Add metadata writer to image_classifier in model_maker

PiperOrigin-RevId: 485926985

mediapipe/model_maker/python/core/tasks/custom_model.py

@@ -52,6 +52,7 @@ class CustomModel(abc.ABC):
     """Prints a summary of the model."""
     self._model.summary()
 
+# TODO: Remove this method when all tasks use Metadata writer
   def export_tflite(
       self,
       export_dir: str,
@@ -62,7 +63,7 @@ class CustomModel(abc.ABC):
 
     Args:
       export_dir: The directory to save exported files.
-      tflite_filename: File name to save tflite model. The full export path is
+      tflite_filename: File name to save TFLite model. The full export path is
         {export_dir}/{tflite_filename}.
       quantization_config: The configuration for model quantization.
       preprocess: A callable to preprocess the representative dataset for
@@ -73,11 +74,11 @@ class CustomModel(abc.ABC):
       tf.io.gfile.makedirs(export_dir)
 
     tflite_filepath = os.path.join(export_dir, tflite_filename)
-    # TODO: Populate metadata to the exported TFLite model.
-    model_util.export_tflite(
+    tflite_model = model_util.convert_to_tflite(
         model=self._model,
-        tflite_filepath=tflite_filepath,
         quantization_config=quantization_config,
         preprocess=preprocess)
+    model_util.save_tflite(
+        tflite_model=tflite_model, tflite_file=tflite_filepath)
     tf.compat.v1.logging.info(
         'TensorFlow Lite model exported successfully: %s' % tflite_filepath)

mediapipe/model_maker/python/core/utils/model_util.py

@@ -89,28 +89,25 @@ def get_steps_per_epoch(steps_per_epoch: Optional[int] = None,
     return len(train_data) // batch_size
 
 
-def export_tflite(
+def convert_to_tflite(
     model: tf.keras.Model,
-    tflite_filepath: str,
     quantization_config: Optional[quantization.QuantizationConfig] = None,
     supported_ops: Tuple[tf.lite.OpsSet,
                          ...] = (tf.lite.OpsSet.TFLITE_BUILTINS,),
-    preprocess: Optional[Callable[..., bool]] = None):
-  """Converts the model to tflite format and saves it.
+    preprocess: Optional[Callable[..., bool]] = None) -> bytearray:
+  """Converts the input Keras model to TFLite format.
 
   Args:
-    model: model to be converted to tflite.
-    tflite_filepath: File path to save tflite model.
+    model: Keras model to be converted to TFLite.
     quantization_config: Configuration for post-training quantization.
     supported_ops: A list of supported ops in the converted TFLite file.
     preprocess: A callable to preprocess the representative dataset for
       quantization. The callable takes three arguments in order: feature, label,
       and is_training.
-  """
-  if tflite_filepath is None:
-    raise ValueError(
-        "TFLite filepath couldn't be None when exporting to tflite.")
 
+  Returns:
+    bytearray of TFLite model
+  """
   with tempfile.TemporaryDirectory() as temp_dir:
     save_path = os.path.join(temp_dir, 'saved_model')
     model.save(save_path, include_optimizer=False, save_format='tf')
@@ -122,9 +119,22 @@ def export_tflite(
 
     converter.target_spec.supported_ops = supported_ops
     tflite_model = converter.convert()
+    return tflite_model
 
-  with tf.io.gfile.GFile(tflite_filepath, 'wb') as f:
+
+def save_tflite(tflite_model: bytearray, tflite_file: str) -> None:
+  """Saves TFLite file to tflite_file.
+
+  Args:
+    tflite_model: A valid flatbuffer representing the TFLite model.
+    tflite_file: File path to save TFLite model.
+  """
+  if tflite_file is None:
+    raise ValueError("TFLite filepath can't be None when exporting to TFLite.")
+  with tf.io.gfile.GFile(tflite_file, 'wb') as f:
     f.write(tflite_model)
+  tf.compat.v1.logging.info(
+      'TensorFlow Lite model exported successfully to: %s' % tflite_file)
 
 
 class WarmUp(tf.keras.optimizers.schedules.LearningRateSchedule):
@@ -176,14 +186,12 @@ class WarmUp(tf.keras.optimizers.schedules.LearningRateSchedule):
 class LiteRunner(object):
   """A runner to do inference with the TFLite model."""
 
-  def __init__(self, tflite_filepath: str):
-    """Initializes Lite runner with tflite model file.
+  def __init__(self, tflite_model: bytearray):
+    """Initializes Lite runner from TFLite model buffer.
 
