talksik/mediapipe
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Project import generated by Copybara.

Browse Source 
GitOrigin-RevId: f7d09ed033907b893638a8eb4148efa11c0f09a6
This commit is contained in:
MediaPipe Team
2020-11-04 19:09:58 -05:00
committed by chuoling
parent a8d6ce95c4
commit f96eadd6df
250 changed files with 15261 additions and 4620 deletions
Download Patch File Download Diff File Expand all files Collapse all files
mediapipe/python/BUILD
+20 -8
View File
@@ -16,14 +16,6 @@ load("@pybind11_bazel//:build_defs.bzl", "pybind_extension")
licenses(["notice"]) # Apache 2.0
cc_library(
name = "builtin_calculators",
deps = [
"//mediapipe/calculators/core:pass_through_calculator",
"//mediapipe/graphs/pose_tracking:upper_body_pose_tracking_cpu_deps",
],
)
pybind_extension(
name = "_framework_bindings",
srcs = ["framework_bindings.cc"],
@@ -50,5 +42,25 @@ pybind_extension(
"//mediapipe/python/pybind:resource_util",
"//mediapipe/python/pybind:timestamp",
"//mediapipe/python/pybind:validated_graph_config",
# Type registration.
"//mediapipe/framework:basic_types_registration",
"//mediapipe/framework/formats:classification_registration",
"//mediapipe/framework/formats:detection_registration",
"//mediapipe/framework/formats:landmark_registration",
],
)
cc_library(
name = "builtin_calculators",
deps = [
"//mediapipe/calculators/core:pass_through_calculator",
"//mediapipe/calculators/core:split_normalized_landmark_list_calculator",
"//mediapipe/modules/face_detection:face_detection_front_cpu",
"//mediapipe/modules/face_landmark:face_landmark_front_cpu",
"//mediapipe/modules/hand_landmark:hand_landmark_tracking_cpu",
"//mediapipe/modules/palm_detection:palm_detection_cpu",
"//mediapipe/modules/pose_detection:pose_detection_cpu",
"//mediapipe/modules/pose_landmark:pose_landmark_upper_body_by_roi_cpu",
"//mediapipe/modules/pose_landmark:pose_landmark_upper_body_smoothed_cpu",
],
)
mediapipe/python/pybind/calculator_graph.cc
+6
View File
@@ -31,6 +31,10 @@
namespace mediapipe {
namespace python {
// A mutex to guard the output stream observer python callback function.
// Only one python callback can run at once.
absl::Mutex callback_mutex;
template <typename T>
T ParseProto(const py::object& proto_object) {
T proto;
@@ -393,6 +397,8 @@ void CalculatorGraphSubmodule(pybind11::module* module) {
pybind11::function callback_fn) {
RaisePyErrorIfNotOk(self->ObserveOutputStream(
stream_name, [callback_fn, stream_name](const Packet& packet) {
// Acquire a mutex so that only one callback_fn can run at once.
absl::MutexLock lock(&callback_mutex);
callback_fn(stream_name, packet);
return mediapipe::OkStatus();
}));
mediapipe/python/solution_base.py
+472
View File
@@ -0,0 +1,472 @@
# Copyright 2020 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.
# Lint as: python3
"""MediaPipe SolutionBase module.
MediaPipe SolutionBase is the common base class for the high-level MediaPipe
Solution APIs such as BlazeFace, hand tracking, and BlazePose. The SolutionBase
class contains the shared logic among the high-level Solution APIs including
graph initialization, processing image/audio data, and graph shutdown. Thus,
users can easily create new MediaPipe Solution APIs on top of the SolutionBase
class.
"""
import collections
import enum
import os
from typing import Any, Iterable, List, Mapping, NamedTuple, Optional, Union
import numpy as np
from google.protobuf import descriptor
# resources dependency
from mediapipe.framework import calculator_pb2
# pylint: disable=unused-import
from mediapipe.framework.formats import detection_pb2
from mediapipe.calculators.image import image_transformation_calculator_pb2
from mediapipe.calculators.tensor import tensors_to_detections_calculator_pb2
from mediapipe.calculators.util import landmarks_smoothing_calculator_pb2
from mediapipe.calculators.util import logic_calculator_pb2
from mediapipe.calculators.util import thresholding_calculator_pb2
from mediapipe.framework.formats import classification_pb2
from mediapipe.framework.formats import landmark_pb2
from mediapipe.framework.formats import rect_pb2
# pylint: enable=unused-import
from mediapipe.python._framework_bindings import calculator_graph
from mediapipe.python._framework_bindings import image_frame
from mediapipe.python._framework_bindings import packet
from mediapipe.python._framework_bindings import resource_util
from mediapipe.python._framework_bindings import validated_graph_config
import mediapipe.python.packet_creator as packet_creator
import mediapipe.python.packet_getter as packet_getter
RGB_CHANNELS = 3
# TODO: Enable calculator options modification for more calculators.
CALCULATOR_TO_OPTIONS = {
'ImageTransformationCalculator':
image_transformation_calculator_pb2
.ImageTransformationCalculatorOptions,
'LandmarksSmoothingCalculator':
landmarks_smoothing_calculator_pb2.LandmarksSmoothingCalculatorOptions,
'LogicCalculator':
logic_calculator_pb2.LogicCalculatorOptions,
'ThresholdingCalculator':
thresholding_calculator_pb2.ThresholdingCalculatorOptions,
'TensorsToDetectionsCalculator':
tensors_to_detections_calculator_pb2
.TensorsToDetectionsCalculatorOptions,
}
# TODO: Support more packet data types, such as "Any" type.
@enum.unique
class _PacketDataType(enum.Enum):
"""The packet data types supported by the SolutionBase class."""
STRING = 'string'
BOOL = 'bool'
INT = 'int'
FLOAT = 'float'
AUDIO = 'matrix'
IMAGE = 'image_frame'
PROTO = 'proto'
PROTO_LIST = 'proto_list'
@staticmethod
def from_registered_name(registered_name: str) -> '_PacketDataType':
return NAME_TO_TYPE[registered_name]
NAME_TO_TYPE: Mapping[str, '_PacketDataType'] = {
'string':
_PacketDataType.STRING,
'bool':
_PacketDataType.BOOL,
'int':
_PacketDataType.INT,
'float':
_PacketDataType.FLOAT,
'::mediapipe::Matrix':
_PacketDataType.AUDIO,
'::mediapipe::ImageFrame':
_PacketDataType.IMAGE,
'::mediapipe::Classification':
_PacketDataType.PROTO,
'::mediapipe::ClassificationList':
_PacketDataType.PROTO,
'::mediapipe::Detection':
_PacketDataType.PROTO,
'::mediapipe::DetectionList':
_PacketDataType.PROTO,
'::mediapipe::Landmark':
_PacketDataType.PROTO,
'::mediapipe::NormalizedLandmark':
_PacketDataType.PROTO,
'::mediapipe::Rect':
_PacketDataType.PROTO,
'::mediapipe::NormalizedRect':
_PacketDataType.PROTO,
'::mediapipe::NormalizedLandmarkList':
_PacketDataType.PROTO,
'::std::vector<::mediapipe::Classification>':
_PacketDataType.PROTO_LIST,
'::std::vector<::mediapipe::ClassificationList>':
_PacketDataType.PROTO_LIST,
'::std::vector<::mediapipe::Detection>':
_PacketDataType.PROTO_LIST,
'::std::vector<::mediapipe::DetectionList>':
_PacketDataType.PROTO_LIST,
'::std::vector<::mediapipe::Landmark>':
_PacketDataType.PROTO_LIST,
'::std::vector<::mediapipe::NormalizedLandmark>':
_PacketDataType.PROTO_LIST,
'::std::vector<::mediapipe::NormalizedLandmarkList>':
_PacketDataType.PROTO_LIST,
'::std::vector<::mediapipe::Rect>':
_PacketDataType.PROTO_LIST,
'::std::vector<::mediapipe::NormalizedRect>':
_PacketDataType.PROTO_LIST,
}
class SolutionBase:
"""The common base class for the high-level MediaPipe Solution APIs.
The SolutionBase class contains the shared logic among the high-level solution
APIs including graph initialization, processing image/audio data, and graph
shutdown.
