 TensorFlow.org에서 보기 |
 Google Colab에서 실행 |
 GitHub에서 소스 보기 |
 노트북 다운로드 |
TensorFlow Lite 모델 메타데이터는 표준 모델 설명 형식입니다. 여기에는 일반 모델 정보, 입력/출력 및 관련 파일에 대한 풍부한 의미 체계가 포함되어 있어 모델을 이해하기 쉽고 교환 가능하게 만들 수 있습니다.
모델 메타데이터는 현재 다음 두 가지 주요 사용 사례에서 이용됩니다.
TensorFlow Lite 작업 라이브러리 및 codegen 도구를 사용하여 손쉬운 모델 추론이 가능합니다. 모델 메타데이터는 이미지 분류의 레이블 파일, 오디오 분류의 오디오 입력 샘플링 비율, 자연어 모델의 입력 문자열 처리를 위한 토크나이저 유형 등 추론 시 필요한 필수 정보를 포함합니다.
모델 작성자가 모델 입력/출력에 대한 설명 또는 모델 사용 방법과 같은 문서를 포함할 수 있습니다. 모델 사용자는 Netron과 같은 시각화 도구를 통해 이러한 문서를 볼 수 있습니다.
TensorFlow Lite Metadata Writer API는 사용하기 쉬운 API를 제공하여 TFLite 작업 라이브러리에서 지원하는 주된 ML 작업에 대한 모델 메타데이터를 생성합니다. 이 노트북은 아래의 다음 작업에 대해 메타데이터를 채우는 방법에 대한 예를 보여줍니다.
BERT 자연어 분류기 및 BERT 질문 답변자를 위한 메타데이터 작성자가 곧 제공될 예정입니다.
지원되지 않는 사용 사례에 대한 메타데이터를 추가하려면 Flatbuffers Python API를 사용하세요. 여기에서 튜토리얼을 참조하세요.
전제 조건
TensorFlow Lite 지원 Pypi 패키지를 설치합니다.
pip install tflite-support-nightly
작업 라이브러리 및 Codegen에 대한 모델 메타데이터 만들기
이미지 분류기
지원되는 모델 형식에 대한 자세한 내용은 이미지 분류기 모델 호환성 요구 사항을 참조하세요.
1단계: 필요한 패키지를 가져옵니다.
from tflite_support.metadata_writers import image_classifier
from tflite_support.metadata_writers import writer_utils
2022-12-14 20:32:09.693742: W tensorflow/compiler/xla/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libnvinfer.so.7'; dlerror: libnvinfer.so.7: cannot open shared object file: No such file or directory 2022-12-14 20:32:09.693841: W tensorflow/compiler/xla/stream_executor/platform/default/dso_loader.cc:64] Could not load dynamic library 'libnvinfer_plugin.so.7'; dlerror: libnvinfer_plugin.so.7: cannot open shared object file: No such file or directory 2022-12-14 20:32:09.693852: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Cannot dlopen some TensorRT libraries. If you would like to use Nvidia GPU with TensorRT, please make sure the missing libraries mentioned above are installed properly.
2단계: 예제 이미지 분류기 mobilenet_v2_1.0_224.tflite 및 레이블 파일을 다운로드합니다.
curl -L https://github.com/tensorflow/tflite-support/raw/master/tensorflow_lite_support/metadata/python/tests/testdata/image_classifier/mobilenet_v2_1.0_224.tflite -o mobilenet_v2_1.0_224.tflitecurl -L https://github.com/tensorflow/tflite-support/raw/master/tensorflow_lite_support/metadata/python/tests/testdata/image_classifier/labels.txt -o mobilenet_labels.txt
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 0
100 13.3M 100 13.3M 0 0 17.7M 0 --:--:-- --:--:-- --:--:-- 126M
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 0
100 10484 100 10484 0 0 26951 0 --:--:-- --:--:-- --:--:-- 26951
3단계: 메타데이터 작성기를 만들고 채웁니다.
