From cb52432159ecfea69c6df54be2cb56fd569f275f Mon Sep 17 00:00:00 2001 From: kinaryml Date: Thu, 8 Sep 2022 06:23:03 -0700 Subject: [PATCH] Added image classification implementation files and associated tests --- .../tasks/python/components/containers/BUILD | 10 + .../components/containers/classifications.py | 169 ++++++++++ mediapipe/tasks/python/test/vision/BUILD | 38 ++- .../test/vision/image_classification_test.py | 301 ++++++++++++++++++ mediapipe/tasks/python/vision/BUILD | 20 ++ .../python/vision/image_classification.py | 227 +++++++++++++ 6 files changed, 764 insertions(+), 1 deletion(-) create mode 100644 mediapipe/tasks/python/components/containers/classifications.py create mode 100644 mediapipe/tasks/python/test/vision/image_classification_test.py create mode 100644 mediapipe/tasks/python/vision/image_classification.py diff --git a/mediapipe/tasks/python/components/containers/BUILD b/mediapipe/tasks/python/components/containers/BUILD index 2bc951220..eb3acdd97 100644 --- a/mediapipe/tasks/python/components/containers/BUILD +++ b/mediapipe/tasks/python/components/containers/BUILD @@ -47,3 +47,13 @@ py_library( "//mediapipe/tasks/python/core:optional_dependencies", ], ) + +py_library( + name = "classifications", + srcs = ["classifications.py"], + deps = [ + ":category", + "//mediapipe/tasks/cc/components/containers:classifications_py_pb2", + "//mediapipe/tasks/python/core:optional_dependencies", + ], +) diff --git a/mediapipe/tasks/python/components/containers/classifications.py b/mediapipe/tasks/python/components/containers/classifications.py new file mode 100644 index 000000000..19c5decde --- /dev/null +++ b/mediapipe/tasks/python/components/containers/classifications.py @@ -0,0 +1,169 @@ +# Copyright 2022 The TensorFlow 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. +"""Classifications data class.""" + +import dataclasses +from typing import Any, List, Optional + +from mediapipe.tasks.cc.components.containers import classifications_pb2 +from mediapipe.tasks.python.components.containers import category as category_module +from mediapipe.tasks.python.core.optional_dependencies import doc_controls + +_ClassificationEntryProto = classifications_pb2.ClassificationEntry +_ClassificationsProto = classifications_pb2.Classifications +_ClassificationResultProto = classifications_pb2.ClassificationResult + + +@dataclasses.dataclass +class ClassificationEntry: + """List of predicted classes (aka labels) for a given classifier head. + + Attributes: + categories: The array of predicted categories, usually sorted by descending + scores (e.g. from high to low probability). + timestamp_ms: The optional timestamp (in milliseconds) associated to the + classification entry. This is useful for time series use cases, e.g., + audio classification. + """ + + categories: List[category_module.Category] + timestamp_ms: Optional[int] = None + + @doc_controls.do_not_generate_docs + def to_pb2(self) -> _ClassificationEntryProto: + """Generates a ClassificationEntry protobuf object.""" + return _ClassificationEntryProto( + categories=[category.to_pb2() for category in self.categories], + timestamp_ms=self.timestamp_ms) + + @classmethod + @doc_controls.do_not_generate_docs + def create_from_pb2( + cls, pb2_obj: _ClassificationEntryProto) -> 'ClassificationEntry': + """Creates a `ClassificationEntry` object from the given protobuf object.""" + return ClassificationEntry( + categories=[ + category_module.Category.create_from_pb2(category) + for category in pb2_obj.categories + ], + timestamp_ms=pb2_obj.timestamp_ms) + + def __eq__(self, other: Any) -> bool: + """Checks if this object is equal to the given object. + + Args: + other: The object to be compared with. + + Returns: + True if the objects are equal. + """ + if not isinstance(other, ClassificationEntry): + return False + + return self.to_pb2().