     Args:
-      tflite_filepath: File path to the TFLite model.
+      tflite_model: A valid flatbuffer representing the TFLite model.
     """
-    with tf.io.gfile.GFile(tflite_filepath, 'rb') as f:
-      tflite_model = f.read()
     self.interpreter = tf.lite.Interpreter(model_content=tflite_model)
     self.interpreter.allocate_tensors()
     self.input_details = self.interpreter.get_input_details()
@@ -250,9 +258,9 @@ class LiteRunner(object):
     return output_tensors
 
 
-def get_lite_runner(tflite_filepath: str) -> 'LiteRunner':
-  """Returns a `LiteRunner` from file path to TFLite model."""
-  lite_runner = LiteRunner(tflite_filepath)
+def get_lite_runner(tflite_buffer: bytearray) -> 'LiteRunner':
+  """Returns a `LiteRunner` from flatbuffer of the TFLite model."""
+  lite_runner = LiteRunner(tflite_buffer)
   return lite_runner
 
 

mediapipe/model_maker/python/core/utils/model_util_test.py

@@ -95,13 +95,12 @@ class ModelUtilTest(tf.test.TestCase, parameterized.TestCase):
             'name': 'test'
         })
 
-  def test_export_tflite(self):
+  def test_convert_to_tflite(self):
     input_dim = 4
     model = test_util.build_model(input_shape=[input_dim], num_classes=2)
-    tflite_file = os.path.join(self.get_temp_dir(), 'model.tflite')
-    model_util.export_tflite(model, tflite_file)
+    tflite_model = model_util.convert_to_tflite(model)
     test_util.test_tflite(
-        keras_model=model, tflite_file=tflite_file, size=[1, input_dim])
+        keras_model=model, tflite_model=tflite_model, size=[1, input_dim])
 
   @parameterized.named_parameters(
       dict(
@@ -118,25 +117,32 @@ class ModelUtilTest(tf.test.TestCase, parameterized.TestCase):
           testcase_name='float16_quantize',
           config=quantization.QuantizationConfig.for_float16(),
           model_size=1468))
-  def test_export_tflite_quantized(self, config, model_size):
+  def test_convert_to_tflite_quantized(self, config, model_size):
     input_dim = 16
     num_classes = 2
     max_input_value = 5
     model = test_util.build_model(
         input_shape=[input_dim], num_classes=num_classes)
-    tflite_file = os.path.join(self.get_temp_dir(), 'model_quantized.tflite')
 
-    model_util.export_tflite(
-        model=model, tflite_filepath=tflite_file, quantization_config=config)
+    tflite_model = model_util.convert_to_tflite(
+        model=model, quantization_config=config)
     self.assertTrue(
         test_util.test_tflite(
             keras_model=model,
-            tflite_file=tflite_file,
+            tflite_model=tflite_model,
             size=[1, input_dim],
             high=max_input_value,
             atol=1e-00))
-    self.assertNear(os.path.getsize(tflite_file), model_size, 300)
+    self.assertNear(len(tflite_model), model_size, 300)
 
+  def test_save_tflite(self):
+    input_dim = 4
+    model = test_util.build_model(input_shape=[input_dim], num_classes=2)
+    tflite_model = model_util.convert_to_tflite(model)
+    tflite_file = os.path.join(self.get_temp_dir(), 'model.tflite')
+    model_util.save_tflite(tflite_model=tflite_model, tflite_file=tflite_file)
+    test_util.test_tflite_file(
+        keras_model=model, tflite_file=tflite_file, size=[1, input_dim])
 
 if __name__ == '__main__':
   tf.test.main()

mediapipe/model_maker/python/core/utils/test_util.py

@@ -79,13 +79,13 @@ def build_model(input_shape: List[int], num_classes: int) -> tf.keras.Model:
   return model
 
 
-def is_same_output(tflite_file: str,
+def is_same_output(tflite_model: bytearray,
                    keras_model: tf.keras.Model,
                    input_tensors: Union[List[tf.Tensor], tf.Tensor],
                    atol: float = 1e-04) -> bool:
   """Returns if the output of TFLite model and keras model are identical."""
   # Gets output from lite model.
-  lite_runner = model_util.get_lite_runner(tflite_file)
+  lite_runner = model_util.get_lite_runner(tflite_model)
   lite_output = lite_runner.run(input_tensors)
 
   # Gets output from keras model.
@@ -95,12 +95,41 @@ def is_same_output(tflite_file: str,
 
 
 def test_tflite(keras_model: tf.keras.Model,
-                tflite_file: str,
+                tflite_model: bytearray,
                 size: Union[int, List[int]],
                 high: float = 1,
                 atol: float = 1e-04) -> bool:
   """Verifies if the output of TFLite model and TF Keras model are identical.
 