ImageClassifierWriter = image_classifier.MetadataWriter
_MODEL_PATH = "mobilenet_v2_1.0_224.tflite"
# Task Library expects label files that are in the same format as the one below.
_LABEL_FILE = "mobilenet_labels.txt"
_SAVE_TO_PATH = "mobilenet_v2_1.0_224_metadata.tflite"
# Normalization parameters is required when reprocessing the image. It is
# optional if the image pixel values are in range of [0, 255] and the input
# tensor is quantized to uint8. See the introduction for normalization and
# quantization parameters below for more details.
# https://www.tensorflow.org/lite/models/convert/metadata#normalization_and_quantization_parameters)
_INPUT_NORM_MEAN = 127.5
_INPUT_NORM_STD = 127.5
# Create the metadata writer.
writer = ImageClassifierWriter.create_for_inference(
writer_utils.load_file(_MODEL_PATH), [_INPUT_NORM_MEAN], [_INPUT_NORM_STD],
[_LABEL_FILE])
# Verify the metadata generated by metadata writer.
print(writer.get_metadata_json())
# Populate the metadata into the model.
writer_utils.save_file(writer.populate(), _SAVE_TO_PATH)
{
"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 classified.",
"content": {
"content_properties_type": "ImageProperties",
"content_properties": {
"color_space": "RGB"
}
},
"process_units": [
{
"options_type": "NormalizationOptions",
"options": {
"mean": [
127.5
],
"std": [
127.5
]
}
}
],
"stats": {
"max": [
1.0
],
"min": [
-1.0
]
}
}
],
"output_tensor_metadata": [
{
"name": "probability",
"description": "Probabilities of the labels respectively.",
"content": {
"content_properties_type": "FeatureProperties",
"content_properties": {
}
},
"stats": {
"max": [
1.0
],
"min": [
0.0
]
},
"associated_files": [
{
"name": "mobilenet_labels.txt",
"description": "Labels for categories that the model can recognize.",
"type": "TENSOR_AXIS_LABELS"
}
]
}
]
}
]
}
물체 감지기
지원되는 모델 형식에 대한 자세한 내용은 물체 감지기 모델 호환성 요구 사항을 참조하세요.
1단계: 필요한 패키지를 가져옵니다.
from tflite_support.metadata_writers import object_detector
from tflite_support.metadata_writers import writer_utils
2단계: 예제 물체 감지기 ssd_mobilenet_v1.tflite 및 레이블 파일을 다운로드합니다.
curl -L https://github.com/tensorflow/tflite-support/raw/master/tensorflow_lite_support/metadata/python/tests/testdata/object_detector/ssd_mobilenet_v1.tflite -o ssd_mobilenet_v1.tflitecurl -L https://github.com/tensorflow/tflite-support/raw/master/tensorflow_lite_support/metadata/python/tests/testdata/object_detector/labelmap.txt -o ssd_mobilenet_labels.txt
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 0
100 4085k 100 4085k 0 0 3561k 0 0:00:01 0:00:01 --:--:-- 8978k
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 0
100 661 100 661 0 0 1703 0 --:--:-- --:--:-- --:--:-- 1703
3단계: 메타데이터 작성기를 만들고 채웁니다.
ObjectDetectorWriter = object_detector.MetadataWriter
_MODEL_PATH = "ssd_mobilenet_v1.tflite"
# Task Library expects label files that are in the same format as the one below.
_LABEL_FILE = "ssd_mobilenet_labels.txt"
_SAVE_TO_PATH = "ssd_mobilenet_v1_metadata.tflite"
# Normalization parameters is required when reprocessing the image. It is
# optional if the image pixel values are in range of [0, 255] and the input
# tensor is quantized to uint8. See the introduction for normalization and
# quantization parameters below for more details.
# https://www.tensorflow.org/lite/models/convert/metadata#normalization_and_quantization_parameters)
_INPUT_NORM_MEAN = 127.5
_INPUT_NORM_STD = 127.5
# Create the metadata writer.
writer = ObjectDetectorWriter.create_for_inference(
writer_utils.load_file(_MODEL_PATH), [_INPUT_NORM_MEAN], [_INPUT_NORM_STD],
[_LABEL_FILE])
# Verify the metadata generated by metadata writer.
print(writer.get_metadata_json())
