"
#include "mediapipe/framework/port.h"
#include "mediapipe/framework/port/ret_check.h"
#include "mediapipe/gpu/gpu_buffer_format.h"
#include "mediapipe/gpu/gpu_origin.pb.h"
#include "mediapipe/util/resource_util.h"
#if !MEDIAPIPE_DISABLE_GPU
#include "mediapipe/gpu/gpu_buffer.h"
@@ -50,50 +43,12 @@
#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>
@@ -103,7 +58,6 @@ constexpr char kImageFrameTag[] = "IMAGE";
constexpr char kGpuBufferTag[] = "IMAGE_GPU";
constexpr char kTensorsTag[] = "TENSORS";
constexpr char kMatrixTag[] = "MATRIX";
} // namespace
namespace mediapipe {
@@ -155,7 +109,7 @@ class TensorConverterCalculator : public CalculatorBase {
private:
absl::Status InitGpu(CalculatorContext* cc);
absl::Status LoadOptions(CalculatorContext* cc, bool use_gpu);
absl::Status LoadOptions(CalculatorContext* cc);
template <class T>
absl::Status NormalizeImage(const ImageFrame& image_frame,
bool flip_vertically, float* tensor_ptr);
@@ -191,8 +145,7 @@ 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)
<< "Only one input tag of {IMAGE, IMAGE_GPU, MATRIX} may be specified";
1);
if (cc->Inputs().HasTag(kImageFrameTag)) {
cc->Inputs().Tag(kImageFrameTag).Set<ImageFrame>();
@@ -220,6 +173,8 @@ 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;
@@ -232,8 +187,6 @@ absl::Status TensorConverterCalculator::Open(CalculatorContext* cc) {
}
#endif // !MEDIAPIPE_DISABLE_GPU
MP_RETURN_IF_ERROR(LoadOptions(cc, use_gpu_));
return absl::OkStatus();
}
@@ -406,7 +359,6 @@ absl::Status TensorConverterCalculator::ProcessGPU(CalculatorContext* cc) {
glActiveTexture(GL_TEXTURE1);
glBindTexture(GL_TEXTURE_2D, 0);
#endif // MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_31
glFlush();
src.Release();
return absl::OkStatus();
}));
@@ -426,34 +378,23 @@ absl::Status TensorConverterCalculator::InitGpu(CalculatorContext* cc) {
// Get input image sizes.
const auto& input =
cc->Inputs().Tag(kGpuBufferTag).Get<mediapipe::GpuBuffer>();
mediapipe::GpuBufferFormat format = input.format();
mediapipe::ImageFormat::Format format =
mediapipe::ImageFormatForGpuBufferFormat(input.format());
const bool include_alpha = (max_num_channels_ == 4);
const bool single_channel = (max_num_channels_ == 1);
RET_CHECK(format == mediapipe::GpuBufferFormat::kBGRA32 ||
format == mediapipe::GpuBufferFormat::kRGB24 ||
format == mediapipe::GpuBufferFormat::kRGBA32 ||
format == mediapipe::GpuBufferFormat::kRGBAFloat128 ||
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 (!(format == mediapipe::ImageFormat::GRAY8 ||
format == mediapipe::ImageFormat::SRGB ||
format == mediapipe::ImageFormat::SRGBA))
RET_CHECK_FAIL() << "Unsupported GPU input format.";
if (include_alpha && (format != mediapipe::ImageFormat::SRGBA))
RET_CHECK_FAIL() << "Num input channels is less than desired output.";
#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"glsl(
R"(
#include <metal_stdlib>
using namespace metal;
@@ -472,7 +413,7 @@ 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);",
@@ -482,8 +423,8 @@ 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"glsl(out_buf[linear_index + 1] = pixel.y;
out_buf[linear_index + 2] = pixel.z;)glsl",
single_channel ? "" : R"(out_buf[linear_index + 1] = pixel.y;
out_buf[linear_index + 2] = pixel.z;)",
/*$4=*/include_alpha ? "out_buf[linear_index + 3] = pixel.w;" : "");
NSString* library_source =
@@ -501,17 +442,17 @@ 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]() -> absl::Status {
&single_channel]()
-> absl::Status {
#if MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_31
// Shader to convert GL Texture to Shader Storage Buffer Object (SSBO),
// with normalization to either: [0,1] or [-1,1].
const std::string shader_source = absl::Substitute(
R"glsl( #version 310 es
// Shader to convert GL Texture to Shader Storage Buffer Object (SSBO),
// with normalization to either: [0,1] or [-1,1].
const std::string shader_source = absl::Substitute(
R"( #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;
@@ -525,40 +466,38 @@ 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(),
/*$3=*/
output_range_.has_value()
? absl::Substitute(
"pixel = pixel * float($0) + float($1);",
(output_range_->second - output_range_->first),
output_range_->first)
: "",
/*$4=*/flip_vertically_ ? "(width_height.y - 1 - gid.y)" : "gid.y",
/*$5=*/
single_channel
? ""
: R"glsl(output_data.elements[linear_index + 1] = pixel.y;
output_data.elements[linear_index + 2] = pixel.z;)glsl",
/*$6=*/
include_alpha ? "output_data.elements[linear_index + 3] = pixel.w;"
: "",
/*$7=*/max_num_channels_);
GLuint shader = glCreateShader(GL_COMPUTE_SHADER);
const GLchar* sources[] = {shader_source.c_str()};
glShaderSource(shader, 1, sources, NULL);
glCompileShader(shader);
GLint compiled = GL_FALSE;
glGetShaderiv(shader, GL_COMPILE_STATUS, &compiled);
RET_CHECK(compiled == GL_TRUE);
to_buffer_program_ = glCreateProgram();
glAttachShader(to_buffer_program_, shader);
glDeleteShader(shader);
glLinkProgram(to_buffer_program_);
})",
/*$0=*/kWorkgroupSize, /*$1=*/input.width(), /*$2=*/input.height(),
/*$3=*/
output_range_.has_value()
? absl::Substitute("pixel = pixel * float($0) + float($1);",
(output_range_->second - output_range_->first),
output_range_->first)
: "",
/*$4=*/flip_vertically_ ? "(width_height.y - 1 - gid.y)" : "gid.y",
/*$5=*/
single_channel ? ""
: R"(output_data.elements[linear_index + 1] = pixel.y;
output_data.elements[linear_index + 2] = pixel.z;)",
/*$6=*/
include_alpha ? "output_data.elements[linear_index + 3] = pixel.w;"
: "",
/*$7=*/max_num_channels_);
GLuint shader = glCreateShader(GL_COMPUTE_SHADER);
const GLchar* sources[] = {shader_source.c_str()};
glShaderSource(shader, 1, sources, NULL);
glCompileShader(shader);
GLint compiled = GL_FALSE;
glGetShaderiv(shader, GL_COMPILE_STATUS, &compiled);
RET_CHECK(compiled == GL_TRUE);
to_buffer_program_ = glCreateProgram();
glAttachShader(to_buffer_program_, shader);
glDeleteShader(shader);
glLinkProgram(to_buffer_program_);
#else
// OpenGL ES 3.0 fragment shader Texture2d -> Texture2d conversion.