Example usage:
hand_tracker = solution_base.SolutionBase(
binary_graph_path='mediapipe/modules/hand_landmark/hand_landmark_tracking_cpu.binarypb',
side_inputs={'num_hands': 2})
# Read an image and convert the BGR image to RGB.
input_image = cv2.cvtColor(cv2.imread('/tmp/hand.png'), COLOR_BGR2RGB)
results = hand_tracker.process(input_image)
print(results.palm_detections)
print(results.multi_hand_landmarks)
hand_tracker.close()
"""
def __init__(
self,
binary_graph_path: Optional[str] = None,
graph_config: Optional[calculator_pb2.CalculatorGraphConfig] = None,
calculator_params: Optional[Mapping[str, Any]] = None,
side_inputs: Optional[Mapping[str, Any]] = None,
outputs: Optional[List[str]] = None):
"""Initializes the SolutionBase object.
Args:
binary_graph_path: The path to a binary mediapipe graph file (.binarypb).
graph_config: A CalculatorGraphConfig proto message or its text proto
format.
calculator_params: A mapping from the
{calculator_name}.{options_field_name} str to the field value.
side_inputs: A mapping from the side packet name to the packet raw data.
outputs: A list of the graph output stream names to observe. If the list
is empty, all the output streams listed in the graph config will be
automatically observed by default.
Raises:
FileNotFoundError: If the binary graph file can't be found.
RuntimeError: If the underlying calculator graph can't be successfully
initialized or started.
ValueError: If any of the following:
a) If not exactly one of 'binary_graph_path' or 'graph_config' arguments
is provided.
b) If the graph validation process contains error.
c) If the registered type name of the streams and side packets can't be
found.
d) If the calculator options of the calculator listed in
calculator_params is not allowed to be modified.
e) If the calculator options field is a repeated field but the field
value to be set is not iterable.
"""
if bool(binary_graph_path) == bool(graph_config):
raise ValueError(
"Must provide exactly one of 'binary_graph_path' or 'graph_config'.")
# MediaPipe package root path
root_path = os.sep.join( os.path.abspath(__file__).split(os.sep)[:-3])
resource_util.set_resource_dir(root_path)
validated_graph = validated_graph_config.ValidatedGraphConfig()
if binary_graph_path:
validated_graph.initialize(
binary_graph_path=os.path.join(root_path, binary_graph_path))
else:
validated_graph.initialize(graph_config=graph_config)
canonical_graph_config_proto = self._initialize_graph_interface(
validated_graph, side_inputs, outputs)
if calculator_params:
self._modify_calculator_options(canonical_graph_config_proto,
calculator_params)
self._graph = calculator_graph.CalculatorGraph(
graph_config=canonical_graph_config_proto)
self._simulated_timestamp = 0
self._graph_outputs = {}
def callback(stream_name: str, output_packet: packet.Packet) -> None:
self._graph_outputs[stream_name] = output_packet
for stream_name in self._output_stream_type_info.keys():
self._graph.observe_output_stream(stream_name, callback)
input_side_packets = {
name: self._make_packet(self._side_input_type_info[name], data)
for name, data in (side_inputs or {}).items()
}
self._graph.start_run(input_side_packets)
# TODO: Use "inspect.Parameter" to fetch the input argument names and
# types from "_input_stream_type_info" and then auto generate the process
# method signature by "inspect.Signature" in __init__.
def process(
self, input_data: Union[np.ndarray, Mapping[str,
np.ndarray]]) -> NamedTuple:
"""Processes a set of RGB image data and output SolutionOutputs.
Args:
input_data: Either a single numpy ndarray object representing the solo
image input of a graph or a mapping from the stream name to the image
data that represents every input streams of a graph.
Raises:
NotImplementedError: If input_data contains non image data.
RuntimeError: If the underlying graph occurs any error.
ValueError: If the input image data is not three channel RGB.
Returns:
A NamedTuple object that contains the output data of a graph run.
The field names in the NamedTuple object are mapping to the graph output
stream names.
Examples:
solution = solution_base.SolutionBase(graph_config=hand_landmark_graph)
results = solution.process(cv2.imread('/tmp/hand0.png')[:, :, ::-1])
print(results.detection)
results = solution.process(
{'video_in' : cv2.imread('/tmp/hand1.png')[:, :, ::-1]})
print(results.hand_landmarks)
"""
self._graph_outputs.clear()
if isinstance(input_data, np.ndarray):
if len(self._input_stream_type_info.keys()) != 1:
raise ValueError(
"Can't process single image input since the graph has more than one input streams."
)
input_dict = {next(iter(self._input_stream_type_info)): input_data}
else:
input_dict = input_data
# Set the timestamp increment to 33333 us to simulate the 30 fps video
# input.
self._simulated_timestamp += 33333
for stream_name, data in input_dict.items():
if self._input_stream_type_info[stream_name] == _PacketDataType.IMAGE:
if data.shape[2] != RGB_CHANNELS:
raise ValueError('Input image must contain three channel rgb data.')
self._graph.add_packet_to_input_stream(
stream=stream_name,
packet=self._make_packet(_PacketDataType.IMAGE,
data).at(self._simulated_timestamp))
else:
# TODO: Support audio data.
raise NotImplementedError(
f'SolutionBase can only process image data. '
f'{self._input_stream_type_info[stream_name].name} '
f'type is not supported yet.')
self._graph.wait_until_idle()
# Create a NamedTuple object where the field names are mapping to the graph
# output stream names.
solution_outputs = collections.namedtuple(
'SolutionOutputs', self._output_stream_type_info.keys())
for stream_name in self._output_stream_type_info.keys():
if stream_name in self._graph_outputs:
setattr(
solution_outputs, stream_name,
self._get_packet_content(self._output_stream_type_info[stream_name],
self._graph_outputs[stream_name]))
else:
setattr(solution_outputs, stream_name, None)
return solution_outputs
def close(self) -> None:
"""Closes all the input sources and the graph."""
self._graph.close()
self._graph = None
self._input_stream_type_info = None
self._output_stream_type_info = None
def _initialize_graph_interface(
self,
validated_graph: validated_graph_config.ValidatedGraphConfig,
side_inputs: Optional[Mapping[str, Any]] = None,
outputs: Optional[List[str]] = None):
"""Gets graph interface type information and returns the canonical graph config proto."""
canonical_graph_config_proto = calculator_pb2.CalculatorGraphConfig()
canonical_graph_config_proto.ParseFromString(validated_graph.binary_config)
# Gets name from a 'TAG:index:name' str.
def get_name(tag_index_name):
return tag_index_name.split(':')[-1]
# Gets the packet type information of the input streams and output streams
# from the validated calculator graph. The mappings from the stream names to
# the packet data types is for deciding which packet creator and getter
# methods to call in the process() method.
def get_stream_packet_type(packet_tag_index_name):
return _PacketDataType.from_registered_name(
validated_graph.registered_stream_type_name(
get_name(packet_tag_index_name)))
self._input_stream_type_info = {
get_name(tag_index_name): get_stream_packet_type(tag_index_name)
for tag_index_name in canonical_graph_config_proto.input_stream
}
if not outputs:
output_streams = canonical_graph_config_proto.output_stream
else:
output_streams = outputs
self._output_stream_type_info = {
get_name(tag_index_name): get_stream_packet_type(tag_index_name)
for tag_index_name in output_streams
}
# Gets the packet type information of the input side packets from the
# validated calculator graph. The mappings from the side packet names to the
# packet data types is for making the input_side_packets dict for graph
# start_run().
def get_side_packet_type(packet_tag_index_name):
return _PacketDataType.from_registered_name(
validated_graph.registered_side_packet_type_name(
get_name(packet_tag_index_name)))
self._side_input_type_info = {
get_name(tag_index_name): get_side_packet_type(tag_index_name)
for tag_index_name, _ in (side_inputs or {}).items()
}
return canonical_graph_config_proto
def _modify_calculator_options(
self, calculator_graph_config: calculator_pb2.CalculatorGraphConfig,
calculator_params: Mapping[str, Any]) -> None:
"""Modifies the CalculatorOptions of the calculators listed in calculator_params."""
