Merge pull request #3829 from kinaryml:hand-landmarker-python

PiperOrigin-RevId: 487281485

mediapipe/tasks/python/components/containers/landmark_detection_result.py

@@ -53,7 +53,13 @@ class LandmarksDetectionResult:
   def to_pb2(self) -> _LandmarksDetectionResultProto:
     """Generates a LandmarksDetectionResult protobuf object."""
 
+    landmarks = _NormalizedLandmarkListProto()
     classifications = _ClassificationListProto()
+    world_landmarks = _LandmarkListProto()
+
+    for landmark in self.landmarks:
+      landmarks.landmark.append(landmark.to_pb2())
+
     for category in self.categories:
       classifications.classification.append(
           _ClassificationProto(
@@ -63,9 +69,9 @@ class LandmarksDetectionResult:
               display_name=category.display_name))
 
     return _LandmarksDetectionResultProto(
-        landmarks=_NormalizedLandmarkListProto(self.landmarks),
+        landmarks=landmarks,
         classifications=classifications,
-        world_landmarks=_LandmarkListProto(self.world_landmarks),
+        world_landmarks=world_landmarks,
         rect=self.rect.to_pb2())
 
   @classmethod
@@ -73,9 +79,11 @@ class LandmarksDetectionResult:
   def create_from_pb2(
       cls,
       pb2_obj: _LandmarksDetectionResultProto) -> 'LandmarksDetectionResult':
-    """Creates a `LandmarksDetectionResult` object from the given protobuf object.
+    """Creates a `LandmarksDetectionResult` object from the given protobuf object."""
-    """
     categories = []
+    landmarks = []
+    world_landmarks = []
+
     for classification in pb2_obj.classifications.classification:
       categories.append(
           category_module.Category(
@@ -83,14 +91,14 @@ class LandmarksDetectionResult:
               index=classification.index,
               category_name=classification.label,
               display_name=classification.display_name))
+
+    for landmark in pb2_obj.landmarks.landmark:
+      landmarks.append(_NormalizedLandmark.create_from_pb2(landmark))
+
+    for landmark in pb2_obj.world_landmarks.landmark:
+      world_landmarks.append(_Landmark.create_from_pb2(landmark))
     return LandmarksDetectionResult(
-        landmarks=[
+        landmarks=landmarks,
-            _NormalizedLandmark.create_from_pb2(landmark)
-            for landmark in pb2_obj.landmarks.landmark
-        ],
         categories=categories,
-        world_landmarks=[
+        world_landmarks=world_landmarks,
-            _Landmark.create_from_pb2(landmark)
-            for landmark in pb2_obj.world_landmarks.landmark
-        ],
         rect=_NormalizedRect.create_from_pb2(pb2_obj.rect))

mediapipe/tasks/python/test/vision/BUILD

@@ -73,3 +73,26 @@ py_test(
         "//mediapipe/tasks/python/vision/core:vision_task_running_mode",
     ],
 )
+
+py_test(
+    name = "hand_landmarker_test",
+    srcs = ["hand_landmarker_test.py"],
+    data = [
+        "//mediapipe/tasks/testdata/vision:test_images",
+        "//mediapipe/tasks/testdata/vision:test_models",
+        "//mediapipe/tasks/testdata/vision:test_protos",
+    ],
+    deps = [
+        "//mediapipe/python:_framework_bindings",
+        "//mediapipe/tasks/cc/components/containers/proto:landmarks_detection_result_py_pb2",
+        "//mediapipe/tasks/python/components/containers:landmark",
+        "//mediapipe/tasks/python/components/containers:landmark_detection_result",
+        "//mediapipe/tasks/python/components/containers:rect",
+        "//mediapipe/tasks/python/core:base_options",
+        "//mediapipe/tasks/python/test:test_utils",
+        "//mediapipe/tasks/python/vision:hand_landmarker",
+        "//mediapipe/tasks/python/vision/core:image_processing_options",
+        "//mediapipe/tasks/python/vision/core:vision_task_running_mode",
+        "@com_google_protobuf//:protobuf_python",
+    ],
+)