const std::string shader_source = absl::Substitute(
R"glsl(
// OpenGL ES 3.0 fragment shader Texture2d -> Texture2d conversion.
const std::string shader_source = absl::Substitute(
R"(
#if __VERSION__ < 130
#define in varying
#endif // __VERSION__ < 130
@@ -584,51 +523,49 @@ 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",
/*$1=*/
flip_vertically_
? "vec2(sample_coordinate.x, 1.0 - sample_coordinate.y);"
: "sample_coordinate;",
/*$2=*/output_range_.has_value()
? absl::Substitute(
"pixel = pixel * float($0) + float($1);",
(output_range_->second - output_range_->first),
output_range_->first)
: "",
/*$3=*/single_channel ? "" : R"glsl(fragColor.g = pixel.g;
fragColor.b = pixel.b;)glsl",
/*$4=*/
include_alpha ? "fragColor.a = pixel.a;"
: (single_channel ? "" : "fragColor.a = 1.0;"));
})",
/*$0=*/single_channel ? "vec1" : "vec4",
/*$1=*/
flip_vertically_
? "vec2(sample_coordinate.x, 1.0 - sample_coordinate.y);"
: "sample_coordinate;",
/*$2=*/output_range_.has_value()
? absl::Substitute("pixel = pixel * float($0) + float($1);",
(output_range_->second - output_range_->first),
output_range_->first)
: "",
/*$3=*/single_channel ? "" : R"(fragColor.g = pixel.g;
fragColor.b = pixel.b;)",
/*$4=*/
include_alpha ? "fragColor.a = pixel.a;"
: (single_channel ? "" : "fragColor.a = 1.0;"));
const GLint attr_location[NUM_ATTRIBUTES] = {
ATTRIB_VERTEX,
ATTRIB_TEXTURE_POSITION,
};
const GLchar* attr_name[NUM_ATTRIBUTES] = {
"position",
"texture_coordinate",
};
// shader program and params
mediapipe::GlhCreateProgram(
mediapipe::kBasicVertexShader, shader_source.c_str(),
NUM_ATTRIBUTES, &attr_name[0], attr_location, &to_tex2d_program_);
RET_CHECK(to_tex2d_program_) << "Problem initializing the program.";
glUseProgram(to_tex2d_program_);
glUniform1i(glGetUniformLocation(to_tex2d_program_, "frame"), 1);
glGenFramebuffers(1, &framebuffer_);
const GLint attr_location[NUM_ATTRIBUTES] = {
ATTRIB_VERTEX,
ATTRIB_TEXTURE_POSITION,
};
const GLchar* attr_name[NUM_ATTRIBUTES] = {
"position",
"texture_coordinate",
};
// shader program and params
mediapipe::GlhCreateProgram(
mediapipe::kBasicVertexShader, shader_source.c_str(), NUM_ATTRIBUTES,
&attr_name[0], attr_location, &to_tex2d_program_);
RET_CHECK(to_tex2d_program_) << "Problem initializing the program.";
glUseProgram(to_tex2d_program_);
glUniform1i(glGetUniformLocation(to_tex2d_program_, "frame"), 1);
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,
bool use_gpu) {
absl::Status TensorConverterCalculator::LoadOptions(CalculatorContext* cc) {
// Get calculator options specified in the graph.
const auto& options =
cc->Options<::mediapipe::TensorConverterCalculatorOptions>();
@@ -645,7 +582,7 @@ 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());
ABSL_CHECK_GT(output_range_->second, output_range_->first);
CHECK_GT(output_range_->second, output_range_->first);
}
// Custom div and sub values.
@@ -656,16 +593,16 @@ absl::Status TensorConverterCalculator::LoadOptions(CalculatorContext* cc,
}
// Get y-flip mode.
MP_ASSIGN_OR_RETURN(flip_vertically_, ShouldFlipVertically(options, use_gpu));
flip_vertically_ = options.flip_vertically();
// 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();
ABSL_CHECK_GE(max_num_channels_, 1);
ABSL_CHECK_LE(max_num_channels_, 4);
ABSL_CHECK_NE(max_num_channels_, 2);
CHECK_GE(max_num_channels_, 1);
CHECK_LE(max_num_channels_, 4);
CHECK_NE(max_num_channels_, 2);
return absl::OkStatus();
}
@@ -3,7 +3,6 @@ syntax = "proto2";
package mediapipe;
import "mediapipe/framework/calculator.proto";
import "mediapipe/gpu/gpu_origin.proto";
// Full Example:
//
@@ -44,16 +43,8 @@ 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];
@@ -12,15 +12,10 @@
// 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"
@@ -29,10 +24,8 @@
#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"
@@ -47,7 +40,7 @@ constexpr char kTransposeOptionsString[] =
} // namespace
using RandomEngine = std::mt19937_64;
using ::testing::HasSubstr;
using testing::Eq;
const uint32_t kSeed = 1234;
const int kNumSizes = 8;
const int sizes[kNumSizes][2] = {{1, 1}, {12, 1}, {1, 9}, {2, 2},
@@ -60,7 +53,7 @@ class TensorConverterCalculatorTest : public ::testing::Test {
bool row_major_matrix = false) {
RandomEngine random(kSeed);
std::uniform_real_distribution<> uniform_dist(0, 1.0);
auto matrix = std::make_unique<Matrix>();
auto matrix = ::absl::make_unique<Matrix>();
matrix->resize(num_rows, num_columns);
if (row_major_matrix) {
for (int y = 0; y < num_rows; ++y) {
@@ -108,7 +101,7 @@ TEST_F(TensorConverterCalculatorTest, RandomMatrixColMajor) {
tool::AddVectorSink("tensor", &graph_config, &output_packets);
// Run the graph.
graph_ = std::make_unique<CalculatorGraph>();
graph_ = absl::make_unique<CalculatorGraph>();
MP_ASSERT_OK(graph_->Initialize(graph_config));
MP_ASSERT_OK(graph_->StartRun({}));
@@ -117,12 +110,12 @@ TEST_F(TensorConverterCalculatorTest, RandomMatrixColMajor) {
// Wait until the calculator done processing.