+  Args:
+    keras_model: Input TensorFlow Keras model.
+    tflite_model: Input TFLite model flatbuffer.
+    size: Size of the input tesnor.
+    high: Higher boundary of the values in input tensors.
+    atol: Absolute tolerance of the difference between the outputs of Keras
+      model and TFLite model.
+
+  Returns:
+    True if the output of TFLite model and TF Keras model are identical.
+    Otherwise, False.
+  """
+  random_input = create_random_sample(size=size, high=high)
+  random_input = tf.convert_to_tensor(random_input)
+
+  return is_same_output(
+      tflite_model=tflite_model,
+      keras_model=keras_model,
+      input_tensors=random_input,
+      atol=atol)
+
+
+def test_tflite_file(keras_model: tf.keras.Model,
+                     tflite_file: bytearray,
+                     size: Union[int, List[int]],
+                     high: float = 1,
+                     atol: float = 1e-04) -> bool:
+  """Verifies if the output of TFLite model and TF Keras model are identical.
+
   Args:
     keras_model: Input TensorFlow Keras model.
     tflite_file: Input TFLite model file.
@@ -113,11 +142,6 @@ def test_tflite(keras_model: tf.keras.Model,
     True if the output of TFLite model and TF Keras model are identical.
     Otherwise, False.
   """
-  random_input = create_random_sample(size=size, high=high)
-  random_input = tf.convert_to_tensor(random_input)
-
-  return is_same_output(
-      tflite_file=tflite_file,
-      keras_model=keras_model,
-      input_tensors=random_input,
-      atol=atol)
+  with tf.io.gfile.GFile(tflite_file, "rb") as f:
+    tflite_model = f.read()
+  return test_tflite(keras_model, tflite_model, size, high, atol)

mediapipe/model_maker/python/vision/image_classifier/BUILD

@@ -81,6 +81,8 @@ py_library(
         "//mediapipe/model_maker/python/core/utils:model_util",
         "//mediapipe/model_maker/python/core/utils:quantization",
         "//mediapipe/model_maker/python/vision/core:image_preprocessing",
+        "//mediapipe/tasks/python/metadata/metadata_writers:image_classifier",
+        "//mediapipe/tasks/python/metadata/metadata_writers:metadata_writer",
     ],
 )
 
@@ -88,7 +90,11 @@ py_library(
     name = "image_classifier_test_lib",
     testonly = 1,
     srcs = ["image_classifier_test.py"],
-    deps = [":image_classifier_import"],
+    data = ["//mediapipe/model_maker/python/vision/image_classifier/testdata"],
+    deps = [
+        ":image_classifier_import",
+        "//mediapipe/tasks/python/test:test_utils",
+    ],
 )
 
 py_test(

mediapipe/model_maker/python/vision/image_classifier/image_classifier.py

@@ -12,6 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """APIs to train image classifier model."""
+import os
 
 from typing import List, Optional
 
@@ -26,6 +27,8 @@ from mediapipe.model_maker.python.vision.core import image_preprocessing
 from mediapipe.model_maker.python.vision.image_classifier import hyperparameters as hp
 from mediapipe.model_maker.python.vision.image_classifier import model_spec as ms
 from mediapipe.model_maker.python.vision.image_classifier import train_image_classifier_lib
+from mediapipe.tasks.python.metadata.metadata_writers import image_classifier as image_classifier_writer
+from mediapipe.tasks.python.metadata.metadata_writers import metadata_writer
 
 
 class ImageClassifier(classifier.Classifier):
@@ -156,15 +159,32 @@ class ImageClassifier(classifier.Classifier):
       self,
       model_name: str = 'model.tflite',
       quantization_config: Optional[quantization.QuantizationConfig] = None):
-    """Converts the model to the requested formats and exports to a file.
+    """Converts and saves the model to a TFLite file with metadata included.
+
+    Note that only the TFLite file is needed for deployment. This function also
+    saves a metadata.json file to the same directory as the TFLite file which
+    can be used to interpret the metadata content in the TFLite file.
 
     Args:
-      model_name: File name to save tflite model. The full export path is
-        {export_dir}/{tflite_filename}.
+      model_name: File name to save TFLite model with metadata. The full export
+        path is {self._hparams.model_dir}/{model_name}.
       quantization_config: The configuration for model quantization.
     """
-    super().export_tflite(
-        self._hparams.model_dir,
-        model_name,
-        quantization_config,
+    if not tf.io.gfile.exists(self._hparams.model_dir):
+      tf.io.gfile.makedirs(self._hparams.model_dir)
+    tflite_file = os.path.join(self._hparams.model_dir, model_name)
+    metadata_file = os.path.join(self._hparams.model_dir, 'metadata.json')
+
+    tflite_model = model_util.convert_to_tflite(
+        model=self._model,
+        quantization_config=quantization_config,
         preprocess=self._preprocess)
+    writer = image_classifier_writer.MetadataWriter.create(
+        tflite_model,
+        self._model_spec.mean_rgb,
+        self._model_spec.stddev_rgb,
+        labels=metadata_writer.Labels().add(self._label_names))
+    tflite_model_with_metadata, metadata_json = writer.populate()
+    model_util.save_tflite(tflite_model_with_metadata, tflite_file)
+    with open(metadata_file, 'w') as f:
+      f.write(metadata_json)