mediapipe/tasks/python/test/vision/hand_landmarker_test.py

@@ -0,0 +1,428 @@
+# Copyright 2022 The MediaPipe Authors. All Rights Reserved.
+#
+# 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.
+"""Tests for hand landmarker."""
+
+import enum
+from unittest import mock
+
+from absl.testing import absltest
+from absl.testing import parameterized
+import numpy as np
+
+from google.protobuf import text_format
+from mediapipe.python._framework_bindings import image as image_module
+from mediapipe.tasks.cc.components.containers.proto import landmarks_detection_result_pb2
+from mediapipe.tasks.python.components.containers import landmark as landmark_module
+from mediapipe.tasks.python.components.containers import landmark_detection_result as landmark_detection_result_module
+from mediapipe.tasks.python.components.containers import rect as rect_module
+from mediapipe.tasks.python.core import base_options as base_options_module
+from mediapipe.tasks.python.test import test_utils
+from mediapipe.tasks.python.vision import hand_landmarker
+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
+
+_LandmarksDetectionResultProto = landmarks_detection_result_pb2.LandmarksDetectionResult
+_BaseOptions = base_options_module.BaseOptions
+_Rect = rect_module.Rect
+_Landmark = landmark_module.Landmark
+_NormalizedLandmark = landmark_module.NormalizedLandmark
+_LandmarksDetectionResult = landmark_detection_result_module.LandmarksDetectionResult
+_Image = image_module.Image
+_HandLandmarker = hand_landmarker.HandLandmarker
+_HandLandmarkerOptions = hand_landmarker.HandLandmarkerOptions
+_HandLandmarkerResult = hand_landmarker.HandLandmarkerResult
+_RUNNING_MODE = running_mode_module.VisionTaskRunningMode
+_ImageProcessingOptions = image_processing_options_module.ImageProcessingOptions
+
+_HAND_LANDMARKER_BUNDLE_ASSET_FILE = 'hand_landmarker.task'
+_NO_HANDS_IMAGE = 'cats_and_dogs.jpg'
+_TWO_HANDS_IMAGE = 'right_hands.jpg'
+_THUMB_UP_IMAGE = 'thumb_up.jpg'
+_THUMB_UP_LANDMARKS = 'thumb_up_landmarks.pbtxt'
+_POINTING_UP_IMAGE = 'pointing_up.jpg'
+_POINTING_UP_LANDMARKS = 'pointing_up_landmarks.pbtxt'
+_POINTING_UP_ROTATED_IMAGE = 'pointing_up_rotated.jpg'
+_POINTING_UP_ROTATED_LANDMARKS = 'pointing_up_rotated_landmarks.pbtxt'
+_LANDMARKS_ERROR_TOLERANCE = 0.03
+_HANDEDNESS_MARGIN = 0.05
+
+
+def _get_expected_hand_landmarker_result(
+    file_path: str) -> _HandLandmarkerResult:
+  landmarks_detection_result_file_path = test_utils.get_test_data_path(
+      file_path)
+  with open(landmarks_detection_result_file_path, 'rb') as f:
+    landmarks_detection_result_proto = _LandmarksDetectionResultProto()
+    # Use this if a .pb file is available.
+    # landmarks_detection_result_proto.ParseFromString(f.read())
+    text_format.Parse(f.read(), landmarks_detection_result_proto)
+    landmarks_detection_result = _LandmarksDetectionResult.create_from_pb2(
+        landmarks_detection_result_proto)
+  return _HandLandmarkerResult(
+      handedness=[landmarks_detection_result.categories],
+      hand_landmarks=[landmarks_detection_result.landmarks],
+      hand_world_landmarks=[landmarks_detection_result.world_landmarks])
+
+
+class ModelFileType(enum.Enum):
+  FILE_CONTENT = 1
+  FILE_NAME = 2
+
+
+class HandLandmarkerTest(parameterized.TestCase):
+
+  def setUp(self):
+    super().setUp()
+    self.test_image = _Image.create_from_file(
+        test_utils.get_test_data_path(_THUMB_UP_IMAGE))
+    self.model_path = test_utils.get_test_data_path(
+        _HAND_LANDMARKER_BUNDLE_ASSET_FILE)
+
+  def _assert_actual_result_approximately_matches_expected_result(
+      self, actual_result: _HandLandmarkerResult,
+      expected_result: _HandLandmarkerResult):
+    # Expects to have the same number of hands detected.
+    self.assertLen(actual_result.hand_landmarks,
+                   len(expected_result.hand_landmarks))
+    self.assertLen(actual_result.hand_world_landmarks,
+                   len(expected_result.hand_world_landmarks))
+    self.assertLen(actual_result.handedness, len(expected_result.handedness))
+    # Actual landmarks match expected landmarks.
+    self.assertLen(actual_result.hand_landmarks[0],
+                   len(expected_result.hand_landmarks[0]))
+    actual_landmarks = actual_result.hand_landmarks[0]
+    expected_landmarks = expected_result.hand_landmarks[0]
+    for i, rename_me in enumerate(actual_landmarks):
+      self.assertAlmostEqual(
+          rename_me.x,
+          expected_landmarks[i].x,
+          delta=_LANDMARKS_ERROR_TOLERANCE)
+      self.assertAlmostEqual(
+          rename_me.y,
+          expected_landmarks[i].y,
+          delta=_LANDMARKS_ERROR_TOLERANCE)
+    # Actual handedness matches expected handedness.
+    actual_top_handedness = actual_result.handedness[0][0]
+    expected_top_handedness = expected_result.handedness[0][0]
+    self.assertEqual(actual_top_handedness.index, expected_top_handedness.index)
+    self.assertEqual(actual_top_handedness.category_name,
+                     expected_top_handedness.category_name)
+    self.assertAlmostEqual(
+        actual_top_handedness.score,
+        expected_top_handedness.score,
+        delta=_HANDEDNESS_MARGIN)
+
+  def test_create_from_file_succeeds_with_valid_model_path(self):
+    # Creates with default option and valid model file successfully.
+    with _HandLandmarker.create_from_model_path(self.model_path) as landmarker:
+      self.assertIsInstance(landmarker, _HandLandmarker)
+
+  def test_create_from_options_succeeds_with_valid_model_path(self):
+    # Creates with options containing model file successfully.
+    base_options = _BaseOptions(model_asset_path=self.model_path)
+    options = _HandLandmarkerOptions(base_options=base_options)
+    with _HandLandmarker.create_from_options(options) as landmarker:
+      self.assertIsInstance(landmarker, _HandLandmarker)
+
+  def test_create_from_options_fails_with_invalid_model_path(self):
+    # Invalid empty model path.
+    with self.assertRaisesRegex(
+        RuntimeError, 'Unable to open file at /path/to/invalid/model.tflite'):
+      base_options = _BaseOptions(
+          model_asset_path='/path/to/invalid/model.tflite')
+      options = _HandLandmarkerOptions(base_options=base_options)
+      _HandLandmarker.create_from_options(options)
+
+  def test_create_from_options_succeeds_with_valid_model_content(self):
+    # Creates with options containing model content successfully.
+    with open(self.model_path, 'rb') as f:
+      base_options = _BaseOptions(model_asset_buffer=f.read())
+      options = _HandLandmarkerOptions(base_options=base_options)
+      landmarker = _HandLandmarker.create_from_options(options)
+      self.assertIsInstance(landmarker, _HandLandmarker)
+
+  @parameterized.parameters(