MP_ASSERT_OK(graph_->WaitUntilIdle());
ASSERT_EQ(output_packets.size(), 1);
EXPECT_EQ(1, output_packets.size());
// Get and process results.
const std::vector<Tensor>& tensor_vec =
output_packets[0].Get<std::vector<Tensor>>();
ASSERT_EQ(tensor_vec.size(), 1);
EXPECT_EQ(1, tensor_vec.size());
const Tensor* tensor = &tensor_vec[0];
EXPECT_EQ(Tensor::ElementType::kFloat32, tensor->element_type());
@@ -134,7 +127,7 @@ 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_FLOAT_EQ(tensor_buffer[i], expected) << "at i = " << i;
EXPECT_EQ(expected, tensor_buffer[i]) << "at i = " << i;
}
// Fully close graph at end, otherwise calculator+tensors are destroyed
@@ -170,7 +163,7 @@ TEST_F(TensorConverterCalculatorTest, RandomMatrixRowMajor) {
tool::AddVectorSink("tensor", &graph_config, &output_packets);
// Run the graph.
graph_ = std::make_unique<CalculatorGraph>();
graph_ = absl::make_unique<CalculatorGraph>();
MP_ASSERT_OK(graph_->Initialize(graph_config));
MP_ASSERT_OK(graph_->StartRun({}));
@@ -179,12 +172,12 @@ TEST_F(TensorConverterCalculatorTest, RandomMatrixRowMajor) {
// Wait until the calculator done processing.
MP_ASSERT_OK(graph_->WaitUntilIdle());
ASSERT_EQ(output_packets.size(), 1);
EXPECT_EQ(1, output_packets.size());
// Get and process results.
const std::vector<Tensor>& tensor_vec =
output_packets[0].Get<std::vector<Tensor>>();
ASSERT_EQ(tensor_vec.size(), 1);
EXPECT_EQ(1, tensor_vec.size());
const Tensor* tensor = &tensor_vec[0];
EXPECT_EQ(Tensor::ElementType::kFloat32, tensor->element_type());
@@ -196,7 +189,7 @@ 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(tensor_buffer[i], expected) << "at i = " << i;
EXPECT_EQ(expected, tensor_buffer[i]) << "at i = " << i;
}
// Fully close graph at end, otherwise calculator+tensors are destroyed
@@ -234,7 +227,7 @@ TEST_F(TensorConverterCalculatorTest, CustomDivAndSub) {
// 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);
auto input_image = absl::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(
@@ -246,12 +239,12 @@ TEST_F(TensorConverterCalculatorTest, CustomDivAndSub) {
// Get and process results.
const std::vector<Tensor>& tensor_vec =
output_packets[0].Get<std::vector<Tensor>>();
ASSERT_EQ(tensor_vec.size(), 1);
EXPECT_EQ(1, tensor_vec.size());
const Tensor* tensor = &tensor_vec[0];
EXPECT_EQ(Tensor::ElementType::kFloat32, tensor->element_type());
auto view = tensor->GetCpuReadView();
EXPECT_FLOAT_EQ(*view.buffer<float>(), 67.0f);
EXPECT_FLOAT_EQ(67.0f, *view.buffer<float>());
// Fully close graph at end, otherwise calculator+tensors are destroyed
// after calling WaitUntilDone().
@@ -266,29 +259,32 @@ TEST_F(TensorConverterCalculatorTest, SetOutputRange) {
for (std::pair<float, float> range : range_values) {
CalculatorGraph graph;
CalculatorGraphConfig graph_config =
mediapipe::ParseTextProtoOrDie<CalculatorGraphConfig>(absl::Substitute(
R"pb(
input_stream: "input_image"
node {
calculator: "TensorConverterCalculator"
input_stream: "IMAGE:input_image"
output_stream: "TENSORS:tensor"
options {
[mediapipe.TensorConverterCalculatorOptions.ext] {
output_tensor_float_range { min: $0 max: $1 }
}
}
mediapipe::ParseTextProtoOrDie<CalculatorGraphConfig>(
absl::Substitute(R"(
input_stream: "input_image"
node {
calculator: "TensorConverterCalculator"
input_stream: "IMAGE:input_image"
output_stream: "TENSORS:tensor"
options {
[mediapipe.TensorConverterCalculatorOptions.ext] {
output_tensor_float_range {
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 = std::make_unique<ImageFrame>(ImageFormat::GRAY8, 1, 1);
auto input_image = absl::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(
@@ -296,23 +292,26 @@ TEST_F(TensorConverterCalculatorTest, SetOutputRange) {
// Wait until the calculator finishes processing.
MP_ASSERT_OK(graph.WaitUntilIdle());
ASSERT_EQ(output_packets.size(), 1);
EXPECT_THAT(output_packets.size(), Eq(1));
// Get and process results.
const std::vector<Tensor>& tensor_vec =
output_packets[0].Get<std::vector<Tensor>>();
ASSERT_EQ(tensor_vec.size(), 1);
EXPECT_THAT(tensor_vec.size(), Eq(1));
const Tensor* tensor = &tensor_vec[0];
// Calculate the expected normalized value:
float expected_value =
float normalized_value =
range.first + (200 * (range.second - range.first)) / 255.0;
EXPECT_EQ(tensor->element_type(), Tensor::ElementType::kFloat32);
EXPECT_THAT(tensor->element_type(), Eq(Tensor::ElementType::kFloat32));
auto view = tensor->GetCpuReadView();
float actual_value = *view.buffer<float>();
EXPECT_FLOAT_EQ(actual_value, expected_value);
float dataf = *view.buffer<float>();
EXPECT_THAT(
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().
@@ -321,153 +320,4 @@ 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
@@ -1,84 +0,0 @@
// 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.
#include "mediapipe/calculators/tensor/tensor_to_joints_calculator.h"
#include <utility>
#include "mediapipe/calculators/tensor/tensor_to_joints_calculator.pb.h"
#include "mediapipe/framework/calculator_framework.h"
#include "mediapipe/framework/formats/body_rig.pb.h"
#include "mediapipe/framework/formats/tensor.h"
#include "mediapipe/framework/port/ret_check.h"
namespace mediapipe {
namespace api2 {
namespace {
// Number of values in 6D representation of rotation.
constexpr int kRotation6dSize = 6;
} // namespace
class TensorToJointsCalculatorImpl
: public mediapipe::api2::NodeImpl<TensorToJointsCalculator> {
public:
absl::Status Open(CalculatorContext* cc) override {
const auto& options = cc->Options<TensorToJointsCalculatorOptions>();
// Get number of joints.
