Fix some doc

PiperOrigin-RevId: 419717366

Author: wangtz

lite/codelabs/digit_classifier/android/finish/gradle/wrapper/gradle-wrapper.properties

@@ -1,5 +1,5 @@
 distributionBase=GRADLE_USER_HOME
 distributionPath=wrapper/dists
-distributionUrl=https\://services.gradle.org/distributions/gradle-6.8.3-bin.zip
+distributionUrl=https\://services.gradle.org/distributions/gradle-7.3.2-bin.zip
 zipStoreBase=GRADLE_USER_HOME
 zipStorePath=wrapper/dists

lite/codelabs/digit_classifier/android/start/gradle/wrapper/gradle-wrapper.properties

@@ -1,5 +1,5 @@
 distributionBase=GRADLE_USER_HOME
 distributionPath=wrapper/dists
-distributionUrl=https\://services.gradle.org/distributions/gradle-6.8.3-bin.zip
+distributionUrl=https\://services.gradle.org/distributions/gradle-7.3.2-bin.zip
 zipStoreBase=GRADLE_USER_HOME
 zipStorePath=wrapper/dists

lite/codelabs/flower_classification/android/finish/gradle/wrapper/gradle-wrapper.properties

@@ -1,5 +1,5 @@
 distributionBase=GRADLE_USER_HOME
 distributionPath=wrapper/dists
-distributionUrl=https\://services.gradle.org/distributions/gradle-6.8.3-bin.zip
+distributionUrl=https\://services.gradle.org/distributions/gradle-7.3.2-bin.zip
 zipStoreBase=GRADLE_USER_HOME
 zipStorePath=wrapper/dists

lite/codelabs/flower_classification/android/start/gradle/wrapper/gradle-wrapper.properties

@@ -1,5 +1,5 @@
 distributionBase=GRADLE_USER_HOME
 distributionPath=wrapper/dists
-distributionUrl=https\://services.gradle.org/distributions/gradle-6.8.3-bin.zip
+distributionUrl=https\://services.gradle.org/distributions/gradle-7.3.2-bin.zip
 zipStoreBase=GRADLE_USER_HOME
 zipStorePath=wrapper/dists

lite/examples/bert_qa/android/gradle/wrapper/gradle-wrapper.properties

@@ -1,5 +1,5 @@
 distributionBase=GRADLE_USER_HOME
 distributionPath=wrapper/dists
-distributionUrl=https\://services.gradle.org/distributions/gradle-6.8.3-bin.zip
+distributionUrl=https\://services.gradle.org/distributions/gradle-7.3.2-bin.zip
 zipStoreBase=GRADLE_USER_HOME
 zipStorePath=wrapper/dists

lite/examples/digit_classifier/android/gradle/wrapper/gradle-wrapper.properties

@@ -1,5 +1,5 @@
 distributionBase=GRADLE_USER_HOME
 distributionPath=wrapper/dists
-distributionUrl=https\://services.gradle.org/distributions/gradle-6.8.3-bin.zip
+distributionUrl=https\://services.gradle.org/distributions/gradle-7.3.2-bin.zip
 zipStoreBase=GRADLE_USER_HOME
 zipStorePath=wrapper/dists

lite/examples/gesture_classification/android/gradle/wrapper/gradle-wrapper.properties

@@ -1,5 +1,5 @@
 distributionBase=GRADLE_USER_HOME
 distributionPath=wrapper/dists
-distributionUrl=https\://services.gradle.org/distributions/gradle-6.8.3-bin.zip
+distributionUrl=https\://services.gradle.org/distributions/gradle-7.3.2-bin.zip
 zipStoreBase=GRADLE_USER_HOME
 zipStorePath=wrapper/dists

lite/examples/image_classification/android/gradle/wrapper/gradle-wrapper.properties

@@ -1,5 +1,5 @@
 distributionBase=GRADLE_USER_HOME
 distributionPath=wrapper/dists
-distributionUrl=https\://services.gradle.org/distributions/gradle-6.8.3-bin.zip
+distributionUrl=https\://services.gradle.org/distributions/gradle-7.3.2-bin.zip
 zipStoreBase=GRADLE_USER_HOME
 zipStorePath=wrapper/dists

lite/examples/image_classification/raspberry_pi/README.md

@@ -5,7 +5,7 @@ on a Raspberry Pi to perform real-time image classification using images
 streamed from the camera.
 