+      (ModelFileType.FILE_NAME,
+       _get_expected_hand_landmarker_result(_THUMB_UP_LANDMARKS)),
+      (ModelFileType.FILE_CONTENT,
+       _get_expected_hand_landmarker_result(_THUMB_UP_LANDMARKS)))
+  def test_detect(self, model_file_type, expected_detection_result):
+    # Creates hand landmarker.
+    if model_file_type is ModelFileType.FILE_NAME:
+      base_options = _BaseOptions(model_asset_path=self.model_path)
+    elif model_file_type is ModelFileType.FILE_CONTENT:
+      with open(self.model_path, 'rb') as f:
+        model_content = f.read()
+      base_options = _BaseOptions(model_asset_buffer=model_content)
+    else:
+      # Should never happen
+      raise ValueError('model_file_type is invalid.')
+
+    options = _HandLandmarkerOptions(base_options=base_options)
+    landmarker = _HandLandmarker.create_from_options(options)
+
+    # Performs hand landmarks detection on the input.
+    detection_result = landmarker.detect(self.test_image)
+    # Comparing results.
+    self._assert_actual_result_approximately_matches_expected_result(
+        detection_result, expected_detection_result)
+    # Closes the hand landmarker explicitly when the hand landmarker is not used
+    # in a context.
+    landmarker.close()
+
+  @parameterized.parameters(
+      (ModelFileType.FILE_NAME,
+       _get_expected_hand_landmarker_result(_THUMB_UP_LANDMARKS)),
+      (ModelFileType.FILE_CONTENT,
+       _get_expected_hand_landmarker_result(_THUMB_UP_LANDMARKS)))
+  def test_detect_in_context(self, model_file_type, expected_detection_result):
+    # Creates hand landmarker.
+    if model_file_type is ModelFileType.FILE_NAME:
+      base_options = _BaseOptions(model_asset_path=self.model_path)
+    elif model_file_type is ModelFileType.FILE_CONTENT:
+      with open(self.model_path, 'rb') as f:
+        model_content = f.read()
+      base_options = _BaseOptions(model_asset_buffer=model_content)
+    else:
+      # Should never happen
+      raise ValueError('model_file_type is invalid.')
+
+    options = _HandLandmarkerOptions(base_options=base_options)
+    with _HandLandmarker.create_from_options(options) as landmarker:
+      # Performs hand landmarks detection on the input.
+      detection_result = landmarker.detect(self.test_image)
+      # Comparing results.
+      self._assert_actual_result_approximately_matches_expected_result(
+          detection_result, expected_detection_result)
+
+  def test_detect_succeeds_with_num_hands(self):
+    # Creates hand landmarker.
+    base_options = _BaseOptions(model_asset_path=self.model_path)
+    options = _HandLandmarkerOptions(base_options=base_options, num_hands=2)
+    with _HandLandmarker.create_from_options(options) as landmarker:
+      # Load the two hands image.
+      test_image = _Image.create_from_file(
+          test_utils.get_test_data_path(_TWO_HANDS_IMAGE))
+      # Performs hand landmarks detection on the input.
+      detection_result = landmarker.detect(test_image)
+      # Comparing results.
+      self.assertLen(detection_result.handedness, 2)
+
+  def test_detect_succeeds_with_rotation(self):
+    # Creates hand landmarker.
+    base_options = _BaseOptions(model_asset_path=self.model_path)
+    options = _HandLandmarkerOptions(base_options=base_options)
+    with _HandLandmarker.create_from_options(options) as landmarker:
+      # Load the pointing up rotated image.
+      test_image = _Image.create_from_file(
+          test_utils.get_test_data_path(_POINTING_UP_ROTATED_IMAGE))
+      # Set rotation parameters using ImageProcessingOptions.
+      image_processing_options = _ImageProcessingOptions(rotation_degrees=-90)
+      # Performs hand landmarks detection on the input.
+      detection_result = landmarker.detect(test_image, image_processing_options)
+      expected_detection_result = _get_expected_hand_landmarker_result(
+          _POINTING_UP_ROTATED_LANDMARKS)
+      # Comparing results.
+      self._assert_actual_result_approximately_matches_expected_result(
+          detection_result, expected_detection_result)
+
+  def test_detect_fails_with_region_of_interest(self):
+    # Creates hand landmarker.
+    base_options = _BaseOptions(model_asset_path=self.model_path)
+    options = _HandLandmarkerOptions(base_options=base_options)
+    with self.assertRaisesRegex(
+        ValueError, "This task doesn't support region-of-interest."):
+      with _HandLandmarker.create_from_options(options) as landmarker:
+        # Set the `region_of_interest` parameter using `ImageProcessingOptions`.
+        image_processing_options = _ImageProcessingOptions(
+            region_of_interest=_Rect(0, 0, 1, 1))
+        # Attempt to perform hand landmarks detection on the cropped input.
+        landmarker.detect(self.test_image, image_processing_options)
+
+  def test_empty_detection_outputs(self):
+    options = _HandLandmarkerOptions(
+        base_options=_BaseOptions(model_asset_path=self.model_path))
+    with _HandLandmarker.create_from_options(options) as landmarker:
+      # Load the image with no hands.
+      no_hands_test_image = _Image.create_from_file(
+          test_utils.get_test_data_path(_NO_HANDS_IMAGE))
+      # Performs hand landmarks detection on the input.
+      detection_result = landmarker.detect(no_hands_test_image)
+      self.assertEmpty(detection_result.hand_landmarks)
+      self.assertEmpty(detection_result.hand_world_landmarks)
+      self.assertEmpty(detection_result.handedness)
+
+  def test_missing_result_callback(self):
+    options = _HandLandmarkerOptions(
+        base_options=_BaseOptions(model_asset_path=self.model_path),
+        running_mode=_RUNNING_MODE.LIVE_STREAM)
+    with self.assertRaisesRegex(ValueError,
+                                r'result callback must be provided'):
+      with _HandLandmarker.create_from_options(options) as unused_landmarker:
+        pass
+
+  @parameterized.parameters((_RUNNING_MODE.IMAGE), (_RUNNING_MODE.VIDEO))
+  def test_illegal_result_callback(self, running_mode):
+    options = _HandLandmarkerOptions(
+        base_options=_BaseOptions(model_asset_path=self.model_path),
+        running_mode=running_mode,
+        result_callback=mock.MagicMock())
+    with self.assertRaisesRegex(ValueError,
+                                r'result callback should not be provided'):
+      with _HandLandmarker.create_from_options(options) as unused_landmarker:
+        pass
+
+  def test_calling_detect_for_video_in_image_mode(self):
+    options = _HandLandmarkerOptions(
+        base_options=_BaseOptions(model_asset_path=self.model_path),
+        running_mode=_RUNNING_MODE.IMAGE)
+    with _HandLandmarker.create_from_options(options) as landmarker:
+      with self.assertRaisesRegex(ValueError,