# Reorganizes the calculator options field data by calculator name and puts
# all the field data of the same calculator in a list.
def generate_nested_calculator_params(flat_map):
nested_map = {}
for compound_name, field_value in flat_map.items():
calculator_and_field_name = compound_name.split('.')
if len(calculator_and_field_name) != 2:
raise ValueError(
f'The key "{compound_name}" in the calculator_params is invalid.')
calculator_name = calculator_and_field_name[0]
field_name = calculator_and_field_name[1]
if calculator_name in nested_map:
nested_map[calculator_name].append((field_name, field_value))
else:
nested_map[calculator_name] = [(field_name, field_value)]
return nested_map
def modify_options_fields(calculator_options, options_field_list):
for field_name, field_value in options_field_list:
if field_value is None:
calculator_options.ClearField(field_name)
else:
field_label = calculator_options.DESCRIPTOR.fields_by_name[
field_name].label
if field_label is descriptor.FieldDescriptor.LABEL_REPEATED:
if not isinstance(field_value, Iterable):
raise ValueError(
f'{field_name} is a repeated proto field but the value '
f'to be set is {type(field_value)}, which is not iterable.')
# TODO: Support resetting the entire repeated field
# (array-option) and changing the individual values in the repeated
# field (array-element-option).
calculator_options.ClearField(field_name)
for elem in field_value:
getattr(calculator_options, field_name).append(elem)
else:
setattr(calculator_options, field_name, field_value)
nested_calculator_params = generate_nested_calculator_params(
calculator_params)
num_modified = 0
for node in calculator_graph_config.node:
if node.name not in nested_calculator_params:
continue
options_type = CALCULATOR_TO_OPTIONS.get(node.calculator)
if options_type is None:
raise ValueError(
f'Modifying the calculator options of {node.name} is not supported.'
)
options_field_list = nested_calculator_params[node.name]
if node.HasField('options') and node.node_options:
raise ValueError(
f'Cannot modify the calculator options of {node.name} because it '
f'has both options and node_options fields.')
if node.node_options:
# The "node_options" case for the proto3 syntax.
node_options_modified = False
for elem in node.node_options:
type_name = elem.type_url.split('/')[-1]
if type_name == options_type.DESCRIPTOR.full_name:
calculator_options = options_type.FromString(elem.value)
modify_options_fields(calculator_options, options_field_list)
elem.value = calculator_options.SerializeToString()
node_options_modified = True
break
# There is no existing node_options being modified. Add a new
# node_options instead.
if not node_options_modified:
calculator_options = options_type()
modify_options_fields(calculator_options, options_field_list)
node.node_options.add().Pack(calculator_options)
else:
# The "options" case for the proto2 syntax as well as the fallback
# when the calculator doesn't have either "options" or "node_options".
modify_options_fields(node.options.Extensions[options_type.ext],
options_field_list)
num_modified += 1
# Exits the loop early when every elements in nested_calculator_params
# have been visited.
if num_modified == len(nested_calculator_params):
break
def _make_packet(self, packet_data_type: _PacketDataType,
data: Any) -> packet.Packet:
if packet_data_type == _PacketDataType.IMAGE:
return packet_creator.create_image_frame(
data, image_format=image_frame.ImageFormat.SRGB)
else:
return getattr(packet_creator, 'create_' + packet_data_type.value)(data)
def _get_packet_content(self, packet_data_type: _PacketDataType,
output_packet: packet.Packet) -> Any:
if packet_data_type == _PacketDataType.STRING:
return packet_getter.get_str(output_packet)
elif packet_data_type == _PacketDataType.IMAGE:
return packet_getter.get_image_frame(output_packet).numpy_view()
else:
return getattr(packet_getter, 'get_' + packet_data_type.value)(
output_packet)
mediapipe/python/solution_base_test.py
+288
View File
@@ -0,0 +1,288 @@
# Copyright 2020 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.
# Lint as: python3
"""Tests for mediapipe.python.solution_base."""
from absl.testing import absltest
from absl.testing import parameterized
import numpy as np
from google.protobuf import text_format
from mediapipe.framework import calculator_pb2
from mediapipe.framework.formats import detection_pb2
from mediapipe.python import solution_base
CALCULATOR_OPTIONS_TEST_GRAPH_CONFIG = """
input_stream: 'image_in'
output_stream: 'image_out'
node {
name: 'ImageTransformation'
calculator: 'ImageTransformationCalculator'
input_stream: 'IMAGE:image_in'
output_stream: 'IMAGE:image_out'
options: {
[mediapipe.ImageTransformationCalculatorOptions.ext] {
output_width: 10
output_height: 10
}
}
node_options: {
[type.googleapis.com/mediapipe.ImageTransformationCalculatorOptions] {
output_width: 10
output_height: 10
}
}
}
"""
class SolutionBaseTest(parameterized.TestCase):
def test_invalid_initialization_arguments(self):
with self.assertRaisesRegex(
ValueError,
'Must provide exactly one of \'binary_graph_path\' or \'graph_config\'.'
):
solution_base.SolutionBase()
with self.assertRaisesRegex(
ValueError,
'Must provide exactly one of \'binary_graph_path\' or \'graph_config\'.'
):
solution_base.SolutionBase(
graph_config=calculator_pb2.CalculatorGraphConfig(),
binary_graph_path='/tmp/no_such.binarypb')
@parameterized.named_parameters(('no_graph_input_output_stream', """
node {
calculator: 'PassThroughCalculator'
input_stream: 'in'
output_stream: 'out'
}
""", RuntimeError, 'does not have a corresponding output stream.'),
('calcualtor_io_mismatch', """
node {
calculator: 'PassThroughCalculator'
input_stream: 'in'
input_stream: 'in2'
output_stream: 'out'
}
""", ValueError, 'must use matching tags and indexes.'),
('unkown_registered_stream_type_name', """
input_stream: 'in'
output_stream: 'out'
node {
calculator: 'PassThroughCalculator'
input_stream: 'in'
output_stream: 'out'
}
""", RuntimeError, 'Unable to find the type for stream \"in\".'))
def test_invalid_config(self, text_config, error_type, error_message):
config_proto = text_format.Parse(text_config,
calculator_pb2.CalculatorGraphConfig())
with self.assertRaisesRegex(error_type, error_message):
solution_base.SolutionBase(graph_config=config_proto)
def test_invalid_input_data_type(self):
text_config = """
input_stream: 'input_detections'
output_stream: 'output_detections'
node {
calculator: 'DetectionUniqueIdCalculator'
input_stream: 'DETECTIONS:input_detections'
output_stream: 'DETECTIONS:output_detections'
}
"""
config_proto = text_format.Parse(text_config,
calculator_pb2.CalculatorGraphConfig())
solution = solution_base.SolutionBase(graph_config=config_proto)
detection = detection_pb2.Detection()
text_format.Parse('score: 0.5', detection)
with self.assertRaisesRegex(
NotImplementedError,
'SolutionBase can only process image data. PROTO_LIST type is not supported.'