# Populate the metadata into the model.
writer_utils.save_file(writer.populate(), _SAVE_TO_PATH)
{
"name": "ObjectDetector",
"description": "Identify which of a known set of objects might be present and provide information about their positions within the given image or a video stream.",
"subgraph_metadata": [
{
"input_tensor_metadata": [
{
"name": "image",
"description": "Input image to be detected.",
"content": {
"content_properties_type": "ImageProperties",
"content_properties": {
"color_space": "RGB"
}
},
"process_units": [
{
"options_type": "NormalizationOptions",
"options": {
"mean": [
127.5
],
"std": [
127.5
]
}
}
],
"stats": {
"max": [
255.0
],
"min": [
0.0
]
}
}
],
"output_tensor_metadata": [
{
"name": "location",
"description": "The locations of the detected boxes.",
"content": {
"content_properties_type": "BoundingBoxProperties",
"content_properties": {
"index": [
1,
0,
3,
2
],
"type": "BOUNDARIES"
},
"range": {
"min": 2,
"max": 2
}
},
"stats": {
}
},
{
"name": "category",
"description": "The categories of the detected boxes.",
"content": {
"content_properties_type": "FeatureProperties",
"content_properties": {
},
"range": {
"min": 2,
"max": 2
}
},
"stats": {
},
"associated_files": [
{
"name": "ssd_mobilenet_labels.txt",
"description": "Labels for categories that the model can recognize.",
"type": "TENSOR_VALUE_LABELS"
}
]
},
{
"name": "score",
"description": "The scores of the detected boxes.",
"content": {
"content_properties_type": "FeatureProperties",
"content_properties": {
},
"range": {
"min": 2,
"max": 2
}
},
"stats": {
}
},
{
"name": "number of detections",
"description": "The number of the detected boxes.",
"content": {
"content_properties_type": "FeatureProperties",
"content_properties": {
}
},
"stats": {
}
}
],
"output_tensor_groups": [
{
"name": "detection_result",
"tensor_names": [
"location",
"category",
"score"
]
}
]
}
]
}
이미지 분할기
지원되는 모델 형식에 대한 자세한 내용은 이미지 분할기 모델 호환성 요구 사항을 참조하세요.
1단계: 필요한 패키지를 가져옵니다.
from tflite_support.metadata_writers import image_segmenter
from tflite_support.metadata_writers import writer_utils
2단계: 예제 이미지 분할기 deeplabv3.tflite 및 레이블 파일을 다운로드합니다.
curl -L https://github.com/tensorflow/tflite-support/raw/master/tensorflow_lite_support/metadata/python/tests/testdata/image_segmenter/deeplabv3.tflite -o deeplabv3.tflitecurl -L https://github.com/tensorflow/tflite-support/raw/master/tensorflow_lite_support/metadata/python/tests/testdata/image_segmenter/labelmap.txt -o deeplabv3_labels.txt
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 0
100 2714k 100 2714k 0 0 2769k 0 --:--:-- --:--:-- --:--:-- 40.0M
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 0
100 141 100 141 0 0 453 0 --:--:-- --:--:-- --:--:-- 453
3단계: 메타데이터 작성기를 만들고 채웁니다.
ImageSegmenterWriter = image_segmenter.MetadataWriter
_MODEL_PATH = "deeplabv3.tflite"
# Task Library expects label files that are in the same format as the one below.
_LABEL_FILE = "deeplabv3_labels.txt"
_SAVE_TO_PATH = "deeplabv3_metadata.tflite"
# Normalization parameters is required when reprocessing the image. It is
# optional if the image pixel values are in range of [0, 255] and the input
# tensor is quantized to uint8. See the introduction for normalization and
# quantization parameters below for more details.
# https://www.tensorflow.org/lite/models/convert/metadata#normalization_and_quantization_parameters)
_INPUT_NORM_MEAN = 127.5
_INPUT_NORM_STD = 127.5
# Create the metadata writer.
writer = ImageSegmenterWriter.create_for_inference(
writer_utils.load_file(_MODEL_PATH), [_INPUT_NORM_MEAN], [_INPUT_NORM_STD],
[_LABEL_FILE])
# Verify the metadata generated by metadata writer.
print(writer.get_metadata_json())
# Populate the metadata into the model.
writer_utils.save_file(writer.populate(), _SAVE_TO_PATH)
{
"name": "ImageSegmenter",
"description": "Semantic image segmentation predicts whether each pixel of an image is associated with a certain class.",
"subgraph_metadata": [
{
"input_tensor_metadata": [
{
"name": "image",
"description": "Input image to be segmented.",
"content": {
"content_properties_type": "ImageProperties",
"content_properties": {
"color_space": "RGB"
}
},
"process_units": [
{
"options_type": "NormalizationOptions",
"options": {
"mean": [
127.5
],
"std": [
127.5
]
}
}
],
"stats": {
"max": [
1.0
],
"min": [
-1.0
]
}
}
],
"output_tensor_metadata": [
{
"name": "segmentation_masks",
"description": "Masks over the target objects with high accuracy.",
"content": {
"content_properties_type": "ImageProperties",
"content_properties": {
"color_space": "GRAYSCALE"
},
"range": {
"min": 1,
"max": 2
}
},
"stats": {
},
"associated_files": [
{
"name": "deeplabv3_labels.txt",
"description": "Labels for categories that the model can recognize.",
"type": "TENSOR_AXIS_LABELS"
}
]
}
]
}
]
}
지원되는 모델 형식에 대한 자세한 내용은 자연어 분류기 모델 호환성 요구 사항을 참조하세요.