+                                  r'not initialized with the video mode'):
+        landmarker.detect_for_video(self.test_image, 0)
+
+  def test_calling_detect_async_in_image_mode(self):
+    options = _HandLandmarkerOptions(
+        base_options=_BaseOptions(model_asset_path=self.model_path),
+        running_mode=_RUNNING_MODE.IMAGE)
+    with _HandLandmarker.create_from_options(options) as landmarker:
+      with self.assertRaisesRegex(ValueError,
+                                  r'not initialized with the live stream mode'):
+        landmarker.detect_async(self.test_image, 0)
+
+  def test_calling_detect_in_video_mode(self):
+    options = _HandLandmarkerOptions(
+        base_options=_BaseOptions(model_asset_path=self.model_path),
+        running_mode=_RUNNING_MODE.VIDEO)
+    with _HandLandmarker.create_from_options(options) as landmarker:
+      with self.assertRaisesRegex(ValueError,
+                                  r'not initialized with the image mode'):
+        landmarker.detect(self.test_image)
+
+  def test_calling_detect_async_in_video_mode(self):
+    options = _HandLandmarkerOptions(
+        base_options=_BaseOptions(model_asset_path=self.model_path),
+        running_mode=_RUNNING_MODE.VIDEO)
+    with _HandLandmarker.create_from_options(options) as landmarker:
+      with self.assertRaisesRegex(ValueError,
+                                  r'not initialized with the live stream mode'):
+        landmarker.detect_async(self.test_image, 0)
+
+  def test_detect_for_video_with_out_of_order_timestamp(self):
+    options = _HandLandmarkerOptions(
+        base_options=_BaseOptions(model_asset_path=self.model_path),
+        running_mode=_RUNNING_MODE.VIDEO)
+    with _HandLandmarker.create_from_options(options) as landmarker:
+      unused_result = landmarker.detect_for_video(self.test_image, 1)
+      with self.assertRaisesRegex(
+          ValueError, r'Input timestamp must be monotonically increasing'):
+        landmarker.detect_for_video(self.test_image, 0)
+
+  @parameterized.parameters(
+      (_THUMB_UP_IMAGE, 0,
+       _get_expected_hand_landmarker_result(_THUMB_UP_LANDMARKS)),
+      (_POINTING_UP_IMAGE, 0,
+       _get_expected_hand_landmarker_result(_POINTING_UP_LANDMARKS)),
+      (_POINTING_UP_ROTATED_IMAGE, -90,
+       _get_expected_hand_landmarker_result(_POINTING_UP_ROTATED_LANDMARKS)),
+      (_NO_HANDS_IMAGE, 0, _HandLandmarkerResult([], [], [])))
+  def test_detect_for_video(self, image_path, rotation, expected_result):
+    test_image = _Image.create_from_file(
+        test_utils.get_test_data_path(image_path))
+    # Set rotation parameters using ImageProcessingOptions.
+    image_processing_options = _ImageProcessingOptions(
+        rotation_degrees=rotation)
+    options = _HandLandmarkerOptions(
+        base_options=_BaseOptions(model_asset_path=self.model_path),
+        running_mode=_RUNNING_MODE.VIDEO)
+    with _HandLandmarker.create_from_options(options) as landmarker:
+      for timestamp in range(0, 300, 30):
+        result = landmarker.detect_for_video(test_image, timestamp,
+                                             image_processing_options)
+        if result.hand_landmarks and result.hand_world_landmarks and result.handedness:
+          self._assert_actual_result_approximately_matches_expected_result(
+              result, expected_result)
+        else:
+          self.assertEqual(result, expected_result)
+
+  def test_calling_detect_in_live_stream_mode(self):
+    options = _HandLandmarkerOptions(
+        base_options=_BaseOptions(model_asset_path=self.model_path),
+        running_mode=_RUNNING_MODE.LIVE_STREAM,
+        result_callback=mock.MagicMock())
+    with _HandLandmarker.create_from_options(options) as landmarker:
+      with self.assertRaisesRegex(ValueError,
+                                  r'not initialized with the image mode'):
+        landmarker.detect(self.test_image)
+
+  def test_calling_detect_for_video_in_live_stream_mode(self):
+    options = _HandLandmarkerOptions(
+        base_options=_BaseOptions(model_asset_path=self.model_path),
+        running_mode=_RUNNING_MODE.LIVE_STREAM,
+        result_callback=mock.MagicMock())
+    with _HandLandmarker.create_from_options(options) as landmarker:
+      with self.assertRaisesRegex(ValueError,
+                                  r'not initialized with the video mode'):
+        landmarker.detect_for_video(self.test_image, 0)
+
+  def test_detect_async_calls_with_illegal_timestamp(self):
+    options = _HandLandmarkerOptions(
+        base_options=_BaseOptions(model_asset_path=self.model_path),
+        running_mode=_RUNNING_MODE.LIVE_STREAM,
+        result_callback=mock.MagicMock())
+    with _HandLandmarker.create_from_options(options) as landmarker:
+      landmarker.detect_async(self.test_image, 100)
+      with self.assertRaisesRegex(
+          ValueError, r'Input timestamp must be monotonically increasing'):
+        landmarker.detect_async(self.test_image, 0)
+
+  @parameterized.parameters(
+      (_THUMB_UP_IMAGE, 0,
+       _get_expected_hand_landmarker_result(_THUMB_UP_LANDMARKS)),
+      (_POINTING_UP_IMAGE, 0,
+       _get_expected_hand_landmarker_result(_POINTING_UP_LANDMARKS)),
+      (_POINTING_UP_ROTATED_IMAGE, -90,
+       _get_expected_hand_landmarker_result(_POINTING_UP_ROTATED_LANDMARKS)),
+      (_NO_HANDS_IMAGE, 0, _HandLandmarkerResult([], [], [])))
+  def test_detect_async_calls(self, image_path, rotation, expected_result):
+    test_image = _Image.create_from_file(
+        test_utils.get_test_data_path(image_path))
+    # Set rotation parameters using ImageProcessingOptions.
+    image_processing_options = _ImageProcessingOptions(
+        rotation_degrees=rotation)
+    observed_timestamp_ms = -1
+
+    def check_result(result: _HandLandmarkerResult, output_image: _Image,
+                     timestamp_ms: int):
+      if result.hand_landmarks and result.hand_world_landmarks and result.handedness:
+        self._assert_actual_result_approximately_matches_expected_result(
+            result, expected_result)
+      else:
+        self.assertEqual(result, expected_result)
+      self.assertTrue(
+          np.array_equal(output_image.numpy_view(), test_image.numpy_view()))
+      self.assertLess(observed_timestamp_ms, timestamp_ms)
+      self.observed_timestamp_ms = timestamp_ms
+
+    options = _HandLandmarkerOptions(
+        base_options=_BaseOptions(model_asset_path=self.model_path),
+        running_mode=_RUNNING_MODE.LIVE_STREAM,
+        result_callback=check_result)
+    with _HandLandmarker.create_from_options(options) as landmarker:
+      for timestamp in range(0, 300, 30):
+        landmarker.detect_async(test_image, timestamp, image_processing_options)
+
+
+if __name__ == '__main__':
+  absltest.main()

