diff --git a/mediapipe/tasks/python/test/vision/BUILD b/mediapipe/tasks/python/test/vision/BUILD index bb495338d..c01645ee9 100644 --- a/mediapipe/tasks/python/test/vision/BUILD +++ b/mediapipe/tasks/python/test/vision/BUILD @@ -19,3 +19,21 @@ package(default_visibility = ["//mediapipe/tasks:internal"]) licenses(["notice"]) # TODO: This test fails in OSS + +py_test( + name = "image_classification_test", + srcs = ["image_classification_test.py"], + data = [ + "//mediapipe/tasks/testdata/vision:test_images", + "//mediapipe/tasks/testdata/vision:test_models", + ], + deps = [ + "//mediapipe/tasks/python/components/containers:category", + "//mediapipe/tasks/python/components/containers:classifications", + "//mediapipe/tasks/python/core:base_options", + "//mediapipe/tasks/python/test:test_util", + "//mediapipe/tasks/python/vision:image_classification", + "//mediapipe/tasks/python/vision/core:vision_task_running_mode", + "@absl_py//absl/testing:parameterized", + ], +) \ No newline at end of file diff --git a/mediapipe/tasks/python/test/vision/image_classification_test.py b/mediapipe/tasks/python/test/vision/image_classification_test.py new file mode 100644 index 000000000..a0c6b9ef6 --- /dev/null +++ b/mediapipe/tasks/python/test/vision/image_classification_test.py @@ -0,0 +1,291 @@ +# 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 image classifier.""" + +import enum + +from absl.testing import absltest +from absl.testing import parameterized + +from mediapipe.python._framework_bindings import image as image_module +from mediapipe.tasks.python.components import classifier_options +from mediapipe.tasks.python.components.containers import category as category_module +from mediapipe.tasks.python.components.containers import classifications as classifications_module +from mediapipe.tasks.python.core import base_options as base_options_module +from mediapipe.tasks.python.test import test_util +from mediapipe.tasks.python.vision import image_classification +from mediapipe.tasks.python.vision.core import vision_task_running_mode as running_mode_module + +_BaseOptions = base_options_module.BaseOptions +_ClassifierOptions = classifier_options.ClassifierOptions +_Category = category_module.Category +_ClassificationEntry = classifications_module.ClassificationEntry +_Classifications = classifications_module.Classifications +_ClassificationResult = classifications_module.ClassificationResult +_Image = image_module.Image +_ImageClassifier = image_classification.ImageClassifier +_ImageClassifierOptions = image_classification.ImageClassifierOptions +_RUNNING_MODE = running_mode_module.VisionTaskRunningMode + +_MODEL_FILE = 'mobilenet_v2_1.0_224.tflite' +_IMAGE_FILE = 'burger.jpg' +_EXPECTED_CLASSIFICATION_RESULT = _ClassificationResult( + classifications=[ + _Classifications( + entries=[ + _ClassificationEntry( + categories=[ + _Category( + index=934, + score=0.7952049970626831, + display_name='', + category_name='cheeseburger'), + _Category( + index=932, + score=0.02732999622821808, + display_name='', + category_name='bagel'), + _Category( + index=925, + score=0.01933487318456173, + display_name='', + category_name='guacamole'), + _Category( + index=963, + score=0.006279350258409977, + display_name='', + category_name='meat loaf') + ], + timestamp_ms=0 + ) + ], + head_index=0, + head_name='probability') + ]) +_ALLOW_LIST = ['cheeseburger', 'guacamole'] +_DENY_LIST = ['cheeseburger'] +_SCORE_THRESHOLD = 0.5 +_MAX_RESULTS = 3 + + +class ModelFileType(enum.Enum): + FILE_CONTENT = 1 + FILE_NAME = 2 + + +class ImageClassifierTest(parameterized.TestCase): + + def setUp(self): + super().setUp() + self.test_image = test_util.read_test_image( + test_util.get_test_data_path(_IMAGE_FILE)) + self.model_path = test_util.get_test_data_path(_MODEL_FILE) + + def test_create_from_file_succeeds_with_valid_model_path(self): + # Creates with default option and valid model file successfully. + with _ImageClassifier.create_from_model_path(self.model_path) as classifier: + self.assertIsInstance(classifier, _ImageClassifier) + + def test_create_from_options_succeeds_with_valid_model_path(self): + # Creates with options containing model file successfully. + base_options = _BaseOptions(file_name=self.model_path) + options = _ImageClassifierOptions(base_options=base_options) + with _ImageClassifier.create_from_options(options) as classifier: + self.assertIsInstance(classifier, _ImageClassifier) + + def test_create_from_options_fails_with_invalid_model_path(self): + # Invalid empty model path. + with self.assertRaisesRegex( + ValueError, + r"ExternalFile must specify at least one of 'file_content', " + r"'file_name' or 'file_descriptor_meta'."): + base_options = _BaseOptions(file_name='') + options = _ImageClassifierOptions(base_options=base_options) + _ImageClassifier.