RET_CHECK_GE(options.num_joints(), 0);
num_joints_ = options.num_joints();
// Get start index.
start_index_ = options.start_index();
return absl::OkStatus();
}
absl::Status Process(CalculatorContext* cc) override {
// Skip if Tensor is empty.
if (kInTensor(cc).IsEmpty()) {
return absl::OkStatus();
}
// Get raw floats from the Tensor.
const Tensor& tensor = kInTensor(cc).Get();
RET_CHECK_EQ(tensor.shape().num_elements(),
num_joints_ * kRotation6dSize + start_index_)
<< "Unexpected number of values in Tensor";
const float* raw_floats = tensor.GetCpuReadView().buffer<float>();
// Convert raw floats into Joint rotations.
JointList joints;
for (int joint_idx = 0; joint_idx < num_joints_; ++joint_idx) {
Joint* joint = joints.add_joint();
for (int idx_6d = 0; idx_6d < kRotation6dSize; ++idx_6d) {
joint->add_rotation_6d(
raw_floats[start_index_ + joint_idx * kRotation6dSize + idx_6d]);
}
}
kOutJoints(cc).Send(std::move(joints));
return absl::OkStatus();
}
private:
int num_joints_ = 0;
int start_index_ = 0;
};
MEDIAPIPE_NODE_IMPLEMENTATION(TensorToJointsCalculatorImpl);
} // namespace api2
} // namespace mediapipe
@@ -1,64 +0,0 @@
// 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_TENSOR_TENSOR_TO_JOINTS_CALCULATOR_H_
#define MEDIAPIPE_CALCULATORS_TENSOR_TENSOR_TO_JOINTS_CALCULATOR_H_
#include <memory>
#include "mediapipe/framework/api2/node.h"
#include "mediapipe/framework/calculator_framework.h"
#include "mediapipe/framework/formats/body_rig.pb.h"
#include "mediapipe/framework/formats/tensor.h"
namespace mediapipe {
namespace api2 {
// A calculator to convert Tensors to JointList.
//
// Calculator fills in only rotation of the joints leaving visibility undefined.
//
// Input:
// TENSOR - std::vector<Tensor> with kFloat32 values
// Vector of tensors to be converted to joints. Only the first tensor will
// be used. Number of values is expected to be multiple of six.
//
// Output:
// JOINTS - JointList
// List of joints with rotations extracted from given tensor and undefined
// visibility.
//
// Example:
// node {
// calculator: "TensorToJointsCalculator"
// input_stream: "TENSOR:tensor"
// output_stream: "JOINTS:joints"
// options: {
// [mediapipe.TensorToJointsCalculatorOptions.ext] {
// num_joints: 56
// start_index: 3
// }
// }
// }
class TensorToJointsCalculator : public NodeIntf {
public:
static constexpr Input<mediapipe::Tensor> kInTensor{"TENSOR"};
static constexpr Output<mediapipe::JointList> kOutJoints{"JOINTS"};
MEDIAPIPE_NODE_INTERFACE(TensorToJointsCalculator, kInTensor, kOutJoints);
};
} // namespace api2
} // namespace mediapipe
#endif // MEDIAPIPE_CALCULATORS_TENSOR_TENSOR_TO_JOINTS_CALCULATOR_H_
@@ -1,32 +0,0 @@
// 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.
syntax = "proto2";
package mediapipe;
import "mediapipe/framework/calculator.proto";
message TensorToJointsCalculatorOptions {
extend CalculatorOptions {
optional TensorToJointsCalculatorOptions ext = 406440177;
}
// Number of joints from the output of the model. Calculator will expect the
// tensor to contain `6 * num_joints + start_index` values.
optional int32 num_joints = 1;
// Index to start reading 6 value blocks from.
optional int32 start_index = 2 [default = 0];
}
@@ -1,123 +0,0 @@
// 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.
#include <cstdint>
#include <memory>
#include <string>
#include <utility>
#include <vector>
#include "absl/strings/substitute.h"
#include "mediapipe/framework/calculator_framework.h"
#include "mediapipe/framework/calculator_runner.h"
#include "mediapipe/framework/formats/body_rig.pb.h"
#include "mediapipe/framework/formats/tensor.h"
#include "mediapipe/framework/port/gmock.h"
#include "mediapipe/framework/port/gtest.h"
#include "mediapipe/framework/port/parse_text_proto.h"
#include "mediapipe/framework/port/status_matchers.h"
#include "mediapipe/framework/timestamp.h"
namespace mediapipe {
namespace api2 {
namespace {
using Node = ::mediapipe::CalculatorGraphConfig::Node;
struct TensorToJointsTestCase {
std::string test_name;
int num_joints;
int start_index;
std::vector<float> raw_values;
std::vector<std::vector<float>> expected_rotations;
};
using TensorToJointsTest = ::testing::TestWithParam<TensorToJointsTestCase>;
TEST_P(TensorToJointsTest, TensorToJointsTest) {
const TensorToJointsTestCase& tc = GetParam();
// Prepare graph.
mediapipe::CalculatorRunner runner(ParseTextProtoOrDie<Node>(absl::Substitute(
R"(
calculator: "TensorToJointsCalculator"
input_stream: "TENSOR:tensor"
output_stream: "JOINTS:joints"
options: {
[mediapipe.TensorToJointsCalculatorOptions.ext] {
num_joints: $0
start_index: $1
}
}
)",
tc.num_joints, tc.start_index)));
// Prepare tensor.
Tensor tensor(Tensor::ElementType::kFloat32,
Tensor::Shape{1, 1, static_cast<int>(tc.raw_values.size()), 1});
float* tensor_buffer = tensor.GetCpuWriteView().buffer<float>();
ASSERT_NE(tensor_buffer, nullptr);
for (int i = 0; i < tc.raw_values.size(); ++i) {
tensor_buffer[i] = tc.raw_values[i];
}
// Send tensor to the graph.
runner.MutableInputs()->Tag("TENSOR").packets.push_back(
mediapipe::MakePacket<Tensor>(std::move(tensor)).At(Timestamp(0)));
// Run the graph.