1단계: 필요한 패키지를 가져옵니다.
from tflite_support.metadata_writers import nl_classifier
from tflite_support.metadata_writers import metadata_info
from tflite_support.metadata_writers import writer_utils
2단계: 예제 자연어 분류기 movie_review.tflite, 레이블 파일 및 어휘 파일을 다운로드합니다.
curl -L https://github.com/tensorflow/tflite-support/raw/master/tensorflow_lite_support/metadata/python/tests/testdata/nl_classifier/movie_review.tflite -o movie_review.tflitecurl -L https://github.com/tensorflow/tflite-support/raw/master/tensorflow_lite_support/metadata/python/tests/testdata/nl_classifier/labels.txt -o movie_review_labels.txtcurl -L https://storage.googleapis.com/download.tensorflow.org/models/tflite_support/nl_classifier/vocab.txt -o movie_review_vocab.txt
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 0
100 628k 100 628k 0 0 854k 0 --:--:-- --:--:-- --:--:-- 854k
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 0
100 17 100 17 0 0 52 0 --:--:-- --:--:-- --:--:-- 0
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
100 122k 100 122k 0 0 2978k 0 --:--:-- --:--:-- --:--:-- 2978k
3단계: 메타데이터 작성기를 만들고 채웁니다.
NLClassifierWriter = nl_classifier.MetadataWriter
_MODEL_PATH = "movie_review.tflite"
# Task Library expects label files and vocab files that are in the same formats
# as the ones below.
_LABEL_FILE = "movie_review_labels.txt"
_VOCAB_FILE = "movie_review_vocab.txt"
# NLClassifier supports tokenize input string using the regex tokenizer. See
# more details about how to set up RegexTokenizer below:
# https://github.com/tensorflow/tflite-support/blob/master/tensorflow_lite_support/metadata/python/metadata_writers/metadata_info.py#L130
_DELIM_REGEX_PATTERN = r"[^\w\']+"
_SAVE_TO_PATH = "moview_review_metadata.tflite"
# Create the metadata writer.
writer = nl_classifier.MetadataWriter.create_for_inference(
writer_utils.load_file(_MODEL_PATH),
metadata_info.RegexTokenizerMd(_DELIM_REGEX_PATTERN, _VOCAB_FILE),
[_LABEL_FILE])
# Verify the metadata generated by metadata writer.
print(writer.get_metadata_json())
# Populate the metadata into the model.
writer_utils.save_file(writer.populate(), _SAVE_TO_PATH)
{
"name": "NLClassifier",
"description": "Classify the input text into a set of known categories.",
"subgraph_metadata": [
{
"input_tensor_metadata": [
{
"name": "input_text",
"description": "Embedding vectors representing the input text to be classified.",
"content": {
"content_properties_type": "FeatureProperties",
"content_properties": {
}
},
"process_units": [
{
"options_type": "RegexTokenizerOptions",
"options": {
"delim_regex_pattern": "[^\\w\\']+",
"vocab_file": [
{
"name": "movie_review_vocab.txt",
"description": "Vocabulary file to convert natural language words to embedding vectors.",
"type": "VOCABULARY"
}
]
}
}
],
"stats": {
}
}
],
"output_tensor_metadata": [
{
"name": "probability",
"description": "Probabilities of the labels respectively.",
"content": {
"content_properties_type": "FeatureProperties",
"content_properties": {
}
},
"stats": {
"max": [
1.0
],
"min": [
0.0
]
},
"associated_files": [
{
"name": "movie_review_labels.txt",
"description": "Labels for categories that the model can recognize.",
"type": "TENSOR_AXIS_LABELS"
}
]
}
]
}
]
}
오디오 분류기
지원되는 모델 형식에 대한 자세한 내용은 오디오 분류기 모델 호환성 요구 사항을 참조하세요.