 At the end of this page, there are extra steps to accelerate the example using
-the Coral USB Accelerator, which increases the inference speed by ~10x.
+the Coral USB Accelerator to increase inference speed.
 
 
 ## Set up your hardware

lite/examples/image_classification/raspberry_pi/image_classifier.py

@@ -14,12 +14,13 @@
 """A wrapper for TensorFlow Lite image classification models."""
 
 import dataclasses
+import json
 import platform
 from typing import List
-import zipfile
 
 import cv2
 import numpy as np
+from tflite_support import metadata
 
 # pylint: disable=g-import-not-at-top
 try:
@@ -76,11 +77,6 @@ def edgetpu_lib_name():
 class ImageClassifier(object):
   """A wrapper class for a TFLite image classification model."""
 
-  _mean = 127
-  """Default mean normalization parameter for float model."""
-  _std = 128
-  """Default std normalization parameter for float model."""
-
   def __init__(
       self,
       model_path: str,
@@ -96,21 +92,27 @@ def __init__(
         ValueError: If the TFLite model is invalid.
         OSError: If the current OS isn't supported by EdgeTPU.
     """
+    # Load metadata from model.
+    displayer = metadata.MetadataDisplayer.with_model_file(model_path)
+
+    # Save model metadata for preprocessing later.
+    model_metadata = json.loads(displayer.get_metadata_json())
+    process_units = model_metadata['subgraph_metadata'][0][
+        'input_tensor_metadata'][0]['process_units']
+    mean = 127.5
+    std = 127.5
+    for option in process_units:
+      if option['options_type'] == 'NormalizationOptions':
+        mean = option['options']['mean'][0]
+        std = option['options']['std'][0]
+    self._mean = mean
+    self._std = std
+
     # Load label list from metadata.
-    try:
-      with zipfile.ZipFile(model_path) as model_with_metadata:
-        if not model_with_metadata.namelist():
-          raise ValueError('Invalid TFLite model: no label file found.')
-
-        file_name = model_with_metadata.namelist()[0]
-        with model_with_metadata.open(file_name) as label_file:
-          label_list = label_file.read().splitlines()
-          self._labels_list = [label.decode('ascii') for label in label_list]
-    except zipfile.BadZipFile:
-      print(
-          'ERROR: Please use models trained with Model Maker or downloaded from TensorFlow Hub.'
-      )
-      raise ValueError('Invalid TFLite model: no metadata found.')
+    file_name = displayer.get_packed_associated_file_list()[0]
+    label_map_file = displayer.get_associated_file_buffer(file_name).decode()
+    label_list = list(filter(len, label_map_file.splitlines()))
+    self._label_list = label_list
 
     # Initialize TFLite model.
     if options.enable_edgetpu:
@@ -191,7 +193,7 @@ def _postprocess(self, output_tensor: np.ndarray) -> List[Category]:
         range(len(output_tensor)), key=lambda k: output_tensor[k], reverse=True)
 
     categories = [
-        Category(label=self._labels_list[idx], score=output_tensor[idx])
+        Category(label=self._label_list[idx], score=output_tensor[idx])
         for idx in prob_descending
     ]
 

lite/examples/object_detection/raspberry_pi/requirements_pypi.txt renamed to lite/examples/image_classification/raspberry_pi/requirements.txt

@@ -1,3 +1,5 @@
 argparse
 numpy>=1.20.0  # To ensure compatibility with OpenCV on Raspberry Pi.
 opencv-python~=4.5.3.56
+tflite-runtime>=2.7.0
+tflite-support>=0.3.1

lite/examples/image_classification/raspberry_pi/requirements_pypi.txt

@@ -7,8 +7,8 @@ else
 fi
 
 # Install Python dependencies.
-python3 -m pip install -r requirements_pypi.txt
-python3 -m pip install -r requirements_tflite.txt
+python3 -m pip install pip --upgrade
+python3 -m pip install -r requirements.txt
 
 # Download TF Lite model with metadata.
 FILE=${DATA_DIR}/efficientnet_lite0.tflite
@@ -24,4 +24,4 @@ if [ ! -f "$FILE" ]; then
     -L 'https://storage.googleapis.com/download.tensorflow.org/models/tflite/edgetpu/efficientnet-edgetpu-M_quant_edgetpu.tflite' \
     -o ${FILE}
 fi
-echo -e "Downloaded files are in ${DATA_DIR}"
+echo -e "Downloaded files are in ${DATA_DIR}"

lite/examples/image_classification/raspberry_pi/requirements_tflite.txt

@@ -1,5 +1,5 @@
 distributionBase=GRADLE_USER_HOME
 distributionPath=wrapper/dists
-distributionUrl=https\://services.gradle.org/distributions/gradle-6.8.3-bin.zip
+distributionUrl=https\://services.gradle.org/distributions/gradle-7.3.2-bin.zip
 zipStoreBase=GRADLE_USER_HOME
 zipStorePath=wrapper/dists

lite/examples/image_classification/raspberry_pi/setup.sh

@@ -5,7 +5,7 @@ a Raspberry Pi to perform real-time image segmentation using images streamed
 from the camera.
 