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(file_content=f.read()) + options = _ImageClassifierOptions(base_options=base_options) + classifier = _ImageClassifier.create_from_options(options) + self.assertIsInstance(classifier, _ImageClassifier) + + @parameterized.parameters( + (ModelFileType.FILE_NAME, 4, _EXPECTED_CLASSIFICATION_RESULT), + (ModelFileType.FILE_CONTENT, 4, _EXPECTED_CLASSIFICATION_RESULT)) + def test_classify(self, model_file_type, max_results, + expected_classification_result): + # Creates classifier. + if model_file_type is ModelFileType.FILE_NAME: + base_options = _BaseOptions(file_name=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(file_content=model_content) + else: + # Should never happen + raise ValueError('model_file_type is invalid.') + + classifier_options = _ClassifierOptions(max_results=max_results) + options = _ImageClassifierOptions( + base_options=base_options, classifier_options=classifier_options) + classifier = _ImageClassifier.create_from_options(options) + + # Performs image classification on the input. + image_result = classifier.classify(self.test_image) + # Comparing results. + self.assertEqual(image_result, expected_classification_result) + # Closes the classifier explicitly when the classifier is not used in + # a context. + classifier.close() + + @parameterized.parameters( + (ModelFileType.FILE_NAME, 4, _EXPECTED_CLASSIFICATION_RESULT), + (ModelFileType.FILE_CONTENT, 4, _EXPECTED_CLASSIFICATION_RESULT)) + def test_classify_in_context(self, model_file_type, max_results, + expected_classification_result): + if model_file_type is ModelFileType.FILE_NAME: + base_options = _BaseOptions(file_name=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(file_content=model_content) + else: + # Should never happen + raise ValueError('model_file_type is invalid.') + + classifier_options = _ClassifierOptions(max_results=max_results) + options = _ImageClassifierOptions( + base_options=base_options, classifier_options=classifier_options) + with _ImageClassifier.create_from_options(options) as classifier: + # Performs object detection on the input. + image_result = classifier.classify(self.test_image) + # Comparing results. + self.assertEqual(image_result, expected_classification_result) + + def test_score_threshold_option(self): + classifier_options = _ClassifierOptions(score_threshold=_SCORE_THRESHOLD) + options = _ImageClassifierOptions( + base_options=_BaseOptions(file_name=self.model_path), + classifier_options=classifier_options) + with _ImageClassifier.create_from_options(options) as classifier: + # Performs image classification on the input. + image_result = classifier.classify(self.test_image) + classifications = image_result.classifications + + for classification in classifications: + for entry in classification.entries: + score = entry.categories[0].score + self.assertGreaterEqual( + score, _SCORE_THRESHOLD, + f'Classification with score lower than threshold found. ' + f'{classification}') + + def test_max_results_option(self): + classifier_options = _ClassifierOptions(score_threshold=_SCORE_THRESHOLD) + options = _ImageClassifierOptions( + base_options=_BaseOptions(file_name=self.model_path), + classifier_options=classifier_options) + with _ImageClassifier.create_from_options(options) as classifier: + # Performs image classification on the input. + image_result = classifier.classify(self.test_image) + categories = image_result.classifications[0].entries[0].categories + + self.assertLessEqual( + len(categories), _MAX_RESULTS, 'Too many results returned.') + + def test_allow_list_option(self): + classifier_options = _ClassifierOptions(category_allowlist=_ALLOW_LIST) + options = _ImageClassifierOptions( + base_options=_BaseOptions(file_name=self.model_path), + classifier_options=classifier_options) + with _ImageClassifier.create_from_options(options) as classifier: + # Performs image classification on the input. + image_result = classifier.classify(self.test_image) + classifications = image_result.classifications + + for classification in classifications: + for entry in classification.entries: + label = entry.categories[0].category_name + self.assertIn(label, _ALLOW_LIST, + f'Label {label} found but not in label allow list') + + def test_deny_list_option(self): + classifier_options = _ClassifierOptions(category_denylist=_DENY_LIST) + options = _ImageClassifierOptions( + base_options=_BaseOptions(file_name=self.model_path), + classifier_options=classifier_options) + with _ImageClassifier.create_from_options(options) as classifier: + # Performs image classification on the input. + image_result = classifier.classify(self.test_image) + classifications = image_result.classifications + + for classification in