mediapipe/tasks/python/vision/BUILD

@@ -102,3 +102,26 @@ py_library(
         "//mediapipe/tasks/python/vision/core:vision_task_running_mode",
     ],
 )
+
+py_library(
+    name = "hand_landmarker",
+    srcs = [
+        "hand_landmarker.py",
+    ],
+    deps = [
+        "//mediapipe/framework/formats:classification_py_pb2",
+        "//mediapipe/framework/formats:landmark_py_pb2",
+        "//mediapipe/python:_framework_bindings",
+        "//mediapipe/python:packet_creator",
+        "//mediapipe/python:packet_getter",
+        "//mediapipe/tasks/cc/vision/hand_landmarker/proto:hand_landmarker_graph_options_py_pb2",
+        "//mediapipe/tasks/python/components/containers:category",
+        "//mediapipe/tasks/python/components/containers:landmark",
+        "//mediapipe/tasks/python/core:base_options",
+        "//mediapipe/tasks/python/core:optional_dependencies",
+        "//mediapipe/tasks/python/core:task_info",
+        "//mediapipe/tasks/python/vision/core:base_vision_task_api",
+        "//mediapipe/tasks/python/vision/core:image_processing_options",
+        "//mediapipe/tasks/python/vision/core:vision_task_running_mode",
+    ],
+)

mediapipe/tasks/python/vision/gesture_recognizer.py

@@ -122,19 +122,21 @@ def _build_recognition_result(
   for proto in hand_landmarks_proto_list:
     hand_landmarks = landmark_pb2.NormalizedLandmarkList()
     hand_landmarks.MergeFrom(proto)
-    hand_landmarks_results.append([
+    hand_landmarks_list = []
-        landmark_module.NormalizedLandmark.create_from_pb2(hand_landmark)
+    for hand_landmark in hand_landmarks.landmark:
-        for hand_landmark in hand_landmarks.landmark
+      hand_landmarks_list.append(
-    ])
+          landmark_module.NormalizedLandmark.create_from_pb2(hand_landmark))
+    hand_landmarks_results.append(hand_landmarks_list)
 