 At the end of this page, there are extra steps to accelerate the example using
-the Coral USB Accelerator, which increases the inference speed by ~10x.
+the Coral USB Accelerator to increase inference speed.
 
 ## Set up your hardware
 

lite/examples/image_segmentation/android/gradle/wrapper/gradle-wrapper.properties

@@ -15,12 +15,13 @@
 
 import dataclasses
 import enum
+import json
 import platform
 from typing import List, Tuple
-import zipfile
 
 import cv2
 import numpy as np
+from tflite_support import metadata
 
 # pylint: disable=g-import-not-at-top
 try:
@@ -388,11 +389,6 @@ def _label_to_color_image(label: np.ndarray) -> np.ndarray:
 class ImageSegmenter(object):
   """A wrapper class for a TFLite image segmentation model."""
 
-  _mean = 127.5
-  """Default mean normalization parameter for float model."""
-  _std = 127.5
-  """Default std normalization parameter for float model."""
-
   def __init__(
       self,
       model_path: str,
@@ -408,21 +404,28 @@ def __init__(
         ValueError: If the TFLite model is invalid.
         OSError: If the current OS isn't supported by EdgeTPU.
     """
+    # Load metadata from model.
+    displayer = metadata.MetadataDisplayer.with_model_file(model_path)
+
+    # Save model metadata for preprocessing later.
+    model_metadata = json.loads(displayer.get_metadata_json())
+    process_units = model_metadata['subgraph_metadata'][0][
+        'input_tensor_metadata'][0]['process_units']
+
+    mean = 127.5
+    std = 127.5
+    for option in process_units:
+      if option['options_type'] == 'NormalizationOptions':
+        mean = option['options']['mean'][0]
+        std = option['options']['std'][0]
+    self._mean = mean
+    self._std = std
+
     # Load label list from metadata.
-    try:
-      with zipfile.ZipFile(model_path) as model_with_metadata:
-        if not model_with_metadata.namelist():
-          raise ValueError('Invalid TFLite model: no label file found.')
-
-        file_name = model_with_metadata.namelist()[0]
-        with model_with_metadata.open(file_name) as label_file:
-          label_list = label_file.read().splitlines()
-          self._label_list = [label.decode('ascii') for label in label_list]
-    except zipfile.BadZipFile:
-      print(
-          'ERROR: Please use models trained with Model Maker or downloaded from TensorFlow Hub.'
-      )
-      raise ValueError('Invalid TFLite model: no metadata found.')
+    file_name = displayer.get_packed_associated_file_list()[0]
+    label_map_file = displayer.get_associated_file_buffer(file_name).decode()
+    label_list = list(filter(len, label_map_file.splitlines()))
+    self._label_list = label_list
 
     # Initialize TFLite model.
     if options.enable_edgetpu:

lite/examples/image_segmentation/raspberry_pi/README.md

@@ -0,0 +1,5 @@
+argparse
+numpy>=1.20.0  # To ensure compatibility with OpenCV on Raspberry Pi.
+opencv-python~=4.5.3.56
+tflite-runtime>=2.7.0
+tflite-support>=0.3.1

lite/examples/image_segmentation/raspberry_pi/image_segmenter.py

@@ -7,8 +7,8 @@ else
 fi
 
 # Install Python dependencies
-python3 -m pip install -r requirements_pypi.txt
-python3 -m pip install -r requirements_tflite.txt
+python3 -m pip install pip --upgrade
+python3 -m pip install -r requirements.txt
 
 # Download TF Lite models with metadata.
 FILE=${DATA_DIR}/deeplabv3.tflite
@@ -25,4 +25,4 @@ if [ ! -f "$FILE" ]; then
     -o ${FILE}
 fi
 