1단계: 필요한 패키지를 가져옵니다.
from tflite_support.metadata_writers import audio_classifier
from tflite_support.metadata_writers import metadata_info
from tflite_support.metadata_writers import writer_utils
2단계: 예제 오디오 분류기 yamnet.tflite 및 레이블 파일을 다운로드합니다.
curl -L https://github.com/tensorflow/tflite-support/raw/master/tensorflow_lite_support/metadata/python/tests/testdata/audio_classifier/yamnet_wavin_quantized_mel_relu6.tflite -o yamnet.tflitecurl -L https://github.com/tensorflow/tflite-support/raw/master/tensorflow_lite_support/metadata/python/tests/testdata/audio_classifier/yamnet_521_labels.txt -o yamnet_labels.txt
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 0
100 4022k 100 4022k 0 0 3409k 0 0:00:01 0:00:01 --:--:-- 3409k
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 0
100 6230 100 6230 0 0 19904 0 --:--:-- --:--:-- --:--:-- 0
3단계: 메타데이터 작성기를 만들고 채웁니다.
AudioClassifierWriter = audio_classifier.MetadataWriter
_MODEL_PATH = "yamnet.tflite"
# Task Library expects label files that are in the same format as the one below.
_LABEL_FILE = "yamnet_labels.txt"
# Expected sampling rate of the input audio buffer.
_SAMPLE_RATE = 16000
# Expected number of channels of the input audio buffer. Note, Task library only
# support single channel so far.
_CHANNELS = 1
_SAVE_TO_PATH = "yamnet_metadata.tflite"
# Create the metadata writer.
writer = AudioClassifierWriter.create_for_inference(
writer_utils.load_file(_MODEL_PATH), _SAMPLE_RATE, _CHANNELS, [_LABEL_FILE])
# Verify the metadata generated by metadata writer.
print(writer.get_metadata_json())
# Populate the metadata into the model.
writer_utils.save_file(writer.populate(), _SAVE_TO_PATH)
{
"name": "AudioClassifier",
"description": "Identify the most prominent type in the audio clip from a known set of categories.",
"subgraph_metadata": [
{
"input_tensor_metadata": [
{
"name": "audio_clip",
"description": "Input audio clip to be classified.",
"content": {
"content_properties_type": "AudioProperties",
"content_properties": {
"sample_rate": 16000,
"channels": 1
}
},
"stats": {
}
}
],
"output_tensor_metadata": [
{
"name": "probability",
"description": "Scores of the labels respectively.",
"content": {
"content_properties_type": "FeatureProperties",
"content_properties": {
}
},
"stats": {
"max": [
1.0
],
"min": [
0.0
]
},
"associated_files": [
{
"name": "yamnet_labels.txt",
"description": "Labels for categories that the model can recognize.",
"type": "TENSOR_AXIS_LABELS"
}
]
}
]
}
]
}
시맨틱 정보로 모델 메타데이터 생성하기
Metadata Writer API를 통해 모델 및 각 텐서에 대한 보다 자세한 설명 정보를 입력하여 모델 이해도를 높일 수 있습니다. 이를 위해 각 메타데이터 작성자의 'create_from_metadata_info' 메서드를 이용할 수 있습니다. 일반적으로 'create_from_metadata_info' 매개변수(예: general_md, input_md 및 output_md)를 통해 데이터를 채울 수 있습니다. 이미지 분류기를 위한 풍부한 모델 메타데이터를 생성하려면 아래 예를 참조하세요.
1단계: 필요한 패키지를 가져옵니다.
from tflite_support.metadata_writers import image_classifier
from tflite_support.metadata_writers import metadata_info
from tflite_support.metadata_writers import writer_utils
from tflite_support import metadata_schema_py_generated as _metadata_fb
2단계: 예제 이미지 분류기 mobilenet_v2_1.0_224.tflite 및 레이블 파일을 다운로드합니다.
curl -L https://github.com/tensorflow/tflite-support/raw/master/tensorflow_lite_support/metadata/python/tests/testdata/image_classifier/mobilenet_v2_1.0_224.tflite -o mobilenet_v2_1.0_224.tflitecurl -L https://github.com/tensorflow/tflite-support/raw/master/tensorflow_lite_support/metadata/python/tests/testdata/image_classifier/labels.txt -o mobilenet_labels.txt
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 0
100 13.3M 100 13.3M 0 0 46.9M 0 --:--:-- --:--:-- --:--:-- 46.9M
% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 0
100 10484 100 10484 0 0 82551 0 --:--:-- --:--:-- --:--:-- 82551
3단계: 모델 및 텐서 정보를 생성합니다.