):
solution.process({'input_detections': detection})
def test_invalid_input_image_data(self):
text_config = """
input_stream: 'image_in'
output_stream: 'image_out'
node {
calculator: 'ImageTransformationCalculator'
input_stream: 'IMAGE:image_in'
output_stream: 'IMAGE:transformed_image_in'
}
node {
calculator: 'ImageTransformationCalculator'
input_stream: 'IMAGE:transformed_image_in'
output_stream: 'IMAGE:image_out'
}
"""
config_proto = text_format.Parse(text_config,
calculator_pb2.CalculatorGraphConfig())
solution = solution_base.SolutionBase(graph_config=config_proto)
with self.assertRaisesRegex(
ValueError, 'Input image must contain three channel rgb data.'):
solution.process(np.arange(36, dtype=np.uint8).reshape(3, 3, 4))
@parameterized.named_parameters(('graph_without_side_packets', """
input_stream: 'image_in'
output_stream: 'image_out'
node {
calculator: 'ImageTransformationCalculator'
input_stream: 'IMAGE:image_in'
output_stream: 'IMAGE:transformed_image_in'
}
node {
calculator: 'ImageTransformationCalculator'
input_stream: 'IMAGE:transformed_image_in'
output_stream: 'IMAGE:image_out'
}
""", None), ('graph_with_side_packets', """
input_stream: 'image_in'
input_side_packet: 'allow_signal'
input_side_packet: 'rotation_degrees'
output_stream: 'image_out'
node {
calculator: 'ImageTransformationCalculator'
input_stream: 'IMAGE:image_in'
input_side_packet: 'ROTATION_DEGREES:rotation_degrees'
output_stream: 'IMAGE:transformed_image_in'
}
node {
calculator: 'GateCalculator'
input_stream: 'transformed_image_in'
input_side_packet: 'ALLOW:allow_signal'
output_stream: 'image_out_to_transform'
}
node {
calculator: 'ImageTransformationCalculator'
input_stream: 'IMAGE:image_out_to_transform'
input_side_packet: 'ROTATION_DEGREES:rotation_degrees'
output_stream: 'IMAGE:image_out'
}""", {
'allow_signal': True,
'rotation_degrees': 0
}))
def test_solution_process(self, text_config, side_inputs):
self._process_and_verify(
config_proto=text_format.Parse(text_config,
calculator_pb2.CalculatorGraphConfig()),
side_inputs=side_inputs)
def test_invalid_calculator_options(self):
text_config = """
input_stream: 'image_in'
output_stream: 'image_out'
node {
calculator: 'ImageTransformationCalculator'
input_stream: 'IMAGE:image_in'
output_stream: 'IMAGE:transformed_image_in'
}
node {
name: 'SignalGate'
calculator: 'GateCalculator'
input_stream: 'transformed_image_in'
input_side_packet: 'ALLOW:allow_signal'
output_stream: 'image_out_to_transform'
}
node {
calculator: 'ImageTransformationCalculator'
input_stream: 'IMAGE:image_out_to_transform'
output_stream: 'IMAGE:image_out'
}
"""
config_proto = text_format.Parse(text_config,
calculator_pb2.CalculatorGraphConfig())
with self.assertRaisesRegex(
ValueError,
'Modifying the calculator options of SignalGate is not supported.'):
solution_base.SolutionBase(
graph_config=config_proto,
calculator_params={'SignalGate.invalid_field': 'I am invalid'})
def test_calculator_has_both_options_and_node_options(self):
config_proto = text_format.Parse(CALCULATOR_OPTIONS_TEST_GRAPH_CONFIG,
calculator_pb2.CalculatorGraphConfig())
with self.assertRaisesRegex(ValueError,
'has both options and node_options fields.'):
solution_base.SolutionBase(
graph_config=config_proto,
calculator_params={
'ImageTransformation.output_width': 0,
'ImageTransformation.output_height': 0
})
def test_modifying_calculator_proto2_options(self):
config_proto = text_format.Parse(CALCULATOR_OPTIONS_TEST_GRAPH_CONFIG,
calculator_pb2.CalculatorGraphConfig())
# To test proto2 options only, remove the proto3 node_options field from the
# graph config.
self.assertEqual('ImageTransformation', config_proto.node[0].name)
config_proto.node[0].ClearField('node_options')
self._process_and_verify(
config_proto=config_proto,
calculator_params={
'ImageTransformation.output_width': 0,
'ImageTransformation.output_height': 0
})
def test_modifying_calculator_proto3_node_options(self):
config_proto = text_format.Parse(CALCULATOR_OPTIONS_TEST_GRAPH_CONFIG,
calculator_pb2.CalculatorGraphConfig())
# To test proto3 node options only, remove the proto2 options field from the
# graph config.
self.assertEqual('ImageTransformation', config_proto.node[0].name)
config_proto.node[0].ClearField('options')
self._process_and_verify(
config_proto=config_proto,
calculator_params={
'ImageTransformation.output_width': 0,
'ImageTransformation.output_height': 0
})
def test_adding_calculator_options(self):
config_proto = text_format.Parse(CALCULATOR_OPTIONS_TEST_GRAPH_CONFIG,
calculator_pb2.CalculatorGraphConfig())
# To test a calculator with no options field, remove both proto2 options and
# proto3 node_options fields from the graph config.
self.assertEqual('ImageTransformation', config_proto.node[0].name)
config_proto.node[0].ClearField('options')
config_proto.node[0].ClearField('node_options')
self._process_and_verify(
config_proto=config_proto,
calculator_params={
'ImageTransformation.output_width': 0,
'ImageTransformation.output_height': 0
})
def _process_and_verify(self,
config_proto,
side_inputs=None,
calculator_params=None):
input_image = np.arange(27, dtype=np.uint8).reshape(3, 3, 3)
solution = solution_base.SolutionBase(
graph_config=config_proto,
side_inputs=side_inputs,
calculator_params=calculator_params)
outputs = solution.process(input_image)
self.assertTrue(np.array_equal(input_image, outputs.image_out))
outputs2 = solution.process({'image_in': input_image})
self.assertTrue(np.array_equal(input_image, outputs2.image_out))
solution.close()
if __name__ == '__main__':
absltest.main()
mediapipe/python/solutions/__init__.py
+20
View File
@@ -0,0 +1,20 @@
# Copyright 2020 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.
"""MediaPipe Solutions Python API."""
import mediapipe.python.solutions.drawing_utils
import mediapipe.python.solutions.face_mesh
import mediapipe.python.solutions.hands
import mediapipe.python.solutions.pose
mediapipe/python/solutions/drawing_utils.py
+114
View File
@@ -0,0 +1,114 @@
# Copyright 2020 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.
# Lint as: python3
"""MediaPipe solution drawing utils."""
import math
from typing import List, Tuple, Union
import cv2
import dataclasses
import numpy as np
from mediapipe.framework.formats import landmark_pb2
RGB_CHANNELS = 3
RED_COLOR = (0, 0, 255)
@dataclasses.dataclass
class DrawingSpec:
# Color for drawing the annotation. Default to the green color.
color: Tuple[int, int, int] = (0, 255, 0)
# Thickness for drawing the annotation. Default to 2 pixels.
thickness: int = 2
# Circle radius. Default to 2 pixels.
circle_radius: int = 2
def _normalized_to_pixel_coordinates(
normalized_x: float, normalized_y: float, image_width: int,
image_height: int) -> Union[None, Tuple[int, int]]:
"""Converts normalized value pair to pixel coordinates."""
# Checks if the float value is between 0 and 1.
def is_valid_normalized_value(value: float) -> bool:
return (value > 0 or math.isclose(0, value)) and (value < 1 or
math.isclose(1, value))
if not (is_valid_normalized_value(normalized_x) and
is_valid_normalized_value(normalized_y)):
# TODO: Draw coordinates even if it's outside of the image bounds.
return None
x_px = min(math.floor(normalized_x * image_width), image_width - 1)
y_px = min(math.floor(normalized_y * image_height), image_height - 1)
return x_px, y_px
def draw_landmarks(
image: np.ndarray,
landmark_list: landmark_pb2.NormalizedLandmarkList,
connections: List[Tuple[int, int]] = None,
landmark_drawing_spec: DrawingSpec = DrawingSpec(color=RED_COLOR),
connection_drawing_spec: DrawingSpec = DrawingSpec()):
"""Draws the landmarks and the connections on the image.
Args:
image: A three channel RGB image represented as numpy ndarray.
landmark_list: A normalized landmark list proto message to be annotated on
the image.
connections: A list of landmark index tuples that specifies how landmarks to
be connected in the drawing.
landmark_drawing_spec: A DrawingSpec object that specifies the landmarks'
drawing settings such as color, line thickness, and circle radius.
connection_drawing_spec: A DrawingSpec object that specifies the
connections' drawing settings such as color and line thickness.
Raises:
ValueError: If one of the followings:
a) If the input image is not three channel RGB.
b) If any connetions contain invalid landmark index.
"""
if not landmark_list:
return
if image.shape[2] != RGB_CHANNELS:
raise ValueError('Input image must contain three channel rgb data.')
image_rows, image_cols, _ = image.shape
idx_to_coordinates = {}
for idx, landmark in enumerate(landmark_list.landmark):
if landmark.visibility < 0 or landmark.presence < 0:
continue
landmark_px = _normalized_to_pixel_coordinates(landmark.x, landmark.y,
image_cols, image_rows)
if landmark_px:
idx_to_coordinates[idx] = landmark_px
if connections:
num_landmarks = len(landmark_list.landmark)
# Draws the connections if the start and end landmarks are both visible.
for connection in connections:
start_idx = connection[0]
end_idx = connection[1]
if not (0 <= start_idx < num_landmarks and 0 <= end_idx < num_landmarks):
raise ValueError(f'Landmark index is out of range. Invalid connection '
f'from landmark #{start_idx} to landmark #{end_idx}.')