__eq__(other.to_pb2()) + + +@dataclasses.dataclass +class Classifications: + """Represents the classifications for a given classifier head. + + Attributes: + entries: A list of `ClassificationEntry` objects. + head_index: The index of the classifier head these categories refer to. + This is useful for multi-head models. + head_name: The name of the classifier head, which is the corresponding + tensor metadata name. + """ + + entries: List[ClassificationEntry] + head_index: int + head_name: str + + @doc_controls.do_not_generate_docs + def to_pb2(self) -> _ClassificationsProto: + """Generates a Classifications protobuf object.""" + return _ClassificationsProto( + entries=[entry.to_pb2() for entry in self.entries], + head_index=self.head_index, + head_name=self.head_name) + + @classmethod + @doc_controls.do_not_generate_docs + def create_from_pb2(cls, pb2_obj: _ClassificationsProto) -> 'Classifications': + """Creates a `Classifications` object from the given protobuf object.""" + return Classifications( + entries=[ + ClassificationEntry.create_from_pb2(entry) + for entry in pb2_obj.entries + ], + head_index=pb2_obj.head_index, + head_name=pb2_obj.head_name) + + def __eq__(self, other: Any) -> bool: + """Checks if this object is equal to the given object. + + Args: + other: The object to be compared with. + + Returns: + True if the objects are equal. + """ + if not isinstance(other, Classifications): + return False + + return self.to_pb2().__eq__(other.to_pb2()) + + +@dataclasses.dataclass +class ClassificationResult: + """Contains one set of results per classifier head. + + Attributes: + classifications: A list of `Classifications` objects. + """ + + classifications: List[Classifications] + + @doc_controls.do_not_generate_docs + def to_pb2(self) -> _ClassificationResultProto: + """Generates a ClassificationResult protobuf object.""" + return _ClassificationResultProto( + classifications=[ + classification.to_pb2() for classification in self.classifications + ]) + + @classmethod + @doc_controls.do_not_generate_docs + def create_from_pb2( + cls, pb2_obj: _ClassificationResultProto) -> 'ClassificationResult': + """Creates a `ClassificationResult` object from the given protobuf object.""" + return ClassificationResult( + classifications=[ + Classifications.create_from_pb2(classification) + for classification in pb2_obj.classifications + ]) + + def __eq__(self, other: Any) -> bool: + """Checks if this object is equal to the given object. + + Args: + other: The object to be compared with. + + Returns: + True if the objects are equal. + """ + if not isinstance(other, ClassificationResult): + return False + + return self.to_pb2().__eq__(other.to_pb2()) diff --git a/mediapipe/tasks/python/test/vision/BUILD b/mediapipe/tasks/python/test/vision/BUILD index bb495338d..a63c36b55 100644 --- a/mediapipe/tasks/python/test/vision/BUILD +++ b/mediapipe/tasks/python/test/vision/BUILD @@ -18,4 +18,40 @@ package(default_visibility = ["//mediapipe/tasks:internal"]) licenses(["notice"]) -# TODO: This test fails in OSS +py_test( + name = "object_detector_test", + srcs = ["object_detector_test.py"], + data = [ + "//mediapipe/tasks/testdata/vision:test_images", + "//mediapipe/tasks/testdata/vision:test_models", + ], + deps = [ + # build rule placeholder: numpy dep, + "//mediapipe/tasks/python/components/containers:bounding_box", + "//mediapipe/tasks/python/components/containers:category", + "//mediapipe/tasks/python/components/containers:detections", + "//mediapipe/tasks/python/core:base_options", + "//mediapipe/tasks/python/test:test_util", + "//mediapipe/tasks/python/vision:object_detector", + "//mediapipe/tasks/python/vision/core:vision_task_running_mode", + "@absl_py//absl/testing:parameterized", + ], +) + +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", + ], +) 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..3650c547c --- /dev/null +++ b/mediapipe/tasks/python/test/vision/image_classification_test.py @@ -0,0 +1,301 @@ +# 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.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 +_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.') + + options = _ImageClassifierOptions( + base_options=base_options, max_results=max_results) + 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.') + + options = _ImageClassifierOptions( + base_options=base_options, max_results=max_results) + 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): + options = _ImageClassifierOptions( + base_options=_BaseOptions(file_name=self.model_path), + score_threshold=_SCORE_THRESHOLD) + 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): + options = _ImageClassifierOptions( + base_options=_BaseOptions(file_name=self.model_path), + max_results=_MAX_RESULTS) + 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): + options = _ImageClassifierOptions( + base_options=_BaseOptions(file_name=self.model_path), + category_allowlist=_ALLOW_LIST) + 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): + options = _ImageClassifierOptions( + base_options=_BaseOptions(file_name=self.model_path), + category_denylist=_DENY_LIST) + 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.'): + options = _ImageClassifierOptions( + base_options=_BaseOptions(file_name=self.model_path), + category_allowlist=['foo'], + category_denylist=['bar']) + with _ImageClassifier.create_from_options(options) as unused_classifier: + pass + + def test_empty_classification_outputs(self): + options = _ImageClassifierOptions( + base_options=_BaseOptions(file_name=self.model_path), score_threshold=1) + 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 + + # @parameterized.parameters((0, _EXPECTED_CLASSIFICATION_RESULT), + # (1, _ClassificationResult(classifications=[]))) + # def test_classify_async_calls(self, threshold, expected_result): + # observed_timestamp_ms = -1 + # + # def check_result(result: _ClassificationResult, timestamp_ms: int): + # self.assertEqual(result, expected_result) + # self.assertLess(observed_timestamp_ms, timestamp_ms) + # self.observed_timestamp_ms = timestamp_ms + # + # options = _ImageClassifierOptions( + # base_options=_BaseOptions(file_name=self.model_path), + # running_mode=_RUNNING_MODE.LIVE_STREAM, + # max_results=4, + # score_threshold=threshold, + # result_callback=check_result) + # classifier = _ImageClassifier.create_from_options(options) + # for timestamp in range(0, 300, 30): + # classifier.classify_async(self.test_image, timestamp) + # classifier.close() + + +if __name__ == '__main__': + absltest.main() diff --git a/mediapipe/tasks/python/vision/BUILD b/mediapipe/tasks/python/vision/BUILD index 7ff818610..7a27da179 100644 --- a/mediapipe/tasks/python/vision/BUILD +++ b/mediapipe/tasks/python/vision/BUILD @@ -36,3 +36,23 @@ py_library( "//mediapipe/tasks/python/vision/core:vision_task_running_mode", ], ) + +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/components:classifier_options_py_pb2", + "//mediapipe/tasks/cc/vision/image_classification:image_classifier_options_py_pb2", + "//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/image_classification.py b/mediapipe/tasks/python/vision/image_classification.py new file mode 100644 index 000000000..efe6aa11d --- /dev/null +++ b/mediapipe/tasks/python/vision/image_classification.py @@ -0,0 +1,227 @@ +# 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.components import classifier_options_pb2 +from mediapipe.tasks.cc.vision.image_classification import image_classifier_options_pb2 +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 +_ClassifierOptionsProto = classifier_options_pb2.ClassifierOptions +_ImageClassifierOptionsProto = image_classifier_options_pb2.ImageClassifierOptions +_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 + display_names_locale: Optional[str] = None + max_results: Optional[int] = None + score_threshold: Optional[float] = None + category_allowlist: Optional[List[str]] = None + category_denylist: Optional[List[str]] = None + 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 = _ClassifierOptionsProto( + display_names_locale=self.display_names_locale, + max_results=self.max_results, + score_threshold=self.score_threshold, + category_allowlist=self.category_allowlist, + category_denylist=self.category_denylist) + + 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 + ]) + + def classify_async(self, image: image_module.Image, timestamp_ms: int) -> None: + """Sends live image data (an Image with a unique timestamp) to perform image + classification. + + This method will return immediately after the input image is accepted. The + results will be available via the `result_callback` provided in the + `ImageClassifierOptions`. The `detect_async` method is designed to process + live stream data such as camera input. To lower the overall latency, image + classifier may drop the input images if needed. In other words, it's not + guaranteed to have output per input image. The `result_callback` provides: + - A classification result object that contains a list of classifications. + - The input image that the image classifier runs on. + - The input timestamp in milliseconds. + + Args: + image: MediaPipe Image. + timestamp_ms: The timestamp of the input image in milliseconds. + + Raises: + ValueError: If the current input timestamp is smaller than what the image + classifier has already processed. + """ + self._send_live_stream_data({ + _IMAGE_IN_STREAM_NAME: + packet_creator.create_image(image).at(timestamp_ms) + })