MP_ASSERT_OK(runner.Run());
const auto& output_packets = runner.Outputs().Tag("JOINTS").packets;
EXPECT_EQ(1, output_packets.size());
const auto& joints = output_packets[0].Get<JointList>();
EXPECT_EQ(joints.joint_size(), tc.expected_rotations.size());
for (int i = 0; i < joints.joint_size(); ++i) {
const Joint& joint = joints.joint(i);
std::vector<float> expected_rotation_6d = tc.expected_rotations[i];
EXPECT_EQ(joint.rotation_6d_size(), expected_rotation_6d.size())
<< "Unexpected joint #" << i << " rotation";
for (int j = 0; j < joint.rotation_6d_size(); ++j) {
EXPECT_EQ(joint.rotation_6d(j), expected_rotation_6d[j])
<< "Unexpected joint #" << i << " rotation";
}
EXPECT_FALSE(joint.has_visibility());
}
}
INSTANTIATE_TEST_SUITE_P(
TensorToJointsTests, TensorToJointsTest,
testing::ValuesIn<TensorToJointsTestCase>({
{"Empty", 0, 3, {0, 0, 0}, {}},
{"Single",
1,
3,
{0, 0, 0, 10, 11, 12, 13, 14, 15},
{{10, 11, 12, 13, 14, 15}}},
{"Double",
2,
3,
{0, 0, 0, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21},
{{10, 11, 12, 13, 14, 15}, {16, 17, 18, 19, 20, 21}}},
}),
[](const testing::TestParamInfo<TensorToJointsTest::ParamType>& info) {
return info.param.test_name;
});
} // namespace
} // namespace api2
} // namespace mediapipe
@@ -110,8 +110,8 @@ absl::Status TensorsToClassificationCalculator::Open(CalculatorContext* cc) {
sort_by_descending_score_ = options.sort_by_descending_score();
if (options.has_label_map_path()) {
std::string string_path;
MP_ASSIGN_OR_RETURN(string_path,
PathToResourceAsFile(options.label_map_path()));
ASSIGN_OR_RETURN(string_path,
PathToResourceAsFile(options.label_map_path()));
std::string label_map_string;
MP_RETURN_IF_ERROR(
mediapipe::GetResourceContents(string_path, &label_map_string));
@@ -15,7 +15,6 @@
#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"
@@ -84,7 +83,7 @@ void ConvertRawValuesToAnchors(const float* raw_anchors, int num_boxes,
void ConvertAnchorsToRawValues(const std::vector<Anchor>& anchors,
int num_boxes, float* raw_anchors) {
ABSL_CHECK_EQ(anchors.size(), num_boxes);
CHECK_EQ(anchors.size(), num_boxes);
int box = 0;
for (const auto& anchor : anchors) {
raw_anchors[box * kNumCoordsPerBox + 0] = anchor.y_center();
@@ -330,7 +329,7 @@ absl::Status TensorsToDetectionsCalculator::Process(CalculatorContext* cc) {
} else if (status.code() == absl::StatusCode::kFailedPrecondition) {
// For initialization error because of hardware limitation, fallback to
// CPU processing.
ABSL_LOG(WARNING) << status.message();
LOG(WARNING) << status.message();
} else {
// For other error, let the error propagates.
return status;
@@ -669,7 +668,7 @@ absl::Status TensorsToDetectionsCalculator::ProcessGPU(
output_detections));
#else
ABSL_LOG(ERROR) << "GPU input on non-Android not supported yet.";
LOG(ERROR) << "GPU input on non-Android not supported yet.";
#endif // !defined(MEDIAPIPE_DISABLE_GL_COMPUTE)
return absl::OkStatus();
}
@@ -704,18 +703,18 @@ absl::Status TensorsToDetectionsCalculator::LoadOptions(CalculatorContext* cc) {
num_boxes_ = options_.num_boxes();
num_coords_ = options_.num_coords();
box_output_format_ = GetBoxFormat(options_);
ABSL_CHECK_NE(options_.max_results(), 0)
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.
ABSL_CHECK_EQ(options_.num_values_per_keypoint(), 2);
CHECK_EQ(options_.num_values_per_keypoint(), 2);
// Check if the output size is equal to the requested boxes and keypoints.
ABSL_CHECK_EQ(options_.num_keypoints() * options_.num_values_per_keypoint() +
kNumCoordsPerBox,
num_coords_);
CHECK_EQ(options_.num_keypoints() * options_.num_values_per_keypoint() +
kNumCoordsPerBox,
num_coords_);
if (kSideInIgnoreClasses(cc).IsConnected()) {
RET_CHECK(!kSideInIgnoreClasses(cc).IsEmpty());
@@ -1155,12 +1154,11 @@ void main() {
}
// TODO support better filtering.
if (class_index_set_.is_allowlist) {
ABSL_CHECK_EQ(class_index_set_.values.size(),
IsClassIndexAllowed(0) ? num_classes_ : num_classes_ - 1)
CHECK_EQ(class_index_set_.values.size(),
IsClassIndexAllowed(0) ? num_classes_ : num_classes_ - 1)
<< "Only all classes >= class 0 or >= class 1";
} else {
ABSL_CHECK_EQ(class_index_set_.values.size(),
IsClassIndexAllowed(0) ? 0 : 1)
CHECK_EQ(class_index_set_.values.size(), IsClassIndexAllowed(0) ? 0 : 1)
<< "Only ignore class 0 is allowed";
}
@@ -1381,12 +1379,11 @@ kernel void scoreKernel(
// TODO support better filtering.
