mediapipe-nightly 0.10.21.post20250114__cp312-cp312-manylinux_2_28_x86_64.whl

mediapipe/tasks/python/vision/core/__init__.py

@@ -0,0 +1,14 @@
+"""Copyright 2022 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/tasks/python/vision/core/base_vision_task_api.py

@@ -0,0 +1,226 @@
+# Copyright 2022 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 vision task base api."""
+
+import math
+from typing import Callable, Mapping, Optional
+
+from mediapipe.framework import calculator_pb2
+from mediapipe.python._framework_bindings import image as image_module
+from mediapipe.python._framework_bindings import packet as packet_module
+from mediapipe.python._framework_bindings import task_runner as task_runner_module
+from mediapipe.tasks.python.components.containers import rect as rect_module
+from mediapipe.tasks.python.core.optional_dependencies import doc_controls
+from mediapipe.tasks.python.vision.core import image_processing_options as image_processing_options_module
+from mediapipe.tasks.python.vision.core import vision_task_running_mode as running_mode_module
+
+_TaskRunner = task_runner_module.TaskRunner
+_Packet = packet_module.Packet
+_NormalizedRect = rect_module.NormalizedRect
+_RunningMode = running_mode_module.VisionTaskRunningMode
+_ImageProcessingOptions = image_processing_options_module.ImageProcessingOptions
+
+
+@doc_controls.do_not_generate_docs
+class BaseVisionTaskApi(object):
+  """The base class of the user-facing mediapipe vision task api classes."""
+
+  def __init__(
+      self,
+      graph_config: calculator_pb2.CalculatorGraphConfig,
+      running_mode: _RunningMode,
+      packet_callback: Optional[
+          Callable[[Mapping[str, packet_module.Packet]], None]
+      ] = None,
+  ) -> None:
+    """Initializes the `BaseVisionTaskApi` object.
+
+    Args:
+      graph_config: The mediapipe vision task graph config proto.
+      running_mode: The running mode of the mediapipe vision task.
+      packet_callback: The optional packet callback for getting results
+        asynchronously in the live stream mode.
+
+    Raises:
+      ValueError: The packet callback is not properly set based on the task's
+      running mode.
+    """
+    if running_mode == _RunningMode.LIVE_STREAM:
+      if packet_callback is None:
+        raise ValueError(
+            'The vision task is in live stream mode, a user-defined result '
+            'callback must be provided.'
+        )
+    elif packet_callback:
+      raise ValueError(
+          'The vision task is in image or video mode, a user-defined result '
+          'callback should not be provided.'
+      )
+    self._runner = _TaskRunner.create(graph_config, packet_callback)
+    self._running_mode = running_mode
+
+  def _process_image_data(
+      self, inputs: Mapping[str, _Packet]
+  ) -> Mapping[str, _Packet]:
+    """A synchronous method to process single image inputs.
+
+    The call blocks the current thread until a failure status or a successful
+    result is returned.
+
+    Args:
+      inputs: A dict contains (input stream name, data packet) pairs.
+
+    Returns:
+      A dict contains (output stream name, data packet) pairs.
+
+    Raises:
+      ValueError: If the task's running mode is not set to image mode.
+    """
+    if self._running_mode != _RunningMode.IMAGE:
+      raise ValueError(
+          'Task is not initialized with the image mode. Current running mode:'
+          + self._running_mode.name
+      )
+    return self._runner.process(inputs)
+
+  def _process_video_data(
+      self, inputs: Mapping[str, _Packet]
+  ) -> Mapping[str, _Packet]:
+    """A synchronous method to process continuous video frames.
+
+    The call blocks the current thread until a failure status or a successful
+    result is returned.
+
+    Args:
+      inputs: A dict contains (input stream name, data packet) pairs.
+
+    Returns:
+      A dict contains (output stream name, data packet) pairs.
+
+    Raises:
+      ValueError: If the task's running mode is not set to the video mode.
+    """
+    if self._running_mode != _RunningMode.VIDEO:
+      raise ValueError(
+          'Task is not initialized with the video mode. Current running mode:'
+          + self._running_mode.name