   hand_world_landmarks_results = []
   for proto in hand_world_landmarks_proto_list:
     hand_world_landmarks = landmark_pb2.LandmarkList()
     hand_world_landmarks.MergeFrom(proto)
-    hand_world_landmarks_results.append([
+    hand_world_landmarks_list = []
-        landmark_module.Landmark.create_from_pb2(hand_world_landmark)
+    for hand_world_landmark in hand_world_landmarks.landmark:
-        for hand_world_landmark in hand_world_landmarks.landmark
+      hand_world_landmarks_list.append(
-    ])
+          landmark_module.Landmark.create_from_pb2(hand_world_landmark))
+    hand_world_landmarks_results.append(hand_world_landmarks_list)
 
   return GestureRecognitionResult(gesture_results, handedness_results,
                                   hand_landmarks_results,

mediapipe/tasks/python/vision/hand_landmarker.py

@@ -0,0 +1,379 @@
+# Copyright 2022 The MediaPipe Authors. All Rights Reserved.
+#
+# 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 hand landmarker task."""
+
+import dataclasses
+from typing import Callable, Mapping, Optional, List
+
+from mediapipe.framework.formats import classification_pb2
+from mediapipe.framework.formats import landmark_pb2
+from mediapipe.python import packet_creator
+from mediapipe.python import packet_getter
+from mediapipe.python._framework_bindings import image as image_module
+from mediapipe.python._framework_bindings import packet as packet_module
+from mediapipe.tasks.cc.vision.hand_landmarker.proto import hand_landmarker_graph_options_pb2
+from mediapipe.tasks.python.components.containers import category as category_module
+from mediapipe.tasks.python.components.containers import landmark as landmark_module
+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
+_HandLandmarkerGraphOptionsProto = hand_landmarker_graph_options_pb2.HandLandmarkerGraphOptions
+_RunningMode = running_mode_module.VisionTaskRunningMode
+_ImageProcessingOptions = image_processing_options_module.ImageProcessingOptions
+_TaskInfo = task_info_module.TaskInfo
+
+_IMAGE_IN_STREAM_NAME = 'image_in'
+_IMAGE_OUT_STREAM_NAME = 'image_out'
+_IMAGE_TAG = 'IMAGE'
+_NORM_RECT_STREAM_NAME = 'norm_rect_in'
+_NORM_RECT_TAG = 'NORM_RECT'
+_HANDEDNESS_STREAM_NAME = 'handedness'
+_HANDEDNESS_TAG = 'HANDEDNESS'
+_HAND_LANDMARKS_STREAM_NAME = 'landmarks'
+_HAND_LANDMARKS_TAG = 'LANDMARKS'
+_HAND_WORLD_LANDMARKS_STREAM_NAME = 'world_landmarks'
+_HAND_WORLD_LANDMARKS_TAG = 'WORLD_LANDMARKS'
+_TASK_GRAPH_NAME = 'mediapipe.tasks.vision.hand_landmarker.HandLandmarkerGraph'
+_MICRO_SECONDS_PER_MILLISECOND = 1000
+
+
+@dataclasses.dataclass
+class HandLandmarkerResult:
+  """The hand landmarks result from HandLandmarker, where each vector element represents a single hand detected in the image.
+
+  Attributes:
+    handedness: Classification of handedness.
+    hand_landmarks: Detected hand landmarks in normalized image coordinates.
+    hand_world_landmarks: Detected hand landmarks in world coordinates.
+  """
+
+  handedness: List[List[category_module.Category]]
+  hand_landmarks: List[List[landmark_module.NormalizedLandmark]]
+  hand_world_landmarks: List[List[landmark_module.Landmark]]
+
+
+def _build_landmarker_result(
+    output_packets: Mapping[str, packet_module.Packet]) -> HandLandmarkerResult:
+  """Constructs a `HandLandmarksDetectionResult` from output packets."""
+  handedness_proto_list = packet_getter.get_proto_list(
+      output_packets[_HANDEDNESS_STREAM_NAME])
+  hand_landmarks_proto_list = packet_getter.get_proto_list(
+      output_packets[_HAND_LANDMARKS_STREAM_NAME])
+  hand_world_landmarks_proto_list = packet_getter.get_proto_list(
+      output_packets[_HAND_WORLD_LANDMARKS_STREAM_NAME])
+
+  handedness_results = []
+  for proto in handedness_proto_list:
+    handedness_categories = []
+    handedness_classifications = classification_pb2.ClassificationList()
+    handedness_classifications.MergeFrom(proto)
+    for handedness in handedness_classifications.classification:
+      handedness_categories.append(
+          category_module.Category(
+              index=handedness.index,
+              score=handedness.score,
+              display_name=handedness.display_name,
+              category_name=handedness.label))
+    handedness_results.append(handedness_categories)
+
+  hand_landmarks_results = []
+  for proto in hand_landmarks_proto_list:
+    hand_landmarks = landmark_pb2.NormalizedLandmarkList()
+    hand_landmarks.MergeFrom(proto)
+    hand_landmarks_list = []
+    for hand_landmark in hand_landmarks.landmark:
+      hand_landmarks_list.append(
+          landmark_module.NormalizedLandmark.create_from_pb2(hand_landmark))
+    hand_landmarks_results.append(hand_landmarks_list)
+
+  hand_world_landmarks_results = []
+  for proto in hand_world_landmarks_proto_list:
+    hand_world_landmarks = landmark_pb2.LandmarkList()
+    hand_world_landmarks.MergeFrom(proto)
+    hand_world_landmarks_list = []
+    for hand_world_landmark in hand_world_landmarks.landmark:
+      hand_world_landmarks_list.append(
+          landmark_module.Landmark.create_from_pb2(hand_world_landmark))
+    hand_world_landmarks_results.append(hand_world_landmarks_list)
+
+  return HandLandmarkerResult(handedness_results, hand_landmarks_results,
+                              hand_world_landmarks_results)