if (class_index_set_.is_allowlist) {
ABSL_CHECK_EQ(class_index_set_.values.size(),
IsClassIndexAllowed(0) ? num_classes_ : num_classes_ - 1)
CHECK_EQ(class_index_set_.values.size(),
IsClassIndexAllowed(0) ? num_classes_ : num_classes_ - 1)
<< "Only all classes >= class 0 or >= class 1";
} else {
ABSL_CHECK_EQ(class_index_set_.values.size(),
IsClassIndexAllowed(0) ? 0 : 1)
CHECK_EQ(class_index_set_.values.size(), IsClassIndexAllowed(0) ? 0 : 1)
<< "Only ignore class 0 is allowed";
}
@@ -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_;
ABSL_CHECK_GT(num_dimensions, 0);
CHECK_GT(num_dimensions, 0);
auto view = input_tensors[0].GetCpuReadView();
auto raw_landmarks = view.buffer<float>();
@@ -174,9 +174,6 @@ class TensorsToSegmentationCalculator : public CalculatorBase {
mediapipe::GlCalculatorHelper gpu_helper_;
GLuint upsample_program_;
#if MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_31
int cached_width_ = 0;
int cached_height_ = 0;
std::unique_ptr<tflite::gpu::gl::GlTexture> small_mask_texture_;
std::unique_ptr<GlProgram> mask_program_31_;
#else
GLuint mask_program_20_;
@@ -267,8 +264,7 @@ absl::Status TensorsToSegmentationCalculator::Process(CalculatorContext* cc) {
{
RET_CHECK(!input_tensors.empty());
RET_CHECK(input_tensors[0].element_type() == Tensor::ElementType::kFloat32);
MP_ASSIGN_OR_RETURN(auto hwc,
GetHwcFromDims(input_tensors[0].shape().dims));
ASSIGN_OR_RETURN(auto hwc, GetHwcFromDims(input_tensors[0].shape().dims));
int tensor_channels = std::get<2>(hwc);
typedef mediapipe::TensorsToSegmentationCalculatorOptions Options;
switch (options_.activation()) {
@@ -311,7 +307,6 @@ absl::Status TensorsToSegmentationCalculator::Close(CalculatorContext* cc) {
upsample_program_ = 0;
#if MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_31
mask_program_31_.reset();
small_mask_texture_.reset();
#else
if (mask_program_20_) glDeleteProgram(mask_program_20_);
mask_program_20_ = 0;
@@ -331,7 +326,7 @@ absl::Status TensorsToSegmentationCalculator::ProcessCpu(
// Get input streams, and dimensions.
const auto& input_tensors =
cc->Inputs().Tag(kTensorsTag).Get<std::vector<Tensor>>();
MP_ASSIGN_OR_RETURN(auto hwc, GetHwcFromDims(input_tensors[0].shape().dims));
ASSIGN_OR_RETURN(auto hwc, GetHwcFromDims(input_tensors[0].shape().dims));
auto [tensor_height, tensor_width, tensor_channels] = hwc;
int output_width = tensor_width, output_height = tensor_height;
if (cc->Inputs().HasTag(kOutputSizeTag)) {
@@ -442,7 +437,7 @@ absl::Status TensorsToSegmentationCalculator::ProcessGpu(
// Get input streams, and dimensions.
const auto& input_tensors =
cc->Inputs().Tag(kTensorsTag).Get<std::vector<Tensor>>();
MP_ASSIGN_OR_RETURN(auto hwc, GetHwcFromDims(input_tensors[0].shape().dims));
ASSIGN_OR_RETURN(auto hwc, GetHwcFromDims(input_tensors[0].shape().dims));
auto [tensor_height, tensor_width, tensor_channels] = hwc;
int output_width = tensor_width, output_height = tensor_height;
if (cc->Inputs().HasTag(kOutputSizeTag)) {
@@ -453,24 +448,21 @@ absl::Status TensorsToSegmentationCalculator::ProcessGpu(
}
// Create initial working mask texture.
#if !(MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_31)
#if MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_31
tflite::gpu::gl::GlTexture small_mask_texture;
#else
mediapipe::GlTexture small_mask_texture;
#endif // MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_31
// Run shader, process mask tensor.
#if MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_31
{
// Only recreate if the size has changed. See b/297809673 for more details.
if (tensor_width != cached_width_ || tensor_height != cached_height_) {
MP_RETURN_IF_ERROR(CreateReadWriteRgbaImageTexture(
tflite::gpu::DataType::UINT8, // GL_RGBA8
{tensor_width, tensor_height}, small_mask_texture_.get()));
cached_width_ = tensor_width;
cached_height_ = tensor_height;
}
MP_RETURN_IF_ERROR(CreateReadWriteRgbaImageTexture(
tflite::gpu::DataType::UINT8, // GL_RGBA8
{tensor_width, tensor_height}, &small_mask_texture));
const int output_index = 0;
glBindImageTexture(output_index, small_mask_texture_->id(), 0, GL_FALSE, 0,
glBindImageTexture(output_index, small_mask_texture.id(), 0, GL_FALSE, 0,
GL_WRITE_ONLY, GL_RGBA8);
auto read_view = input_tensors[0].GetOpenGlBufferReadView();
@@ -555,7 +547,7 @@ absl::Status TensorsToSegmentationCalculator::ProcessGpu(
gpu_helper_.BindFramebuffer(output_texture);
glActiveTexture(GL_TEXTURE1);
#if MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_31
glBindTexture(GL_TEXTURE_2D, small_mask_texture_->id());
glBindTexture(GL_TEXTURE_2D, small_mask_texture.id());
#else
glBindTexture(GL_TEXTURE_2D, small_mask_texture.name());
#endif // MEDIAPIPE_OPENGL_ES_VERSION >= MEDIAPIPE_OPENGL_ES_31
@@ -862,7 +854,6 @@ void main() {
mask_program_31_ = absl::make_unique<GlProgram>();
MP_RETURN_IF_ERROR(GlProgram::CreateWithShader(shader_without_previous,
mask_program_31_.get()));
small_mask_texture_ = absl::make_unique<tflite::gpu::gl::GlTexture>();
#elif MEDIAPIPE_METAL_ENABLED
id<MTLDevice> device = metal_helper_.mtlDevice;
NSString* library_source =
@@ -61,10 +61,9 @@ RunUniversalSentenceEncoderPreprocessorCalculator(absl::string_view text) {
std::string model_buffer =
tasks::core::LoadBinaryContent(kTestModelPath.data());
MP_ASSIGN_OR_RETURN(
std::unique_ptr<ModelMetadataExtractor> metadata_extractor,
ModelMetadataExtractor::CreateFromModelBuffer(model_buffer.data(),
model_buffer.size()));
ASSIGN_OR_RETURN(std::unique_ptr<ModelMetadataExtractor> metadata_extractor,
ModelMetadataExtractor::CreateFromModelBuffer(
model_buffer.data(), model_buffer.size()));
// Run the graph.