+      )
+    return self._runner.process(inputs)
+
+  def _send_live_stream_data(self, inputs: Mapping[str, _Packet]) -> None:
+    """An asynchronous method to send live stream data to the runner.
+
+    The results will be available in the user-defined results callback.
+
+    Args:
+      inputs: A dict contains (input stream name, data packet) pairs.
+
+    Raises:
+      ValueError: If the task's running mode is not set to the live stream
+      mode.
+    """
+    if self._running_mode != _RunningMode.LIVE_STREAM:
+      raise ValueError(
+          'Task is not initialized with the live stream mode. Current running'
+          ' mode:'
+          + self._running_mode.name
+      )
+    self._runner.send(inputs)
+
+  def convert_to_normalized_rect(
+      self,
+      options: _ImageProcessingOptions,
+      image: image_module.Image,
+      roi_allowed: bool = True,
+  ) -> _NormalizedRect:
+    """Converts from ImageProcessingOptions to NormalizedRect, performing sanity checks on-the-fly.
+
+    If the input ImageProcessingOptions is not present, returns a default
+    NormalizedRect covering the whole image with rotation set to 0. If
+    'roi_allowed' is false, an error will be returned if the input
+    ImageProcessingOptions has its 'region_of_interest' field set.
+
+    Args:
+      options: Options for image processing.
+      image: The image to process.
+      roi_allowed: Indicates if the `region_of_interest` field is allowed to be
+        set. By default, it's set to True.
+
+    Returns:
+      A normalized rect proto that represents the image processing options.
+    """
+    normalized_rect = _NormalizedRect(
+        rotation=0, x_center=0.5, y_center=0.5, width=1, height=1
+    )
+    if options is None:
+      return normalized_rect
+
+    if options.rotation_degrees % 90 != 0:
+      raise ValueError('Expected rotation to be a multiple of 90°.')
+
+    # Convert to radians counter-clockwise.
+    normalized_rect.rotation = -options.rotation_degrees * math.pi / 180.0
+
+    if options.region_of_interest:
+      if not roi_allowed:
+        raise ValueError("This task doesn't support region-of-interest.")
+      roi = options.region_of_interest
+      if roi.left >= roi.right or roi.top >= roi.bottom:
+        raise ValueError('Expected Rect with left < right and top < bottom.')
+      if roi.left < 0 or roi.top < 0 or roi.right > 1 or roi.bottom > 1:
+        raise ValueError('Expected Rect values to be in [0,1].')
+      normalized_rect.x_center = (roi.left + roi.right) / 2.0
+      normalized_rect.y_center = (roi.top + roi.bottom) / 2.0
+      normalized_rect.width = roi.right - roi.left
+      normalized_rect.height = roi.bottom - roi.top
+
+    # For 90° and 270° rotations, we need to swap width and height.
+    # This is due to the internal behavior of ImageToTensorCalculator, which:
+    # - first denormalizes the provided rect by multiplying the rect width or
+    #   height by the image width or height, respectively.
+    # - then rotates this by denormalized rect by the provided rotation, and
+    #   uses this for cropping,
+    # - then finally rotates this back.
+    if abs(options.rotation_degrees % 180) != 0:
+      w = normalized_rect.height * image.height / image.width
+      h = normalized_rect.width * image.width / image.height
+      normalized_rect.width = w
+      normalized_rect.height = h
+
+    return normalized_rect
+
+  def close(self) -> None:
+    """Shuts down the mediapipe vision task instance.
+
+    Raises:
+      RuntimeError: If the mediapipe vision task failed to close.
+    """
+    self._runner.close()
+
+  def get_graph_config(self) -> calculator_pb2.CalculatorGraphConfig:
+    """Returns the canonicalized CalculatorGraphConfig of the underlying graph.
+    """
+    return self._runner.get_graph_config()
+
+  def __enter__(self):
+    """Return `self` upon entering the runtime context."""
+    return self
+
+  def __exit__(self, unused_exc_type, unused_exc_value, unused_traceback):
+    """Shuts down the mediapipe vision task instance on exit of the context manager.
+
+    Raises:
+      RuntimeError: If the mediapipe vision task failed to close.
+    """
+    self.close()