if start_idx in idx_to_coordinates and end_idx in idx_to_coordinates:
cv2.line(image, idx_to_coordinates[start_idx],
idx_to_coordinates[end_idx], connection_drawing_spec.color,
connection_drawing_spec.thickness)
# Draws landmark points after finishing the connection lines, which is
# aesthetically better.
for landmark_px in idx_to_coordinates.values():
cv2.circle(image, landmark_px, landmark_drawing_spec.circle_radius,
landmark_drawing_spec.color, landmark_drawing_spec.thickness)
mediapipe/python/solutions/drawing_utils_test.py
+144
View File
@@ -0,0 +1,144 @@
# Copyright 2020 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.
# Lint as: python3
"""Tests for mediapipe.python.solutions.drawing_utils."""
from absl.testing import absltest
from absl.testing import parameterized
import cv2
import numpy as np
from google.protobuf import text_format
from mediapipe.framework.formats import landmark_pb2
from mediapipe.python.solutions import drawing_utils
DEFAULT_CONNECTION_DRAWING_SPEC = drawing_utils.DrawingSpec()
DEFAULT_LANDMARK_DRAWING_SPEC = drawing_utils.DrawingSpec(color=(0, 0, 255))
class DrawingUtilTest(parameterized.TestCase):
def test_invalid_input_image(self):
image = np.arange(18, dtype=np.uint8).reshape(3, 3, 2)
with self.assertRaisesRegex(
ValueError, 'Input image must contain three channel rgb data.'):
drawing_utils.draw_landmarks(image, landmark_pb2.NormalizedLandmarkList())
def test_invalid_connection(self):
landmark_list = text_format.Parse(
'landmark {x: 0.5 y: 0.5} landmark {x: 0.2 y: 0.2}',
landmark_pb2.NormalizedLandmarkList())
image = np.arange(27, dtype=np.uint8).reshape(3, 3, 3)
with self.assertRaisesRegex(ValueError, 'Landmark index is out of range.'):
drawing_utils.draw_landmarks(image, landmark_list, [(0, 2)])
@parameterized.named_parameters(
('landmark_list_has_only_one_element', 'landmark {x: 0.1 y: 0.1}'),
('second_landmark_is_invisible',
'landmark {x: 0.1 y: 0.1} landmark {x: 0.5 y: 0.5 visibility: -1.0}'))
def test_draw_single_landmark_point(self, landmark_list_text):
landmark_list = text_format.Parse(landmark_list_text,
landmark_pb2.NormalizedLandmarkList())
image = np.zeros((100, 100, 3), np.uint8)
expected_result = np.copy(image)
cv2.circle(expected_result, (10, 10),
DEFAULT_LANDMARK_DRAWING_SPEC.circle_radius,
DEFAULT_LANDMARK_DRAWING_SPEC.color,
DEFAULT_LANDMARK_DRAWING_SPEC.thickness)
drawing_utils.draw_landmarks(image, landmark_list)
np.testing.assert_array_equal(image, expected_result)
@parameterized.named_parameters(
('landmarks_have_x_and_y_only',
'landmark {x: 0.1 y: 0.5} landmark {x: 0.5 y: 0.1}'),
('landmark_zero_visibility_and_presence',
'landmark {x: 0.1 y: 0.5 presence: 0.0}'
'landmark {x: 0.5 y: 0.1 visibility: 0.0}'))
def test_draw_landmarks_and_connections(self, landmark_list_text):
landmark_list = text_format.Parse(landmark_list_text,
landmark_pb2.NormalizedLandmarkList())
image = np.zeros((100, 100, 3), np.uint8)
expected_result = np.copy(image)
start_point = (10, 50)
end_point = (50, 10)
cv2.line(expected_result, start_point, end_point,
DEFAULT_CONNECTION_DRAWING_SPEC.color,
DEFAULT_CONNECTION_DRAWING_SPEC.thickness)
cv2.circle(expected_result, start_point,
DEFAULT_LANDMARK_DRAWING_SPEC.circle_radius,
DEFAULT_LANDMARK_DRAWING_SPEC.color,
DEFAULT_LANDMARK_DRAWING_SPEC.thickness)
cv2.circle(expected_result, end_point,
DEFAULT_LANDMARK_DRAWING_SPEC.circle_radius,
DEFAULT_LANDMARK_DRAWING_SPEC.color,
DEFAULT_LANDMARK_DRAWING_SPEC.thickness)
drawing_utils.draw_landmarks(
image=image, landmark_list=landmark_list, connections=[(0, 1)])
np.testing.assert_array_equal(image, expected_result)
def test_min_and_max_coordinate_values(self):
landmark_list = text_format.Parse(
'landmark {x: 0.0 y: 1.0}'
'landmark {x: 1.0 y: 0.0}', landmark_pb2.NormalizedLandmarkList())
image = np.zeros((100, 100, 3), np.uint8)
expected_result = np.copy(image)
start_point = (0, 99)
end_point = (99, 0)
cv2.line(expected_result, start_point, end_point,
DEFAULT_CONNECTION_DRAWING_SPEC.color,
DEFAULT_CONNECTION_DRAWING_SPEC.thickness)
cv2.circle(expected_result, start_point,
DEFAULT_LANDMARK_DRAWING_SPEC.circle_radius,
DEFAULT_LANDMARK_DRAWING_SPEC.color,
DEFAULT_LANDMARK_DRAWING_SPEC.thickness)
cv2.circle(expected_result, end_point,
DEFAULT_LANDMARK_DRAWING_SPEC.circle_radius,
DEFAULT_LANDMARK_DRAWING_SPEC.color,
DEFAULT_LANDMARK_DRAWING_SPEC.thickness)
drawing_utils.draw_landmarks(
image=image, landmark_list=landmark_list, connections=[(0, 1)])
np.testing.assert_array_equal(image, expected_result)
def test_drawing_spec(self):
landmark_list = text_format.Parse(
'landmark {x: 0.1 y: 0.1}'
'landmark {x: 0.8 y: 0.8}', landmark_pb2.NormalizedLandmarkList())
image = np.zeros((100, 100, 3), np.uint8)
landmark_drawing_spec = drawing_utils.DrawingSpec(
color=(0, 0, 255), thickness=5)
connection_drawing_spec = drawing_utils.DrawingSpec(
color=(255, 0, 0), thickness=3)
expected_result = np.copy(image)
start_point = (10, 10)
end_point = (80, 80)
cv2.line(expected_result, start_point, end_point,
connection_drawing_spec.color, connection_drawing_spec.thickness)
cv2.circle(expected_result, start_point,
landmark_drawing_spec.circle_radius, landmark_drawing_spec.color,
landmark_drawing_spec.thickness)
cv2.circle(expected_result, end_point, landmark_drawing_spec.circle_radius,
landmark_drawing_spec.color, landmark_drawing_spec.thickness)
drawing_utils.draw_landmarks(
image=image,
landmark_list=landmark_list,
connections=[(0, 1)],
landmark_drawing_spec=landmark_drawing_spec,
connection_drawing_spec=connection_drawing_spec)
np.testing.assert_array_equal(image, expected_result)
if __name__ == '__main__':
absltest.main()
mediapipe/python/solutions/face_mesh.py
+307
View File
@@ -0,0 +1,307 @@
# Copyright 2020 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.
# Lint as: python3
"""MediaPipe FaceMesh."""
from typing import NamedTuple
import numpy as np
# pylint: disable=unused-import
from mediapipe.calculators.core import gate_calculator_pb2
from mediapipe.calculators.core import split_vector_calculator_pb2
from mediapipe.calculators.tensor import inference_calculator_pb2
from mediapipe.calculators.tensor import tensors_to_classification_calculator_pb2
from mediapipe.calculators.tensor import tensors_to_detections_calculator_pb2
from mediapipe.calculators.tensor import tensors_to_landmarks_calculator_pb2
from mediapipe.calculators.tflite import ssd_anchors_calculator_pb2
from mediapipe.calculators.util import association_calculator_pb2
from mediapipe.calculators.util import detections_to_rects_calculator_pb2
from mediapipe.calculators.util import logic_calculator_pb2
from mediapipe.calculators.util import non_max_suppression_calculator_pb2
from mediapipe.calculators.util import rect_transformation_calculator_pb2
from mediapipe.calculators.util import thresholding_calculator_pb2
# pylint: enable=unused-import
from mediapipe.python.solution_base import SolutionBase
BINARYPB_FILE_PATH = 'mediapipe/modules/face_landmark/face_landmark_front_cpu.binarypb'