model_buffer = writer_utils.load_file("mobilenet_v2_1.0_224.tflite")
# Create general model information.
general_md = metadata_info.GeneralMd(
name="ImageClassifier",
version="v1",
description=("Identify the most prominent object in the image from a "
"known set of categories."),
author="TensorFlow Lite",
licenses="Apache License. Version 2.0")
# Create input tensor information.
input_md = metadata_info.InputImageTensorMd(
name="input image",
description=("Input image to be classified. The expected image is "
"128 x 128, with three channels (red, blue, and green) per "
"pixel. Each element in the tensor is a value between min and "
"max, where (per-channel) min is [0] and max is [255]."),
norm_mean=[127.5],
norm_std=[127.5],
color_space_type=_metadata_fb.ColorSpaceType.RGB,
tensor_type=writer_utils.get_input_tensor_types(model_buffer)[0])
# Create output tensor information.
output_md = metadata_info.ClassificationTensorMd(
name="probability",
description="Probabilities of the 1001 labels respectively.",
label_files=[
metadata_info.LabelFileMd(file_path="mobilenet_labels.txt",
locale="en")
],
tensor_type=writer_utils.get_output_tensor_types(model_buffer)[0])
4단계: 메타데이터 작성기를 만들고 채웁니다.
ImageClassifierWriter = image_classifier.MetadataWriter
# Create the metadata writer.
writer = ImageClassifierWriter.create_from_metadata_info(
model_buffer, general_md, input_md, output_md)
# Verify the metadata generated by metadata writer.
print(writer.get_metadata_json())
# Populate the metadata into the model.
writer_utils.save_file(writer.populate(), _SAVE_TO_PATH)
{
"name": "ImageClassifier",
"description": "Identify the most prominent object in the image from a known set of categories.",
"version": "v1",
"subgraph_metadata": [
{
"input_tensor_metadata": [
{
"name": "input image",
"description": "Input image to be classified. The expected image is 128 x 128, with three channels (red, blue, and green) per pixel. Each element in the tensor is a value between min and max, where (per-channel) min is [0] and max is [255].",
"content": {
"content_properties_type": "ImageProperties",
"content_properties": {
"color_space": "RGB"
}
},
"process_units": [
{
"options_type": "NormalizationOptions",
"options": {
"mean": [
127.5
],
"std": [
127.5
]
}
}
],
"stats": {
"max": [
1.0
],
"min": [
-1.0
]
}
}
],
"output_tensor_metadata": [
{
"name": "probability",
"description": "Probabilities of the 1001 labels respectively.",
"content": {
"content_properties_type": "FeatureProperties",
"content_properties": {
}
},
"stats": {
"max": [
1.0
],
"min": [
0.0
]
},
"associated_files": [
{
"name": "mobilenet_labels.txt",
"description": "Labels for categories that the model can recognize.",
"type": "TENSOR_AXIS_LABELS",
"locale": "en"
}
]
}
]
}
],
"author": "TensorFlow Lite",
"license": "Apache License. Version 2.0"
}
모델에 채워진 메타데이터를 읽습니다.
다음 코드를 통해 TFLite 모델의 메타데이터 및 관련 파일을 표시할 수 있습니다.
from tflite_support import metadata
displayer = metadata.MetadataDisplayer.with_model_file("mobilenet_v2_1.0_224_metadata.tflite")
print("Metadata populated:")
print(displayer.get_metadata_json())
print("Associated file(s) populated:")
for file_name in displayer.get_packed_associated_file_list():
print("file name: ", file_name)
print("file content:")
print(displayer.get_associated_file_buffer(file_name))
Metadata populated:
{
"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 classified.",
"content": {
"content_properties_type": "ImageProperties",
"content_properties": {
"color_space": "RGB"
}
},
"process_units": [
{
"options_type": "NormalizationOptions",
"options": {
"mean": [
127.5
],
"std": [
127.5
]
}
}
],
"stats": {
"max": [