CalculatorGraph graph;
MP_RETURN_IF_ERROR(graph.Initialize(
+10 -37
View File
@@ -13,7 +13,6 @@
# 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"])
@@ -315,7 +314,6 @@ 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",
@@ -368,18 +366,18 @@ 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",
],
@@ -408,13 +406,8 @@ cc_library(
alwayslink = 1,
)
# This dependency removed the following 3 targets because they failed Boq conformance test:
#
# tensorflow_jellyfish_deps
# jfprof_lib
# xprofilez_with_server
#
# If you need them plz consider tensorflow_inference_calculator_no_envelope_loader.
# This dependency removed tensorflow_jellyfish_deps and xprofilez_with_server because they failed
# Boq conformance test. Weigh your use case to see if this will work for you.
cc_library(
name = "tensorflow_inference_calculator_for_boq",
srcs = ["tensorflow_inference_calculator.cc"],
@@ -431,7 +424,7 @@ cc_library(
"//mediapipe/framework/port:status",
"//mediapipe/framework/tool:status_util",
"@com_google_absl//absl/base:core_headers",
"@com_google_absl//absl/log:absl_check",
"@com_google_absl//absl/log:check",
"@com_google_absl//absl/memory",
"@com_google_absl//absl/strings",
"@com_google_absl//absl/synchronization",
@@ -490,10 +483,10 @@ 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": [
@@ -521,10 +514,10 @@ 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": [
@@ -557,7 +550,6 @@ 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",
@@ -635,7 +627,6 @@ 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",
@@ -657,7 +648,6 @@ 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,
@@ -672,7 +662,6 @@ 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,
@@ -688,7 +677,6 @@ 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",
@@ -723,7 +711,6 @@ 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",
@@ -786,7 +773,6 @@ 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",
],
@@ -801,8 +787,6 @@ 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,
@@ -816,7 +800,6 @@ 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,
@@ -830,7 +813,6 @@ 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,
@@ -844,8 +826,6 @@ 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,
@@ -940,24 +920,21 @@ 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:location",
"//mediapipe/framework/formats:location_opencv",
"//mediapipe/framework/port:gtest_main",
"//mediapipe/framework/port:opencv_imgcodecs",
"//mediapipe/util/sequence:media_sequence",
"//mediapipe/util/sequence:media_sequence_util",
"@com_google_absl//absl/log:absl_check",
"@com_google_absl//absl/container:flat_hash_map",
"@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",
],
)
@@ -1140,7 +1117,6 @@ 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",
],
)
@@ -1186,7 +1162,6 @@ 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",
@@ -1268,8 +1243,6 @@ 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",
@@ -12,7 +12,6 @@
// 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"
@@ -29,7 +28,7 @@ namespace mediapipe {
namespace {
absl::Status FillTimeSeriesHeaderIfValid(const Packet& header_packet,
TimeSeriesHeader* header) {
ABSL_CHECK(header);
CHECK(header);
if (header_packet.IsEmpty()) {
return absl::UnknownError("No header found.");
}
@@ -151,7 +151,7 @@ class ObjectDetectionTensorsToDetectionsCalculator : public CalculatorBase {
tf::Tensor input_num_detections_tensor =
tf::Tensor(tf::DT_FLOAT, tf::TensorShape({0}));
if (cc->Inputs().HasTag(kClasses)) {
MP_ASSIGN_OR_RETURN(
ASSIGN_OR_RETURN(
input_num_detections_tensor,
MaybeSqueezeDims(kNumDetections,
cc->Inputs().Tag(kNumDetections).Get<tf::Tensor>()));
@@ -160,12 +160,12 @@ class ObjectDetectionTensorsToDetectionsCalculator : public CalculatorBase {
RET_CHECK_EQ(input_num_detections_tensor.dtype(), tf::DT_FLOAT);
}
MP_ASSIGN_OR_RETURN(
ASSIGN_OR_RETURN(
auto input_boxes_tensor,
MaybeSqueezeDims(kBoxes, cc->Inputs().Tag(kBoxes).Get<tf::Tensor>()));
RET_CHECK_EQ(input_boxes_tensor.dtype(), tf::DT_FLOAT);
MP_ASSIGN_OR_RETURN(
ASSIGN_OR_RETURN(
auto input_scores_tensor,
MaybeSqueezeDims(kScores, cc->Inputs().Tag(kScores).Get<tf::Tensor>()));
RET_CHECK_EQ(input_scores_tensor.dtype(), tf::DT_FLOAT);
@@ -173,7 +173,7 @@ class ObjectDetectionTensorsToDetectionsCalculator : public CalculatorBase {
tf::Tensor input_classes_tensor =
tf::Tensor(tf::DT_FLOAT, tf::TensorShape({0}));
if (cc->Inputs().HasTag(kClasses)) {
MP_ASSIGN_OR_RETURN(
ASSIGN_OR_RETURN(
input_classes_tensor,
MaybeSqueezeDims(kClasses,
cc->Inputs().Tag(kClasses).Get<tf::Tensor>()));
@@ -12,23 +12,21 @@
// 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"
@@ -38,11 +36,7 @@ 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_";
@@ -50,7 +44,6 @@ 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;
@@ -62,24 +55,16 @@ 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 images from the
// OpenCVImageEncoderCalculator,
// * "IMAGE_LABEL", which stores whole image labels from Detection,
// * "FORWARD_FLOW_ENCODED", which stores the encoded optical flow from the same
// calculator,
// * "BBOX" which stores bounding boxes from vector<Detections>,
// * streams with the "FLOAT_FEATURE_${NAME}" pattern, which stores the values
// from vector<float>'s associated with the name ${NAME},
// * "KEYPOINTS" stores a map of 2D keypoints from flat_hash_map<string,
// vector<pair<float, float>>>,
// * "CLIP_MEDIA_ID", which stores the clip's media ID as a string.
// * "CLIP_LABEL_${NAME}" which stores sparse feature labels, ID and scores in
// mediapipe::Detection. In the input Detection, the score field is required,
// and label and label_id are optional but at least one of them should be set.
// "IMAGE_${NAME}", "BBOX_${NAME}", and "KEYPOINTS_${NAME}" will also store
// prefixed versions of each stream, which allows for multiple image streams to
// be included. However, the default names are suppored by more tools.
// The supported input stream tags are "IMAGE", which stores the encoded
// images from the OpenCVImageEncoderCalculator, "FORWARD_FLOW_ENCODED", which
// stores the encoded optical flow from the same calculator, "BBOX" which stores
// bounding boxes from vector<Detections>, and streams with the
// "FLOAT_FEATURE_${NAME}" pattern, which stores the values from vector<float>'s
// associated with the name ${NAME}. "KEYPOINTS" stores a map of 2D keypoints
// from flat_hash_map<string, vector<pair<float, float>>>. "IMAGE_${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 {
@@ -115,9 +100,6 @@ 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()
@@ -130,10 +112,6 @@ 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;
@@ -172,18 +150,9 @@ 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>>();
}
@@ -215,11 +184,6 @@ 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();
@@ -233,19 +197,8 @@ 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;
@@ -274,41 +227,12 @@ 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) -
@@ -419,34 +343,6 @@ 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] == '_') {
@@ -497,7 +393,6 @@ 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());
@@ -510,46 +405,6 @@ 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.score().empty()) {
continue;
}
if (detection.label().empty() && detection.label_id().empty()) {
return absl::InvalidArgumentError(
"detection.label and detection.label_id can't be both empty");
}
// Allow empty label (for indexed feature inputs), but if label is not
// empty, it should have the same size as the score field.