mediapipe/tasks/python/vision/core/image_processing_options.py

@@ -0,0 +1,39 @@
+# Copyright 2022 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 vision options for image processing."""
+
+import dataclasses
+from typing import Optional
+
+from mediapipe.tasks.python.components.containers import rect as rect_module
+
+
+@dataclasses.dataclass
+class ImageProcessingOptions:
+  """Options for image processing.
+
+  If both region-of-interest and rotation are specified, the crop around the
+  region-of-interest is extracted first, then the specified rotation is applied
+   to the crop.
+
+  Attributes:
+    region_of_interest: The optional region-of-interest to crop from the image.
+      If not specified, the full image is used. Coordinates must be in [0,1]
+      with 'left' < 'right' and 'top' < 'bottom'.
+    rotation_degrees: The rotation to apply to the image (or cropped
+      region-of-interest), in degrees clockwise. The rotation must be a multiple
+      (positive or negative) of 90°.
+  """
+  region_of_interest: Optional[rect_module.Rect] = None
+  rotation_degrees: int = 0

mediapipe/tasks/python/vision/core/vision_task_running_mode.py

@@ -0,0 +1,31 @@
+# Copyright 2022 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.
+"""The running mode of MediaPipe Vision Tasks."""
+
+import enum
+
+
+class VisionTaskRunningMode(enum.Enum):
+  """MediaPipe vision task running mode.
+
+  Attributes:
+    IMAGE: The mode for running a mediapipe vision task on single image inputs.
+    VIDEO: The mode for running a mediapipe vision task on the decoded frames
+      of an input video.
+    LIVE_STREAM: The mode for running a mediapipe vision task on a live stream
+      of input data, such as from camera.
+  """
+  IMAGE = 'IMAGE'
+  VIDEO = 'VIDEO'
+  LIVE_STREAM = 'LIVE_STREAM'