1.0
],
"min": [
-1.0
]
}
}
],
"output_tensor_metadata": [
{
"name": "probability",
"description": "Probabilities of the labels respectively.",
"content": {
"content_properties_type": "FeatureProperties",
"content_properties": {
}
},
"stats": {
"max": [
1.0
],
"min": [
0.0
]
},
"associated_files": [
{
"name": "mobilenet_labels.txt",
"description": "Labels for categories that the model can recognize.",
"type": "TENSOR_AXIS_LABELS"
}
]
}
]
}
],
"min_parser_version": "1.0.0"
}
Associated file(s) populated:
file name: mobilenet_labels.txt
file content:
b"background\ntench\ngoldfish\ngreat white shark\ntiger shark\nhammerhead\nelectric ray\nstingray\ncock\nhen\nostrich\nbrambling\ngoldfinch\nhouse finch\njunco\nindigo bunting\nrobin\nbulbul\njay\nmagpie\nchickadee\nwater ouzel\nkite\nbald eagle\nvulture\ngreat grey owl\nEuropean fire salamander\ncommon newt\neft\nspotted salamander\naxolotl\nbullfrog\ntree frog\ntailed frog\nloggerhead\nleatherback turtle\nmud turtle\nterrapin\nbox turtle\nbanded gecko\ncommon iguana\nAmerican chameleon\nwhiptail\nagama\nfrilled lizard\nalligator lizard\nGila monster\ngreen lizard\nAfrican chameleon\nKomodo dragon\nAfrican crocodile\nAmerican alligator\ntriceratops\nthunder snake\nringneck snake\nhognose snake\ngreen snake\nking snake\ngarter snake\nwater snake\nvine snake\nnight snake\nboa constrictor\nrock python\nIndian cobra\ngreen mamba\nsea snake\nhorned viper\ndiamondback\nsidewinder\ntrilobite\nharvestman\nscorpion\nblack and gold garden spider\nbarn spider\ngarden spider\nblack widow\ntarantula\nwolf spider\ntick\ncentipede\nblack grouse\nptarmigan\nruffed grouse\nprairie chicken\npeacock\nquail\npartridge\nAfrican grey\nmacaw\nsulphur-crested cockatoo\nlorikeet\ncoucal\nbee eater\nhornbill\nhummingbird\njacamar\ntoucan\ndrake\nred-breasted merganser\ngoose\nblack swan\ntusker\nechidna\nplatypus\nwallaby\nkoala\nwombat\njellyfish\nsea anemone\nbrain coral\nflatworm\nnematode\nconch\nsnail\nslug\nsea slug\nchiton\nchambered nautilus\nDungeness crab\nrock crab\nfiddler crab\nking crab\nAmerican lobster\nspiny lobster\ncrayfish\nhermit crab\nisopod\nwhite stork\nblack stork\nspoonbill\nflamingo\nlittle blue heron\nAmerican egret\nbittern\ncrane\nlimpkin\nEuropean gallinule\nAmerican coot\nbustard\nruddy turnstone\nred-backed sandpiper\nredshank\ndowitcher\noystercatcher\npelican\nking penguin\nalbatross\ngrey whale\nkiller whale\ndugong\nsea lion\nChihuahua\nJapanese spaniel\nMaltese dog\nPekinese\nShih-Tzu\nBlenheim spaniel\npapillon\ntoy terrier\nRhodesian ridgeback\nAfghan hound\nbasset\nbeagle\nbloodhound\nbluetick\nblack-and-tan coonhound\nWalker hound\nEnglish foxhound\nredbone\nborzoi\nIrish wolfhound\nItalian greyhound\nwhippet\nIbizan hound\nNorwegian elkhound\notterhound\nSaluki\nScottish deerhound\nWeimaraner\nStaffordshire bullterrier\nAmerican Staffordshire terrier\nBedlington terrier\nBorder terrier\nKerry blue terrier\nIrish terrier\nNorfolk terrier\nNorwich terrier\nYorkshire terrier\nwire-haired fox terrier\nLakeland terrier\nSealyham terrier\nAiredale\ncairn\nAustralian terrier\nDandie Dinmont\nBoston bull\nminiature schnauzer\ngiant schnauzer\nstandard schnauzer\nScotch terrier\nTibetan terrier\nsilky terrier\nsoft-coated wheaten terrier\nWest Highland white terrier\nLhasa\nflat-coated retriever\ncurly-coated retriever\ngolden retriever\nLabrador retriever\nChesapeake Bay retriever\nGerman short-haired pointer\nvizsla\nEnglish setter\nIrish setter\nGordon setter\nBrittany spaniel\nclumber\nEnglish springer\nWelsh springer spaniel\ncocker spaniel\nSussex spaniel\nIrish water spaniel\nkuvasz\nschipperke\ngroenendael\nmalinois\nbriard\nkelpie\nkomondor\nOld