classifications: + for entry in classification.entries: + label = entry.categories[0].category_name + self.assertNotIn(label, _DENY_LIST, + f'Label {label} found but in deny list.') + + def test_combined_allowlist_and_denylist(self): + # Fails with combined allowlist and denylist + with self.assertRaisesRegex( + ValueError, + r'`category_allowlist` and `category_denylist` are mutually ' + r'exclusive options.'): + classifier_options = _ClassifierOptions(category_allowlist=['foo'], + category_denylist=['bar']) + options = _ImageClassifierOptions( + base_options=_BaseOptions(file_name=self.model_path), + classifier_options=classifier_options) + with _ImageClassifier.create_from_options(options) as unused_classifier: + pass + + def test_empty_classification_outputs(self): + classifier_options = _ClassifierOptions(score_threshold=1) + options = _ImageClassifierOptions( + base_options=_BaseOptions(file_name=self.model_path), + classifier_options=classifier_options) + with _ImageClassifier.create_from_options(options) as classifier: + # Performs image classification on the input. + image_result = classifier.classify(self.test_image) + self.assertEmpty(image_result.classifications[0].entries[0].categories) + + def test_missing_result_callback(self): + options = _ImageClassifierOptions( + base_options=_BaseOptions(file_name=self.model_path), + running_mode=_RUNNING_MODE.LIVE_STREAM) + with self.assertRaisesRegex(ValueError, + r'result callback must be provided'): + with _ImageClassifier.create_from_options(options) as unused_classifier: + pass + + @parameterized.parameters((_RUNNING_MODE.IMAGE), (_RUNNING_MODE.VIDEO)) + def test_illegal_result_callback(self, running_mode): + + def pass_through(unused_result: _ClassificationResult): + pass + + options = _ImageClassifierOptions( + base_options=_BaseOptions(file_name=self.model_path), + running_mode=running_mode, + result_callback=pass_through) + with self.assertRaisesRegex(ValueError, + r'result callback should not be provided'): + with _ImageClassifier.create_from_options(options) as unused_classifier: + pass + + +if __name__ == '__main__': + absltest.main() diff --git a/mediapipe/tasks/python/vision/core/BUILD b/mediapipe/tasks/python/vision/core/BUILD index c7422969a..b777035d9 100644 --- a/mediapipe/tasks/python/vision/core/BUILD +++ b/mediapipe/tasks/python/vision/core/BUILD @@ -35,3 +35,23 @@ py_library( "//mediapipe/tasks/python/core:optional_dependencies", ], ) + +py_library( + name = "image_classification", + srcs = [ + "image_classification.py", + ], + deps = [ + "//mediapipe/python:_framework_bindings", + "//mediapipe/python:packet_creator", + "//mediapipe/python:packet_getter", + "//mediapipe/tasks/cc/vision/image_classification:image_classifier_options_py_pb2", + "//mediapipe/tasks/python/components:classifier_options", + "//mediapipe/tasks/python/components/containers:classifications", + "//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:vision_task_running_mode", + ], +) diff --git a/mediapipe/tasks/python/vision/core/image_classification.py b/mediapipe/tasks/python/vision/core/image_classification.py new file mode 100644 index 000000000..40a699b90 --- /dev/null +++ b/mediapipe/tasks/python/vision/core/image_classification.py @@ -0,0 +1,180 @@ +# 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 image classifier task.""" + +import dataclasses +from typing import Callable, List, Mapping, 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.python._framework_bindings import packet as packet_module +from mediapipe.python._framework_bindings import task_runner as task_runner_module +from mediapipe.tasks.cc.vision.image_classification import image_classifier_options_pb2 +from mediapipe.tasks.python.components import classifier_options +from mediapipe.tasks.python.components.containers import classifications as classifications_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 vision_task_running_mode as running_mode_module + +_BaseOptions = base_options_module.BaseOptions +_ImageClassifierOptionsProto = image_classifier_options_pb2.ImageClassifierOptions +_ClassifierOptions = classifier_options.ClassifierOptions +_RunningMode = running_mode_module.VisionTaskRunningMode +_TaskInfo = task_info_module.TaskInfo +_TaskRunner = task_runner_module.TaskRunner + +_CLASSIFICATION_RESULT_OUT_STREAM_NAME = 'classification_result_out' +_CLASSIFICATION_RESULT_TAG = 'CLASSIFICATION_RESULT' +_IMAGE_IN_STREAM_NAME = 'image_in' +_IMAGE_TAG = 'IMAGE' +_TASK_GRAPH_NAME = 'mediapipe.tasks.vision.ImageClassifierGraph' + + +@dataclasses.dataclass +class ImageClassifierOptions: + """Options for the image classifier task. + Attributes: + base_options: Base options for the image classifier task. + running_mode: The running mode of the task. Default to the