if (!detection.label().empty()) {
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 score field.
if (!detection.label_id().empty()) {
if (detection.label_id().size() != detection.score().size()) {
return absl::InvalidArgumentError(
"Different size of detection.label_id and detection.score");
}
}
for (int i = 0; i < detection.score().size(); ++i) {
if (!detection.label_id().empty()) {
mpms::AddClipLabelIndex(key, detection.label_id(i),
sequence_.get());
}
if (!detection.label().empty()) {
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 =
@@ -557,36 +412,9 @@ class PackMediaSequenceCalculator : public CalculatorBase {
sizeof(*kFloatContextFeaturePrefixTag) -
1);
RET_CHECK_EQ(cc->InputTimestamp(), Timestamp::PostStream());
for (const auto& value :
cc->Inputs().Tag(tag).Get<std::vector<float>>()) {
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());
}
mpms::SetContextFeatureFloats(
key, cc->Inputs().Tag(tag).Get<std::vector<float>>(),
sequence_.get());
}
if (absl::StartsWith(tag, kFloatFeaturePrefixTag) &&
!cc->Inputs().Tag(tag).IsEmpty()) {
@@ -632,7 +460,6 @@ 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>>()) {
@@ -661,9 +488,6 @@ 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()) {
@@ -677,10 +501,6 @@ 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());
}
}
}
}
@@ -728,14 +548,10 @@ 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_;
};
@@ -12,32 +12,27 @@
// See the License for the specific language governing permissions and
// limitations under the License.
#include <cstdint>
#include <memory>
#include <string>
#include <vector>
#include <algorithm>
#include "absl/log/absl_check.h"
#include "absl/container/flat_hash_map.h"
#include "absl/memory/memory.h"
#include "absl/status/status.h"
#include "absl/strings/str_cat.h"
#include "absl/strings/numbers.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/location.h"
#include "mediapipe/framework/formats/location_opencv.h"
#include "mediapipe/framework/packet.h"
#include "mediapipe/framework/port/gmock.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 {
@@ -59,23 +54,13 @@ 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";
constexpr char kClipLabelAnotherTag[] = "CLIP_LABEL_ANOTHER";
class PackMediaSequenceCalculatorTest : public ::testing::Test {
protected:
@@ -83,14 +68,10 @@ 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 std::vector<std::string>& input_side_packets = {
"SEQUENCE_EXAMPLE:input_sequence"}) {
const bool output_as_zero_timestamp = false) {
CalculatorGraphConfig::Node config;
config.set_calculator("PackMediaSequenceCalculator");
for (const std::string& side_packet : input_side_packets) {
config.add_input_side_packet(side_packet);
}
config.add_input_side_packet("SEQUENCE_EXAMPLE:input_sequence");
config.add_output_stream("SEQUENCE_EXAMPLE:output_sequence");
for (const std::string& stream : input_streams) {
config.add_input_stream(stream);
@@ -332,76 +313,6 @@ 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>();
@@ -457,315 +368,6 @@ 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");
@@ -927,10 +529,6 @@ 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]);
}
}
@@ -1073,10 +671,6 @@ 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]);
}
}
@@ -1167,456 +761,6 @@ TEST_F(PackMediaSequenceCalculatorTest, PacksTwoMaskDetections) {
testing::ElementsAreArray(::std::vector<std::string>({"mask"})));
}
TEST_F(PackMediaSequenceCalculatorTest, PackThreeClipLabels) {
SetUpCalculator(
/*input_streams=*/{"CLIP_LABEL_TEST:test", "CLIP_LABEL_OTHER:test2",
"CLIP_LABEL_ANOTHER:test3"},
/*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)));
// No label for detection_3.
Detection detection_3;
detection_3.add_label_id(3);
detection_3.add_label_id(4);
detection_3.add_score(0.3);
detection_3.add_score(0.4);
runner_->MutableInputs()
->Tag(kClipLabelAnotherTag)
.packets.push_back(MakePacket<Detection>(detection_3).At(Timestamp(3)));
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));
ASSERT_FALSE(mpms::HasClipLabelString("ANOTHER", output_sequence));
ASSERT_THAT(mpms::GetClipLabelIndex("ANOTHER", output_sequence),
testing::ElementsAre(3, 4));
ASSERT_THAT(mpms::GetClipLabelConfidence("ANOTHER", output_sequence),
testing::ElementsAre(0.3, 0.4));
}
TEST_F(PackMediaSequenceCalculatorTest, PackTwoClipLabels_EmptyScore) {
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>();
// No score in detection_1. detection_1 is ignored.
Detection detection_1;
detection_1.add_label("label_1");
detection_1.add_label("label_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());
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_FALSE(mpms::HasClipLabelString("TEST", output_sequence));
ASSERT_FALSE(mpms::HasClipLabelIndex("TEST", output_sequence));
ASSERT_FALSE(mpms::HasClipLabelConfidence("TEST", output_sequence));
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_NoLabelOrLabelIndex) {
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>();
// No label or label_index in detection_1.
Detection detection_1;
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(
"detection.label and detection.label_id can't be both empty")));
}
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 scores 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.score")));
}
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"}, {},
@@ -1921,7 +1065,6 @@ 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());
@@ -12,7 +12,6 @@
// 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"
@@ -100,11 +99,10 @@ class TensorSqueezeDimensionsCalculator : public CalculatorBase {
}
}
if (remove_dims_.empty()) {
ABSL_LOG(ERROR)
<< "TensorSqueezeDimensionsCalculator is squeezing input with "
"no single-dimensions. Calculator will be a no-op.";
ABSL_LOG(ERROR) << "Input to TensorSqueezeDimensionsCalculator has shape "
<< tensor_shape.DebugString();
LOG(ERROR) << "TensorSqueezeDimensionsCalculator is squeezing input with "
"no single-dimensions. Calculator will be a no-op.";
LOG(ERROR) << "Input to TensorSqueezeDimensionsCalculator has shape "
<< tensor_shape.DebugString();
}
}
};
@@ -14,7 +14,6 @@
#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"
@@ -100,7 +99,7 @@ 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.
ABSL_CHECK(3 == input_tensor.dims())
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);

Some files were not shown because too many files have changed in this diff Show More