mediapipe/model_maker/python/vision/image_classifier/image_classifier_test.py

@@ -12,6 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
+import filecmp
 import os
 
 from absl.testing import parameterized
@@ -19,6 +20,7 @@ import numpy as np
 import tensorflow as tf
 
 from mediapipe.model_maker.python.vision import image_classifier
+from mediapipe.tasks.python.test import test_utils
 
 
 def _fill_image(rgb, image_size):
@@ -86,7 +88,7 @@ class ImageClassifierTest(tf.test.TestCase, parameterized.TestCase):
         validation_data=self.test_data)
     self._test_accuracy(model)
 
-  def test_efficientnetlite0_model_with_model_maker_retraining_lib(self):
+  def test_efficientnetlite0_model_train_and_export(self):
     hparams = image_classifier.HParams(
         train_epochs=1, batch_size=1, shuffle=True)
     model = image_classifier.ImageClassifier.create(
@@ -96,6 +98,19 @@ class ImageClassifierTest(tf.test.TestCase, parameterized.TestCase):
         validation_data=self.test_data)
     self._test_accuracy(model)
 
+    # Test export_model
+    model.export_model()
+    output_metadata_file = os.path.join(hparams.model_dir, 'metadata.json')
+    output_tflite_file = os.path.join(hparams.model_dir, 'model.tflite')
+    expected_metadata_file = test_utils.get_test_data_path('metadata.json')
+
+    self.assertTrue(os.path.exists(output_tflite_file))
+    self.assertGreater(os.path.getsize(output_tflite_file), 0)
+
+    self.assertTrue(os.path.exists(output_metadata_file))
+    self.assertGreater(os.path.getsize(output_metadata_file), 0)
+    self.assertTrue(filecmp.cmp(output_metadata_file, expected_metadata_file))
+
   def _test_accuracy(self, model, threshold=0.0):
     _, accuracy = model.evaluate(self.test_data)
     self.assertGreaterEqual(accuracy, threshold)

mediapipe/model_maker/python/vision/image_classifier/testdata/BUILD

@@ -0,0 +1,23 @@
+# 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.
+
+package(
+    default_visibility = ["//mediapipe/model_maker/python/vision/image_classifier:__subpackages__"],
+    licenses = ["notice"],  # Apache 2.0
+)
+
+filegroup(
+    name = "testdata",
+    srcs = ["metadata.json"],
+)

mediapipe/model_maker/python/vision/image_classifier/testdata/metadata.json

@@ -0,0 +1,68 @@
+{
+  "name": "ImageClassifier",
+  "description": "Identify the most prominent object in the image from a known set of categories.",
+  "subgraph_metadata": [
+    {
+      "input_tensor_metadata": [
+        {
+          "name": "image",
+          "description": "Input image to be processed.",
+          "content": {
+            "content_properties_type": "ImageProperties",
+            "content_properties": {
+              "color_space": "RGB"
+            }
+          },
+          "process_units": [
+            {
+              "options_type": "NormalizationOptions",
+              "options": {
+                "mean": [
+                  0.0
+                ],
+                "std": [
+                  255.0
+                ]
+              }
+            }
+          ],
+          "stats": {
+            "max": [
+              1.0
+            ],
+            "min": [
+              0.0
+            ]
+          }
+        }
+      ],
+      "output_tensor_metadata": [
+        {
+          "name": "score",
+          "description": "Score of the labels respectively.",
+          "content": {
+            "content_properties_type": "FeatureProperties",
+            "content_properties": {
+            }
+          },
+          "stats": {
+            "max": [
+              1.0
+            ],
+            "min": [
+              0.0
+            ]
+          },
+          "associated_files": [
+            {
+              "name": "labels.txt",
+              "description": "Labels for categories that the model can recognize.",
+              "type": "TENSOR_AXIS_LABELS"
+            }
+          ]
+        }
+      ]
+    }
+  ],
+  "min_parser_version": "1.0.0"
+}

mediapipe/tasks/python/test/BUILD

@@ -24,7 +24,7 @@ py_library(
     srcs = ["test_utils.py"],
     srcs_version = "PY3",
     visibility = [
-        "//mediapipe/model_maker/python/vision/gesture_recognizer:__pkg__",
+        "//mediapipe/model_maker/python:__subpackages__",
         "//mediapipe/tasks:internal",
     ],
     deps = [