+
+
+@dataclasses.dataclass
+class HandLandmarkerOptions:
+  """Options for the hand landmarker task.
+
+  Attributes:
+    base_options: Base options for the hand landmarker task.
+    running_mode: The running mode of the task. Default to the image mode.
+      HandLandmarker has three running modes: 1) The image mode for detecting
+      hand landmarks on single image inputs. 2) The video mode for detecting
+      hand landmarks on the decoded frames of a video. 3) The live stream mode
+      for detecting hand landmarks on the live stream of input data, such as
+      from camera. In this mode, the "result_callback" below must be specified
+      to receive the detection results asynchronously.
+    num_hands: The maximum number of hands can be detected by the hand
+      landmarker.
+    min_hand_detection_confidence: The minimum confidence score for the hand
+      detection to be considered successful.
+    min_hand_presence_confidence: The minimum confidence score of hand presence
+      score in the hand landmark detection.
+    min_tracking_confidence: The minimum confidence score for the hand tracking
+      to be considered successful.
+    result_callback: The user-defined result callback for processing live stream
+      data. The result callback should only be specified when the running mode
+      is set to the live stream mode.
+  """
+  base_options: _BaseOptions
+  running_mode: _RunningMode = _RunningMode.IMAGE
+  num_hands: Optional[int] = 1
+  min_hand_detection_confidence: Optional[float] = 0.5
+  min_hand_presence_confidence: Optional[float] = 0.5
+  min_tracking_confidence: Optional[float] = 0.5
+  result_callback: Optional[Callable[
+      [HandLandmarkerResult, image_module.Image, int], None]] = None
+
+  @doc_controls.do_not_generate_docs
+  def to_pb2(self) -> _HandLandmarkerGraphOptionsProto:
+    """Generates an HandLandmarkerGraphOptions protobuf object."""
+    base_options_proto = self.base_options.to_pb2()
+    base_options_proto.use_stream_mode = False if self.running_mode == _RunningMode.IMAGE else True
+
+    # Initialize the hand landmarker options from base options.
+    hand_landmarker_options_proto = _HandLandmarkerGraphOptionsProto(
+        base_options=base_options_proto)
+    hand_landmarker_options_proto.min_tracking_confidence = self.min_tracking_confidence
+    hand_landmarker_options_proto.hand_detector_graph_options.num_hands = self.num_hands
+    hand_landmarker_options_proto.hand_detector_graph_options.min_detection_confidence = self.min_hand_detection_confidence
+    hand_landmarker_options_proto.hand_landmarks_detector_graph_options.min_detection_confidence = self.min_hand_presence_confidence
+    return hand_landmarker_options_proto
+
+
+class HandLandmarker(base_vision_task_api.BaseVisionTaskApi):
+  """Class that performs hand landmarks detection on images."""
+
+  @classmethod
+  def create_from_model_path(cls, model_path: str) -> 'HandLandmarker':
+    """Creates an `HandLandmarker` object from a TensorFlow Lite model and the default `HandLandmarkerOptions`.
+
+    Note that the created `HandLandmarker` instance is in image mode, for
+    detecting hand landmarks on single image inputs.
+
+    Args:
+      model_path: Path to the model.
+
+    Returns:
+      `HandLandmarker` object that's created from the model file and the
+      default `HandLandmarkerOptions`.
+
+    Raises:
+      ValueError: If failed to create `HandLandmarker` 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 = HandLandmarkerOptions(
+        base_options=base_options, running_mode=_RunningMode.IMAGE)
+    return cls.create_from_options(options)
+
+  @classmethod
+  def create_from_options(cls,
+                          options: HandLandmarkerOptions) -> 'HandLandmarker':
+    """Creates the `HandLandmarker` object from hand landmarker options.
+
+    Args:
+      options: Options for the hand landmarker task.
+
+    Returns:
+      `HandLandmarker` object that's created from `options`.
+
+    Raises:
+      ValueError: If failed to create `HandLandmarker` object from
+        `HandLandmarkerOptions` such as missing the model.
+      RuntimeError: If other types of error occurred.
+    """
+
+    def packets_callback(output_packets: Mapping[str, packet_module.Packet]):
+      if output_packets[_IMAGE_OUT_STREAM_NAME].is_empty():
+        return
+
+      image = packet_getter.get_image(output_packets[_IMAGE_OUT_STREAM_NAME])
+
+      if output_packets[_HAND_LANDMARKS_STREAM_NAME].is_empty():
+        empty_packet = output_packets[_HAND_LANDMARKS_STREAM_NAME]
+        options.result_callback(
+            HandLandmarkerResult([], [], []), image,
+            empty_packet.timestamp.value // _MICRO_SECONDS_PER_MILLISECOND)
+        return
+
+      hand_landmarks_detection_result = _build_landmarker_result(output_packets)
+      timestamp = output_packets[_HAND_LANDMARKS_STREAM_NAME].timestamp
+      options.result_callback(hand_landmarks_detection_result, image,
+                              timestamp.value // _MICRO_SECONDS_PER_MILLISECOND)
+
+    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([_HANDEDNESS_TAG, _HANDEDNESS_STREAM_NAME]),
+            ':'.join([_HAND_LANDMARKS_TAG, _HAND_LANDMARKS_STREAM_NAME]),
+            ':'.join([