FACE_CONNECTIONS = frozenset([
# Lips.
(61, 146),
(146, 91),
(91, 181),
(181, 84),
(84, 17),
(17, 314),
(314, 405),
(405, 321),
(321, 375),
(375, 291),
(61, 185),
(185, 40),
(40, 39),
(39, 37),
(37, 0),
(0, 267),
(267, 269),
(269, 270),
(270, 409),
(409, 291),
(78, 95),
(95, 88),
(88, 178),
(178, 87),
(87, 14),
(14, 317),
(317, 402),
(402, 318),
(318, 324),
(324, 308),
(78, 191),
(191, 80),
(80, 81),
(81, 82),
(82, 13),
(13, 312),
(312, 311),
(311, 310),
(310, 415),
(415, 308),
# Left eye.
(33, 7),
(7, 163),
(163, 144),
(144, 145),
(145, 153),
(153, 154),
(154, 155),
(155, 133),
(33, 246),
(246, 161),
(161, 160),
(160, 159),
(159, 158),
(158, 157),
(157, 173),
(173, 133),
# Left eyebrow.
(46, 53),
(53, 52),
(52, 65),
(65, 55),
(70, 63),
(63, 105),
(105, 66),
(66, 107),
# Right eye.
(263, 249),
(249, 390),
(390, 373),
(373, 374),
(374, 380),
(380, 381),
(381, 382),
(382, 362),
(263, 466),
(466, 388),
(388, 387),
(387, 386),
(386, 385),
(385, 384),
(384, 398),
(398, 362),
# Right eyebrow.
(276, 283),
(283, 282),
(282, 295),
(295, 285),
(300, 293),
(293, 334),
(334, 296),
(296, 336),
# Face oval.
(10, 338),
(338, 297),
(297, 332),
(332, 284),
(284, 251),
(251, 389),
(389, 356),
(356, 454),
(454, 323),
(323, 361),
(361, 288),
(288, 397),
(397, 365),
(365, 379),
(379, 378),
(378, 400),
(400, 377),
(377, 152),
(152, 148),
(148, 176),
(176, 149),
(149, 150),
(150, 136),
(136, 172),
(172, 58),
(58, 132),
(132, 93),
(93, 234),
(234, 127),
(127, 162),
(162, 21),
(21, 54),
(54, 103),
(103, 67),
(67, 109),
(109, 10)
])
class FaceMesh(SolutionBase):
"""MediaPipe FaceMesh.
MediaPipe FaceMesh processes an RGB image and returns the face landmarks on
each detected face.
Usage examples:
import cv2
import mediapipe as mp
mp_drawing = mp.solutions.drawing_utils
mp_face_mesh = mp.solutions.face_mesh
# For static images:
face_mesh = mp_face_mesh.FaceMesh(
static_image_mode=True,
max_num_faces=1,
min_detection_confidence=0.5)
drawing_spec = mp_drawing.DrawingSpec(thickness=1, circle_radius=1)
for idx, file in enumerate(file_list):
image = cv2.imread(file)
# Convert the BGR image to RGB before processing.
results = face_mesh.process(cv2.cvtColor(image, cv2.COLOR_BGR2RGB))
# Print and draw face mesh landmarks on the image.
if not results.multi_face_landmarks:
continue
annotated_image = image.copy()
for face_landmarks in results.multi_face_landmarks:
print('face_landmarks:', face_landmarks)
mp_drawing.draw_landmarks(
image=annotated_image,
landmark_list=face_landmarks,
connections=mp_face_mesh.FACE_CONNECTIONS,
landmark_drawing_spec=drawing_spec,
connection_drawing_spec=drawing_spec)
cv2.imwrite('/tmp/annotated_image' + str(idx) + '.png', image)
face_mesh.close()
# For webcam input:
face_mesh = mp_face_mesh.FaceMesh(
min_detection_confidence=0.5, min_tracking_confidence=0.5)
drawing_spec = mp_drawing.DrawingSpec(thickness=1, circle_radius=1)
cap = cv2.VideoCapture(0)
while cap.isOpened():
success, image = cap.read()
if not success:
break
# Flip the image horizontally for a later selfie-view display, and convert
# the BGR image to RGB.
image = cv2.cvtColor(cv2.flip(image, 1), cv2.COLOR_BGR2RGB)
# To improve performance, optionally mark the image as not writeable to
# pass by reference.
image.flags.writeable = False
results = face_mesh.process(image)
# Draw the face mesh annotations on the image.
image.flags.writeable = True
image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
if results.multi_face_landmarks:
for face_landmarks in results.multi_face_landmarks:
mp_drawing.draw_landmarks(
image=image,
landmark_list=face_landmarks,
connections=mp_face_mesh.FACE_CONNECTIONS,
landmark_drawing_spec=drawing_spec,
connection_drawing_spec=drawing_spec)
cv2.imshow('MediaPipe FaceMesh', image)
if cv2.waitKey(5) & 0xFF == 27:
break
face_mesh.close()
cap.release()
"""
def __init__(self,
static_image_mode=False,
max_num_faces=2,
min_detection_confidence=0.5,
min_tracking_confidence=0.5):
"""Initializes a MediaPipe FaceMesh object.
Args:
static_image_mode: If set to False, the solution treats the input images
as a video stream. It will try to detect faces in the first input
images, and upon a successful detection further localizes the face
landmarks. In subsequent images, once all "max_num_faces" faces are
detected and the corresponding face landmarks are localized, it simply
tracks those landmarks without invoking another detection until it loses
track of any of the faces. This reduces latency and is ideal for
processing video frames. If set to True, face detection runs on every
input image, ideal for processing a batch of static, possibly unrelated,
images. Default to False.
max_num_faces: Maximum number of faces to detect. Default to 2.
min_detection_confidence: Minimum confidence value ([0.0, 1.0]) from the
face detection model for the detection to be considered successful.
Default to 0.5.
min_tracking_confidence: Minimum confidence value ([0.0, 1.0]) from the
landmark-tracking model for the face landmarks to be considered tracked
successfully, or otherwise face detection will be invoked automatically
on the next input image. Setting it to a higher value can increase
robustness of the solution, at the expense of a higher latency. Ignored
if "static_image_mode" is True, where face detection simply runs on
every image. Default to 0.5.
"""
super().__init__(
binary_graph_path=BINARYPB_FILE_PATH,
side_inputs={
'num_faces': max_num_faces,
'can_skip_detection': not static_image_mode,
},
calculator_params={
'facedetectionfrontcpu__TensorsToDetectionsCalculator.min_score_thresh':
min_detection_confidence,
'facelandmarkcpu__ThresholdingCalculator.threshold':
min_tracking_confidence,
},
outputs=['multi_face_landmarks'])
def process(self, image: np.ndarray) -> NamedTuple:
"""Processes an RGB image and returns the face landmarks on each detected face.
Args:
image: An RGB image represented as a numpy ndarray.
Raises:
RuntimeError: If the underlying graph occurs any error.
ValueError: If the input image is not three channel RGB.
Returns:
A NamedTuple object with a "multi_face_landmarks" field that contains the
face landmarks on each detected face.
"""
return super().process(input_data={'image': image})
mediapipe/python/solutions/hands.py
+228
View File
@@ -0,0 +1,228 @@
# Copyright 2020 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.
# Lint as: python3
"""MediaPipe Hands."""
import enum
from typing import NamedTuple
import numpy as np
# pylint: disable=unused-import
from mediapipe.calculators.core import gate_calculator_pb2
from mediapipe.calculators.core import split_vector_calculator_pb2
from mediapipe.calculators.tensor import image_to_tensor_calculator_pb2
from mediapipe.calculators.tensor import inference_calculator_pb2
from mediapipe.calculators.tensor import tensors_to_classification_calculator_pb2
from mediapipe.calculators.tensor import tensors_to_detections_calculator_pb2
from mediapipe.calculators.tensor import tensors_to_landmarks_calculator_pb2
from mediapipe.calculators.tflite import ssd_anchors_calculator_pb2
from mediapipe.calculators.util import association_calculator_pb2
from mediapipe.calculators.util import detections_to_rects_calculator_pb2
from mediapipe.calculators.util import logic_calculator_pb2
from mediapipe.calculators.util import non_max_suppression_calculator_pb2
from mediapipe.calculators.util import rect_transformation_calculator_pb2
from mediapipe.calculators.util import thresholding_calculator_pb2
# pylint: enable=unused-import
from mediapipe.python.solution_base import SolutionBase
class HandLandmark(enum.IntEnum):
"""The 21 hand landmarks."""