mediapipe/tasks/python/vision/face_aligner.py

@@ -0,0 +1,158 @@
+# 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.
+"""MediaPipe face aligner task."""
+
+import dataclasses
+from typing import Optional
+
+from mediapipe.python import packet_creator
+from mediapipe.python import packet_getter
+from mediapipe.python._framework_bindings import image as image_module
+from mediapipe.tasks.cc.vision.face_stylizer.proto import face_stylizer_graph_options_pb2
+from mediapipe.tasks.python.core import base_options as base_options_module
+from mediapipe.tasks.python.core import task_info as task_info_module
+from mediapipe.tasks.python.core.optional_dependencies import doc_controls
+from mediapipe.tasks.python.vision.core import base_vision_task_api
+from mediapipe.tasks.python.vision.core import image_processing_options as image_processing_options_module
+from mediapipe.tasks.python.vision.core import vision_task_running_mode as running_mode_module
+
+_BaseOptions = base_options_module.BaseOptions
+_FaceStylizerGraphOptionsProto = (
+    face_stylizer_graph_options_pb2.FaceStylizerGraphOptions
+)
+_RunningMode = running_mode_module.VisionTaskRunningMode
+_ImageProcessingOptions = image_processing_options_module.ImageProcessingOptions
+_TaskInfo = task_info_module.TaskInfo
+
+_FACE_ALIGNMENT_IMAGE_NAME = 'face_alignment'
+_FACE_ALIGNMENT_IMAGE_TAG = 'FACE_ALIGNMENT'
+_NORM_RECT_STREAM_NAME = 'norm_rect_in'
+_NORM_RECT_TAG = 'NORM_RECT'
+_IMAGE_IN_STREAM_NAME = 'image_in'
+_IMAGE_OUT_STREAM_NAME = 'image_out'
+_IMAGE_TAG = 'IMAGE'
+_TASK_GRAPH_NAME = 'mediapipe.tasks.vision.face_stylizer.FaceStylizerGraph'
+
+
+@dataclasses.dataclass
+class FaceAlignerOptions:
+  """Options for the face aligner task.
+
+  Attributes:
+    base_options: Base options for the face aligner task.
+  """
+
+  base_options: _BaseOptions
+
+  @doc_controls.do_not_generate_docs
+  def to_pb2(self) -> _FaceStylizerGraphOptionsProto:
+    """Generates a FaceStylizerOptions protobuf object."""
+    base_options_proto = self.base_options.to_pb2()
+    base_options_proto.use_stream_mode = False
+    return _FaceStylizerGraphOptionsProto(base_options=base_options_proto)
+
+
+class FaceAligner(base_vision_task_api.BaseVisionTaskApi):
+  """Class that performs face alignment on images."""
+
+  @classmethod
+  def create_from_model_path(cls, model_path: str) -> 'FaceAligner':
+    """Creates a `FaceAligner` object from a face landmarker task bundle and the default `FaceAlignerOptions`.
+
+    Note that the created `FaceAligner` instance is in image mode, for
+    aligning one face on a single image input.
+
+    Args:
+      model_path: Path to the face landmarker task bundle.
+
+    Returns:
+      `FaceAligner` object that's created from the model file and the default
+      `FaceAlignerOptions`.
+
+    Raises:
+      ValueError: If failed to create `FaceAligner` object from the provided
+        file such as invalid file path.
+      RuntimeError: If other types of error occurred.
+    """
+    base_options = _BaseOptions(model_asset_path=model_path)
+    options = FaceAlignerOptions(base_options=base_options)
+    return cls.create_from_options(options)
+
+  @classmethod
+  def create_from_options(cls, options: FaceAlignerOptions) -> 'FaceAligner':
+    """Creates the `FaceAligner` object from face aligner options.
+
+    Args:
+      options: Options for the face aligner task.
+
+    Returns:
+      `FaceAligner` object that's created from `options`.
+
+    Raises:
+      ValueError: If failed to create `FaceAligner` object from
+        `FaceAlignerOptions` such as missing the model.
+      RuntimeError: If other types of error occurred.
+    """
+    task_info = _TaskInfo(
+        task_graph=_TASK_GRAPH_NAME,
+        input_streams=[
+            ':'.join([_IMAGE_TAG, _IMAGE_IN_STREAM_NAME]),
+            ':'.join([_NORM_RECT_TAG, _NORM_RECT_STREAM_NAME]),
+        ],
+        output_streams=[
+            ':'.join([_FACE_ALIGNMENT_IMAGE_TAG, _FACE_ALIGNMENT_IMAGE_NAME]),
+            ':'.join([_IMAGE_TAG, _IMAGE_OUT_STREAM_NAME]),
+        ],
+        task_options=options,
+    )
+    return cls(
+        task_info.generate_graph_config(enable_flow_limiting=False),
+        _RunningMode.IMAGE,
+        None,
+    )
+
+  def align(
+      self,
+      image: image_module.Image,
+      image_processing_options: Optional[_ImageProcessingOptions] = None,
+  ) -> image_module.Image:
+    """Performs face alignment on the provided MediaPipe Image.
+
+    Only use this method when the FaceAligner is created with the image
+    running mode.
+
+    Args:
+      image: MediaPipe Image.
+      image_processing_options: Options for image processing.
+
+    Returns:
+      The aligned face image. The aligned output image size is the same as the
+      model output size. None if no face is detected on the input image.
+
+    Raises:
+      ValueError: If any of the input arguments is invalid.
+      RuntimeError: If face alignment failed to run.
+    """
+    normalized_rect = self.convert_to_normalized_rect(
+        image_processing_options, image
+    )
+    output_packets = self._process_image_data({
+        _IMAGE_IN_STREAM_NAME: packet_creator.create_image(image),
+        _NORM_RECT_STREAM_NAME: packet_creator.create_proto(
+            normalized_rect.to_pb2()
+        ),
+    })
+    if output_packets[_FACE_ALIGNMENT_IMAGE_NAME].is_empty():
+      return None
+    return packet_getter.get_image(output_packets[_FACE_ALIGNMENT_IMAGE_NAME])