+                _HAND_WORLD_LANDMARKS_TAG, _HAND_WORLD_LANDMARKS_STREAM_NAME
+            ]), ':'.join([_IMAGE_TAG, _IMAGE_OUT_STREAM_NAME])
+        ],
+        task_options=options)
+    return cls(
+        task_info.generate_graph_config(
+            enable_flow_limiting=options.running_mode ==
+            _RunningMode.LIVE_STREAM), options.running_mode,
+        packets_callback if options.result_callback else None)
+
+  def detect(
+      self,
+      image: image_module.Image,
+      image_processing_options: Optional[_ImageProcessingOptions] = None
+  ) -> HandLandmarkerResult:
+    """Performs hand landmarks detection on the given image.
+
+    Only use this method when the HandLandmarker is created with the image
+    running mode.
+
+    The image can be of any size with format RGB or RGBA.
+    TODO: Describes how the input image will be preprocessed after the yuv
+    support is implemented.
+
+    Args:
+      image: MediaPipe Image.
+      image_processing_options: Options for image processing.
+
+    Returns:
+      The hand landmarks detection results.
+
+    Raises:
+      ValueError: If any of the input arguments is invalid.
+      RuntimeError: If hand landmarker detection failed to run.
+    """
+    normalized_rect = self.convert_to_normalized_rect(
+        image_processing_options, roi_allowed=False)
+    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[_HAND_LANDMARKS_STREAM_NAME].is_empty():
+      return HandLandmarkerResult([], [], [])
+
+    return _build_landmarker_result(output_packets)
+
+  def detect_for_video(
+      self,
+      image: image_module.Image,
+      timestamp_ms: int,
+      image_processing_options: Optional[_ImageProcessingOptions] = None
+  ) -> HandLandmarkerResult:
+    """Performs hand landmarks detection on the provided video frame.
+
+    Only use this method when the HandLandmarker is created with the video
+    running mode.
+
+    Only use this method when the HandLandmarker is created with the video
+    running mode. It's required to provide the video frame's timestamp (in
+    milliseconds) along with the video frame. The input timestamps should be
+    monotonically increasing for adjacent calls of this method.
+
+    Args:
+      image: MediaPipe Image.
+      timestamp_ms: The timestamp of the input video frame in milliseconds.
+      image_processing_options: Options for image processing.
+
+    Returns:
+      The hand landmarks detection results.
+
+    Raises:
+      ValueError: If any of the input arguments is invalid.
+      RuntimeError: If hand landmarker detection failed to run.
+    """
+    normalized_rect = self.convert_to_normalized_rect(
+        image_processing_options, roi_allowed=False)
+    output_packets = self._process_video_data({
+        _IMAGE_IN_STREAM_NAME:
+            packet_creator.create_image(image).at(
+                timestamp_ms * _MICRO_SECONDS_PER_MILLISECOND),
+        _NORM_RECT_STREAM_NAME:
+            packet_creator.create_proto(normalized_rect.to_pb2()).at(
+                timestamp_ms * _MICRO_SECONDS_PER_MILLISECOND)
+    })
+
+    if output_packets[_HAND_LANDMARKS_STREAM_NAME].is_empty():
+      return HandLandmarkerResult([], [], [])
+
+    return _build_landmarker_result(output_packets)
+
+  def detect_async(
+      self,
+      image: image_module.Image,
+      timestamp_ms: int,
+      image_processing_options: Optional[_ImageProcessingOptions] = None
+  ) -> None:
+    """Sends live image data to perform hand landmarks detection.
+
+    The results will be available via the "result_callback" provided in the
+    HandLandmarkerOptions. Only use this method when the HandLandmarker is
+    created with the live stream running mode.
+
+    Only use this method when the HandLandmarker is created with the live
+    stream running mode. The input timestamps should be monotonically increasing
+    for adjacent calls of this method. This method will return immediately after
+    the input image is accepted. The results will be available via the
+    `result_callback` provided in the `HandLandmarkerOptions`. The
+    `detect_async` method is designed to process live stream data such as
+    camera input. To lower the overall latency, hand landmarker may drop the
+    input images if needed. In other words, it's not guaranteed to have output
+    per input image.
+
+    The `result_callback` provides:
+      - The hand landmarks detection results.
+      - The input image that the hand landmarker runs on.
+      - The input timestamp in milliseconds.
+
+    Args:
+      image: MediaPipe Image.
+      timestamp_ms: The timestamp of the input image in milliseconds.
+      image_processing_options: Options for image processing.
+
+    Raises:
+      ValueError: If the current input timestamp is smaller than what the
+      hand landmarker has already processed.
+    """
+    normalized_rect = self.convert_to_normalized_rect(
+        image_processing_options, roi_allowed=False)
+    self._send_live_stream_data({
+        _IMAGE_IN_STREAM_NAME:
+            packet_creator.create_image(image).at(
+                timestamp_ms * _MICRO_SECONDS_PER_MILLISECOND),
+        _NORM_RECT_STREAM_NAME:
+            packet_creator.create_proto(normalized_rect.to_pb2()).at(
+                timestamp_ms * _MICRO_SECONDS_PER_MILLISECOND)
+    })