WRIST = 0
THUMB_CMC = 1
THUMB_MCP = 2
THUMB_IP = 3
THUMB_TIP = 4
INDEX_FINGER_MCP = 5
INDEX_FINGER_PIP = 6
INDEX_FINGER_DIP = 7
INDEX_FINGER_TIP = 8
MIDDLE_FINGER_MCP = 9
MIDDLE_FINGER_PIP = 10
MIDDLE_FINGER_DIP = 11
MIDDLE_FINGER_TIP = 12
RING_FINGER_MCP = 13
RING_FINGER_PIP = 14
RING_FINGER_DIP = 15
RING_FINGER_TIP = 16
PINKY_MCP = 17
PINKY_PIP = 18
PINKY_DIP = 19
PINKY_TIP = 20
BINARYPB_FILE_PATH = 'mediapipe/modules/hand_landmark/hand_landmark_tracking_cpu.binarypb'
HAND_CONNECTIONS = frozenset([
(HandLandmark.WRIST, HandLandmark.THUMB_CMC),
(HandLandmark.THUMB_CMC, HandLandmark.THUMB_MCP),
(HandLandmark.THUMB_MCP, HandLandmark.THUMB_IP),
(HandLandmark.THUMB_IP, HandLandmark.THUMB_TIP),
(HandLandmark.WRIST, HandLandmark.INDEX_FINGER_MCP),
(HandLandmark.INDEX_FINGER_MCP, HandLandmark.INDEX_FINGER_PIP),
(HandLandmark.INDEX_FINGER_PIP, HandLandmark.INDEX_FINGER_DIP),
(HandLandmark.INDEX_FINGER_DIP, HandLandmark.INDEX_FINGER_TIP),
(HandLandmark.INDEX_FINGER_MCP, HandLandmark.MIDDLE_FINGER_MCP),
(HandLandmark.MIDDLE_FINGER_MCP, HandLandmark.MIDDLE_FINGER_PIP),
(HandLandmark.MIDDLE_FINGER_PIP, HandLandmark.MIDDLE_FINGER_DIP),
(HandLandmark.MIDDLE_FINGER_DIP, HandLandmark.MIDDLE_FINGER_TIP),
(HandLandmark.MIDDLE_FINGER_MCP, HandLandmark.RING_FINGER_MCP),
(HandLandmark.RING_FINGER_MCP, HandLandmark.RING_FINGER_PIP),
(HandLandmark.RING_FINGER_PIP, HandLandmark.RING_FINGER_DIP),
(HandLandmark.RING_FINGER_DIP, HandLandmark.RING_FINGER_TIP),
(HandLandmark.RING_FINGER_MCP, HandLandmark.PINKY_MCP),
(HandLandmark.WRIST, HandLandmark.PINKY_MCP),
(HandLandmark.PINKY_MCP, HandLandmark.PINKY_PIP),
(HandLandmark.PINKY_PIP, HandLandmark.PINKY_DIP),
(HandLandmark.PINKY_DIP, HandLandmark.PINKY_TIP)
])
class Hands(SolutionBase):
"""MediaPipe Hands.
MediaPipe Hands processes an RGB image and returns the hand landmarks and
handedness (left v.s. right hand) of each detected hand.
Note that it determines handedness assuming the input image is mirrored,
i.e., taken with a front-facing/selfie camera (
https://en.wikipedia.org/wiki/Front-facing_camera) with images flipped
horizontally. If that is not the case, use, for instance, cv2.flip(image, 1)
to flip the image first for a correct handedness output.
Usage examples:
import cv2
import mediapipe as mp
mp_drawing = mp.solutions.drawing_utils
mp_hands = mp.solutions.hands
# For static images:
hands = mp_hands.Hands(
static_image_mode=True,
max_num_hands=2,
min_detection_confidence=0.7)
for idx, file in enumerate(file_list):
# Read an image, flip it around y-axis for correct handedness output (see
# above).
image = cv2.flip(cv2.imread(file), 1)
# Convert the BGR image to RGB before processing.
results = hands.process(cv2.cvtColor(image, cv2.COLOR_BGR2RGB))
# Print handedness and draw hand landmarks on the image.
print('handedness:', results.multi_handedness)
if not results.multi_hand_landmarks:
continue
annotated_image = image.copy()
for hand_landmarks in results.multi_hand_landmarks:
print('hand_landmarks:', hand_landmarks)
mp_drawing.draw_landmarks(
annotated_image, hand_landmarks, mp_hands.HAND_CONNECTIONS)
cv2.imwrite(
'/tmp/annotated_image' + str(idx) + '.png', cv2.flip(image, 1))
hands.close()
# For webcam input:
hands = mp_hands.Hands(
min_detection_confidence=0.7, min_tracking_confidence=0.5)
cap = cv2.VideoCapture(0)
while cap.isOpened():
success, image = cap.read()
if not success:
break
# Flip the image horizontally for a later selfie-view display, and convert
# the BGR image to RGB.
image = cv2.cvtColor(cv2.flip(image, 1), cv2.COLOR_BGR2RGB)
# To improve performance, optionally mark the image as not writeable to
# pass by reference.
image.flags.writeable = False
results = hands.process(image)
# Draw the hand annotations on the image.
image.flags.writeable = True
image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
if results.multi_hand_landmarks:
for hand_landmarks in results.multi_hand_landmarks:
mp_drawing.draw_landmarks(
image, hand_landmarks, mp_hands.HAND_CONNECTIONS)
cv2.imshow('MediaPipe Hands', image)
if cv2.waitKey(5) & 0xFF == 27:
break
hands.close()
cap.release()
"""
def __init__(self,
static_image_mode=False,
max_num_hands=2,
min_detection_confidence=0.7,
min_tracking_confidence=0.5):
"""Initializes a MediaPipe Hand object.
Args:
static_image_mode: If set to False, the solution treats the input images
as a video stream. It will try to detect hands in the first input
images, and upon a successful detection further localizes the hand
landmarks. In subsequent images, once all "max_num_hands" hands are
detected and the corresponding hand landmarks are localized, it simply
tracks those landmarks without invoking another detection until it loses
track of any of the hands. This reduces latency and is ideal for
processing video frames. If set to True, hand detection runs on every
input image, ideal for processing a batch of static, possibly unrelated,
images. Default to False.
max_num_hands: Maximum number of hands to detect. Default to 2.
min_detection_confidence: Minimum confidence value ([0.0, 1.0]) from the
hand detection model for the detection to be considered successful.
Default to 0.7.
min_tracking_confidence: Minimum confidence value ([0.0, 1.0]) from the
landmark-tracking model for the hand landmarks to be considered tracked
successfully, or otherwise hand detection will be invoked automatically
on the next input image. Setting it to a higher value can increase
robustness of the solution, at the expense of a higher latency. Ignored
if "static_image_mode" is True, where hand detection simply runs on
every image. Default to 0.5.
"""
super().__init__(
binary_graph_path=BINARYPB_FILE_PATH,
side_inputs={
'num_hands': max_num_hands,
'can_skip_detection': not static_image_mode,
},
calculator_params={
'palmdetectioncpu__TensorsToDetectionsCalculator.min_score_thresh':
min_detection_confidence,
'handlandmarkcpu__ThresholdingCalculator.threshold':
min_tracking_confidence,
},
outputs=['multi_hand_landmarks', 'multi_handedness'])
def process(self, image: np.ndarray) -> NamedTuple:
"""Processes an RGB image and returns the hand landmarks and handedness of each detected hand.
Args:
image: An RGB image represented as a numpy ndarray.
Raises:
RuntimeError: If the underlying graph occurs any error.
ValueError: If the input image is not three channel RGB.
Returns:
A NamedTuple object with two fields: a "multi_hand_landmarks" field that
contains the hand landmarks on each detected hand and a "multi_handedness"
field that contains the handedness (left v.s. right hand) of the detected
hand.