image mode. + Image classifier task has three running modes: + 1) The image mode for classifying objects on single image inputs. + 2) The video mode for classifying objects on the decoded frames of a + video. + 3) The live stream mode for classifying objects on a live stream of input + data, such as from camera. + display_names_locale: The locale to use for display names specified through + the TFLite Model Metadata. + max_results: The maximum number of top-scored classification results to + return. + score_threshold: Overrides the ones provided in the model metadata. Results + below this value are rejected. + category_allowlist: Allowlist of category names. If non-empty, detection + results whose category name is not in this set will be filtered out. + Duplicate or unknown category names are ignored. Mutually exclusive with + `category_denylist`. + category_denylist: Denylist of category names. If non-empty, detection + results whose category name is in this set will be filtered out. Duplicate + or unknown category names are ignored. Mutually exclusive with + `category_allowlist`. + 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 + classifier_options: _ClassifierOptions = _ClassifierOptions() + result_callback: Optional[ + Callable[[classifications_module.ClassificationResult], + None]] = None + + @doc_controls.do_not_generate_docs + def to_pb2(self) -> _ImageClassifierOptionsProto: + """Generates an ImageClassifierOptions 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 + classifier_options_proto = self.classifier_options.to_pb2() + + return _ImageClassifierOptionsProto( + base_options=base_options_proto, + classifier_options=classifier_options_proto + ) + + +class ImageClassifier(base_vision_task_api.BaseVisionTaskApi): + """Class that performs image classification on images.""" + + @classmethod + def create_from_model_path(cls, model_path: str) -> 'ImageClassifier': + """Creates an `ImageClassifier` object from a TensorFlow Lite model and the default `ImageClassifierOptions`. + Note that the created `ImageClassifier` instance is in image mode, for + detecting objects on single image inputs. + Args: + model_path: Path to the model. + Returns: + `ImageClassifier` object that's created from the model file and the default + `ImageClassifierOptions`. + Raises: + ValueError: If failed to create `ImageClassifier` object from the provided + file such as invalid file path. + RuntimeError: If other types of error occurred. + """ + base_options = _BaseOptions(file_name=model_path) + options = ImageClassifierOptions( + base_options=base_options, running_mode=_RunningMode.IMAGE) + return cls.create_from_options(options) + + @classmethod + def create_from_options(cls, + options: ImageClassifierOptions) -> 'ImageClassifier': + """Creates the `ImageClassifier` object from image classifier options. + Args: + options: Options for the image classifier task. + Returns: + `ImageClassifier` object that's created from `options`. + Raises: + ValueError: If failed to create `ImageClassifier` object from + `ImageClassifierOptions` such as missing the model. + RuntimeError: If other types of error occurred. + """ + + def packets_callback(output_packets: Mapping[str, packet_module.Packet]): + classification_result_proto = packet_getter.get_proto( + output_packets[_CLASSIFICATION_RESULT_OUT_STREAM_NAME]) + + classification_result = classifications_module.ClassificationResult([ + classifications_module.Classifications.create_from_pb2(classification) + for classification in classification_result_proto.classifications + ]) + options.result_callback(classification_result) + + task_info = _TaskInfo( + task_graph=_TASK_GRAPH_NAME, + input_streams=[':'.join([_IMAGE_TAG, _IMAGE_IN_STREAM_NAME])], + output_streams=[ + ':'.join([_CLASSIFICATION_RESULT_TAG, + _CLASSIFICATION_RESULT_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) + + # TODO: Create an Image class for MediaPipe Tasks. + def classify( + self, + image: image_module.Image + ) -> classifications_module.ClassificationResult: + """Performs image classification on the provided MediaPipe Image. + Args: + image: MediaPipe Image. + Returns: + A classification result object that contains a list of classifications. + Raises: + ValueError: If any of the input arguments is invalid. + RuntimeError: If image classification failed to run. + """ + output_packets = self._process_image_data( + {_IMAGE_IN_STREAM_NAME: packet_creator.create_image(image)}) + classification_result_proto = packet_getter.get_proto( + output_packets[_CLASSIFICATION_RESULT_OUT_STREAM_NAME]) + + return classifications_module.ClassificationResult([ + classifications_module.Classifications.create_from_pb2(classification) + for classification in classification_result_proto.classifications + ])