-echo -e "Downloaded files are in ${DATA_DIR}"
+echo -e "Downloaded files are in ${DATA_DIR}"

lite/examples/image_segmentation/raspberry_pi/requirements.txt

@@ -1,5 +1,5 @@
 distributionBase=GRADLE_USER_HOME
 distributionPath=wrapper/dists
-distributionUrl=https\://services.gradle.org/distributions/gradle-6.8.3-bin.zip
+distributionUrl=https\://services.gradle.org/distributions/gradle-7.3.2-bin.zip
 zipStoreBase=GRADLE_USER_HOME
 zipStorePath=wrapper/dists

lite/examples/image_segmentation/raspberry_pi/requirements_pypi.txt

@@ -1,5 +1,5 @@
 distributionBase=GRADLE_USER_HOME
 distributionPath=wrapper/dists
-distributionUrl=https\://services.gradle.org/distributions/gradle-6.8.3-bin.zip
+distributionUrl=https\://services.gradle.org/distributions/gradle-7.3.2-bin.zip
 zipStoreBase=GRADLE_USER_HOME
 zipStorePath=wrapper/dists

lite/examples/image_segmentation/raspberry_pi/requirements_tflite.txt

@@ -6,7 +6,7 @@ the Pi Camera. It draws a bounding box around each detected object in the camera
 preview (when the object score is above a given threshold).
 
 At the end of this page, there are extra steps to accelerate the example using
-the Coral USB Accelerator, which increases the inference speed by ~10x.
+the Coral USB Accelerator to increase inference speed.
 
 ## Set up your hardware
 

lite/examples/image_segmentation/raspberry_pi/setup.sh

@@ -13,12 +13,13 @@
 # limitations under the License.
 """A module to run object detection with a TensorFlow Lite model."""
 
+import json
 import platform
 from typing import List, NamedTuple
-import zipfile
 
 import cv2
 import numpy as np
+from tflite_support import metadata
 
 # pylint: disable=g-import-not-at-top
 try:
@@ -90,11 +91,6 @@ def edgetpu_lib_name():
 class ObjectDetector:
   """A wrapper class for a TFLite object detection model."""
 
-  _mean = 127.5
-  """Default mean normalization parameter for float model."""
-  _std = 127.5
-  """Default std normalization parameter for float model."""
-
   _OUTPUT_LOCATION_NAME = 'location'
   _OUTPUT_CATEGORY_NAME = 'category'
   _OUTPUT_SCORE_NAME = 'score'
@@ -116,21 +112,27 @@ def __init__(
         OSError: If the current OS isn't supported by EdgeTPU.
     """
 
+    # Load metadata from model.
+    displayer = metadata.MetadataDisplayer.with_model_file(model_path)
+
+    # Save model metadata for preprocessing later.
+    model_metadata = json.loads(displayer.get_metadata_json())
+    process_units = model_metadata['subgraph_metadata'][0][
+        'input_tensor_metadata'][0]['process_units']
+    mean = 127.5
+    std = 127.5
+    for option in process_units:
+      if option['options_type'] == 'NormalizationOptions':
+        mean = option['options']['mean'][0]
+        std = option['options']['std'][0]
+    self._mean = mean
+    self._std = std
+
     # Load label list from metadata.
-    try:
-      with zipfile.ZipFile(model_path) as model_with_metadata:
-        if not model_with_metadata.namelist():
-          raise ValueError('Invalid TFLite model: no label file found.')
-
-        file_name = model_with_metadata.namelist()[0]
-        with model_with_metadata.open(file_name) as label_file:
-          label_list = label_file.read().splitlines()
-          self._label_list = [label.decode('ascii') for label in label_list]
-    except zipfile.BadZipFile:
-      print(
-          'ERROR: Please use models trained with Model Maker or downloaded from TensorFlow Hub.'
-      )
-      raise ValueError('Invalid TFLite model: no metadata found.')
+    file_name = displayer.get_packed_associated_file_list()[0]
+    label_map_file = displayer.get_associated_file_buffer(file_name).decode()
+    label_list = list(filter(len, label_map_file.splitlines()))
+    self._label_list = label_list
 
     # Initialize TFLite model.
     if options.enable_edgetpu:

lite/examples/model_personalization/android/gradle/wrapper/gradle-wrapper.properties

@@ -0,0 +1,5 @@
+argparse
+numpy>=1.20.0  # To ensure compatibility with OpenCV on Raspberry Pi.
+opencv-python~=4.5.3.56
+tflite-runtime>=2.7.0
+tflite-support>=0.3.1