"""
return super().process(input_data={'image': image})
mediapipe/python/solutions/pose.py
+213
View File
@@ -0,0 +1,213 @@
# Copyright 2020 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.
# Lint as: python3
"""MediaPipe Pose."""
import enum
from typing import NamedTuple
import numpy as np
# pylint: disable=unused-import
from mediapipe.calculators.core import gate_calculator_pb2
from mediapipe.calculators.core import split_vector_calculator_pb2
from mediapipe.calculators.tensor import image_to_tensor_calculator_pb2
from mediapipe.calculators.tensor import inference_calculator_pb2
from mediapipe.calculators.tensor import tensors_to_classification_calculator_pb2
from mediapipe.calculators.tensor import tensors_to_detections_calculator_pb2
from mediapipe.calculators.tensor import tensors_to_landmarks_calculator_pb2
from mediapipe.calculators.util import detections_to_rects_calculator_pb2
from mediapipe.calculators.util import landmarks_smoothing_calculator_pb2
from mediapipe.calculators.util import logic_calculator_pb2
from mediapipe.calculators.util import non_max_suppression_calculator_pb2
from mediapipe.calculators.util import rect_transformation_calculator_pb2
from mediapipe.calculators.util import thresholding_calculator_pb2
# pylint: enable=unused-import
from mediapipe.python.solution_base import SolutionBase
class PoseLandmark(enum.IntEnum):
"""The 25 (upper-body) pose landmarks."""
NOSE = 0
RIGHT_EYE_INNER = 1
RIGHT_EYE = 2
RIGHT_EYE_OUTER = 3
LEFT_EYE_INNER = 4
LEFT_EYE = 5
LEFT_EYE_OUTER = 6
RIGHT_EAR = 7
LEFT_EAR = 8
MOUTH_RIGHT = 9
MOUTH_LEFT = 10
RIGHT_SHOULDER = 11
LEFT_SHOULDER = 12
RIGHT_ELBOW = 13
LEFT_ELBOW = 14
RIGHT_WRIST = 15
LEFT_WRIST = 16
RIGHT_PINKY = 17
LEFT_PINKY = 18
RIGHT_INDEX = 19
LEFT_INDEX = 20
RIGHT_THUMB = 21
LEFT_THUMB = 22
RIGHT_HIP = 23
LEFT_HIP = 24
BINARYPB_FILE_PATH = 'mediapipe/modules/pose_landmark/pose_landmark_upper_body_smoothed_cpu.binarypb'
POSE_CONNECTIONS = frozenset([
(PoseLandmark.NOSE, PoseLandmark.RIGHT_EYE_INNER),
(PoseLandmark.RIGHT_EYE_INNER, PoseLandmark.RIGHT_EYE),
(PoseLandmark.RIGHT_EYE, PoseLandmark.RIGHT_EYE_OUTER),
(PoseLandmark.RIGHT_EYE_OUTER, PoseLandmark.RIGHT_EAR),
(PoseLandmark.NOSE, PoseLandmark.LEFT_EYE_INNER),
(PoseLandmark.LEFT_EYE_INNER, PoseLandmark.LEFT_EYE),
(PoseLandmark.LEFT_EYE, PoseLandmark.LEFT_EYE_OUTER),
(PoseLandmark.LEFT_EYE_OUTER, PoseLandmark.LEFT_EAR),
(PoseLandmark.MOUTH_RIGHT, PoseLandmark.MOUTH_LEFT),
(PoseLandmark.RIGHT_SHOULDER, PoseLandmark.LEFT_SHOULDER),
(PoseLandmark.RIGHT_SHOULDER, PoseLandmark.RIGHT_ELBOW),
(PoseLandmark.RIGHT_ELBOW, PoseLandmark.RIGHT_WRIST),
(PoseLandmark.RIGHT_WRIST, PoseLandmark.RIGHT_PINKY),
(PoseLandmark.RIGHT_WRIST, PoseLandmark.RIGHT_INDEX),
(PoseLandmark.RIGHT_WRIST, PoseLandmark.RIGHT_THUMB),
(PoseLandmark.RIGHT_PINKY, PoseLandmark.RIGHT_INDEX),
(PoseLandmark.LEFT_SHOULDER, PoseLandmark.LEFT_ELBOW),
(PoseLandmark.LEFT_ELBOW, PoseLandmark.LEFT_WRIST),
(PoseLandmark.LEFT_WRIST, PoseLandmark.LEFT_PINKY),
(PoseLandmark.LEFT_WRIST, PoseLandmark.LEFT_INDEX),
(PoseLandmark.LEFT_WRIST, PoseLandmark.LEFT_THUMB),
(PoseLandmark.LEFT_PINKY, PoseLandmark.LEFT_INDEX),
(PoseLandmark.RIGHT_SHOULDER, PoseLandmark.RIGHT_HIP),
(PoseLandmark.LEFT_SHOULDER, PoseLandmark.LEFT_HIP),
(PoseLandmark.RIGHT_HIP, PoseLandmark.LEFT_HIP)
])
class Pose(SolutionBase):
"""MediaPipe Pose.
MediaPipe Pose processes an RGB image and returns pose landmarks on the most
prominent person detected.
Usage examples:
import cv2
import mediapipe as mp
mp_drawing = mp.solutions.drawing_utils
mp_pose = mp.solutions.pose
# For static images:
pose = mp_pose.Pose(
static_image_mode=True, min_detection_confidence=0.5)
for idx, file in enumerate(file_list):
image = cv2.imread(file)
# Convert the BGR image to RGB before processing.
results = pose.process(cv2.cvtColor(image, cv2.COLOR_BGR2RGB))
# Print and draw pose landmarks on the image.
print(
'nose landmark:',
results.pose_landmarks.landmark[mp_pose.PoseLandmark.NOSE])
annotated_image = image.copy()
mp_drawing.draw_landmarks(
annotated_image, results.pose_landmarks, mp_pose.POSE_CONNECTIONS)
cv2.imwrite('/tmp/annotated_image' + str(idx) + '.png', image)
pose.close()
# For webcam input:
pose = mp_pose.Pose(
min_detection_confidence=0.5, min_tracking_confidence=0.5)
cap = cv2.VideoCapture(0)
while cap.isOpened():
success, image = cap.read()
if not success:
break
# Flip the image horizontally for a later selfie-view display, and convert
# the BGR image to RGB.
image = cv2.cvtColor(cv2.flip(image, 1), cv2.COLOR_BGR2RGB)
# To improve performance, optionally mark the image as not writeable to
# pass by reference.
image.flags.writeable = False
results = pose.process(image)
# Draw the pose annotation on the image.
image.flags.writeable = True
image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
mp_drawing.draw_landmarks(
image, results.pose_landmarks, mp_pose.POSE_CONNECTIONS)
cv2.imshow('MediaPipe Pose', image)
if cv2.waitKey(5) & 0xFF == 27:
break
pose.close()
cap.release()
"""
def __init__(self,
static_image_mode=False,
min_detection_confidence=0.5,
min_tracking_confidence=0.5):
"""Initializes a MediaPipe Pose object.
Args:
static_image_mode: If set to False, the solution treats the input images
as a video stream. It will try to detect the most prominent person in
the very first images, and upon a successful detection further localizes
the pose landmarks. In subsequent images, it then simply tracks those
landmarks without invoking another detection until it loses track, on
reducing computation and latency. If set to True, person detection runs
every input image, ideal for processing a batch of static, possibly
unrelated, images. Default to False.
min_detection_confidence: Minimum confidence value ([0.0, 1.0]) from the
person-detection model for the detection to be considered successful.
Default to 0.5.
min_tracking_confidence: Minimum confidence value ([0.0, 1.0]) from the
landmark-tracking model for the pose landmarks to be considered tracked
successfully, or otherwise person detection will be invoked
automatically on the next input image. Setting it to a higher value can
increase robustness of the solution, at the expense of a higher latency.
Ignored if "static_image_mode" is True, where person detection simply
runs on every image. Default to 0.5.
"""
super().__init__(
binary_graph_path=BINARYPB_FILE_PATH,
side_inputs={
'can_skip_detection': not static_image_mode,
},
calculator_params={
'poselandmarkupperbodycpu__posedetectioncpu__TensorsToDetectionsCalculator.min_score_thresh':
min_detection_confidence,
'poselandmarkupperbodycpu__poselandmarkupperbodybyroicpu__ThresholdingCalculator.threshold':
min_tracking_confidence,
},
outputs=['pose_landmarks'])
def process(self, image: np.ndarray) -> NamedTuple:
"""Processes an RGB image and returns the pose landmarks on the most prominent person detected.
Args:
image: An RGB image represented as a numpy ndarray.
Raises:
RuntimeError: If the underlying graph occurs any error.
ValueError: If the input image is not three channel RGB.
Returns:
A NamedTuple object with a "pose_landmarks" field that contains the pose
landmarks on the most prominent person detected.
"""
return super().process(input_data={'image': image})