English sheepdog\nShetland sheepdog\ncollie\nBorder collie\nBouvier des Flandres\nRottweiler\nGerman shepherd\nDoberman\nminiature pinscher\nGreater Swiss Mountain dog\nBernese mountain dog\nAppenzeller\nEntleBucher\nboxer\nbull mastiff\nTibetan mastiff\nFrench bulldog\nGreat Dane\nSaint Bernard\nEskimo dog\nmalamute\nSiberian husky\ndalmatian\naffenpinscher\nbasenji\npug\nLeonberg\nNewfoundland\nGreat Pyrenees\nSamoyed\nPomeranian\nchow\nkeeshond\nBrabancon griffon\nPembroke\nCardigan\ntoy poodle\nminiature poodle\nstandard poodle\nMexican hairless\ntimber wolf\nwhite wolf\nred wolf\ncoyote\ndingo\ndhole\nAfrican hunting dog\nhyena\nred fox\nkit fox\nArctic fox\ngrey fox\ntabby\ntiger cat\nPersian cat\nSiamese cat\nEgyptian cat\ncougar\nlynx\nleopard\nsnow leopard\njaguar\nlion\ntiger\ncheetah\nbrown bear\nAmerican black bear\nice bear\nsloth bear\nmongoose\nmeerkat\ntiger beetle\nladybug\nground beetle\nlong-horned beetle\nleaf beetle\ndung beetle\nrhinoceros beetle\nweevil\nfly\nbee\nant\ngrasshopper\ncricket\nwalking stick\ncockroach\nmantis\ncicada\nleafhopper\nlacewing\ndragonfly\ndamselfly\nadmiral\nringlet\nmonarch\ncabbage butterfly\nsulphur butterfly\nlycaenid\nstarfish\nsea urchin\nsea cucumber\nwood rabbit\nhare\nAngora\nhamster\nporcupine\nfox squirrel\nmarmot\nbeaver\nguinea pig\nsorrel\nzebra\nhog\nwild boar\nwarthog\nhippopotamus\nox\nwater buffalo\nbison\nram\nbighorn\nibex\nhartebeest\nimpala\ngazelle\nArabian camel\nllama\nweasel\nmink\npolecat\nblack-footed ferret\notter\nskunk\nbadger\narmadillo\nthree-toed sloth\norangutan\ngorilla\nchimpanzee\ngibbon\nsiamang\nguenon\npatas\nbaboon\nmacaque\nlangur\ncolobus\nproboscis monkey\nmarmoset\ncapuchin\nhowler monkey\ntiti\nspider monkey\nsquirrel monkey\nMadagascar cat\nindri\nIndian elephant\nAfrican elephant\nlesser panda\ngiant panda\nbarracouta\neel\ncoho\nrock beauty\nanemone fish\nsturgeon\ngar\nlionfish\npuffer\nabacus\nabaya\nacademic gown\naccordion\nacoustic guitar\naircraft carrier\nairliner\nairship\naltar\nambulance\namphibian\nanalog clock\napiary\napron\nashcan\nassault rifle\nbackpack\nbakery\nbalance beam\nballoon\nballpoint\nBand Aid\nbanjo\nbannister\nbarbell\nbarber chair\nbarbershop\nbarn\nbarometer\nbarrel\nbarrow\nbaseball\nbasketball\nbassinet\nbassoon\nbathing cap\nbath towel\nbathtub\nbeach wagon\nbeacon\nbeaker\nbearskin\nbeer bottle\nbeer glass\nbell cote\nbib\nbicycle-built-for-two\nbikini\nbinder\nbinoculars\nbirdhouse\nboathouse\nbobsled\nbolo tie\nbonnet\nbookcase\nbookshop\nbottlecap\nbow\nbow tie\nbrass\nbrassiere\nbreakwater\nbreastplate\nbroom\nbucket\nbuckle\nbulletproof vest\nbullet train\nbutcher shop\ncab\ncaldron\ncandle\ncannon\ncanoe\ncan opener\ncardigan\ncar mirror\ncarousel\ncarpenter's kit\ncarton\ncar wheel\ncash machine\ncassette\ncassette player\ncastle\ncatamaran\nCD player\ncello\ncellular telephone\nchain\nchainlink fence\nchain mail\nchain saw\nchest\nchiffonier\nchime\nchina cabinet\nChristmas stocking\nchurch\ncinema\ncleaver\ncliff dwelling\ncloak\nclog\ncocktail shaker\ncoffee mug\ncoffeepot\ncoil\ncombination lock\ncomputer keyboard\nconfectionery\ncontainer ship\nconvertible\ncorkscrew\ncornet\ncowboy boot\ncowboy hat\ncradle\ncrane\ncrash helmet\ncrate\ncrib\nCrock Pot\ncroquet ball\ncrutch\ncuirass\ndam\ndesk\ndesktop computer\ndial telephone\ndiaper\ndigital clock\ndigital watch\ndining table\ndishrag\ndishwasher\ndisk brake\ndock\ndogsled\ndome\ndoormat\ndrilling platform\ndrum\ndrumstick\ndumbbell\nDutch oven\nelectric fan\nelectric guitar\nelectric locomotive\nentertainment center\nenvelope\nespresso maker\nface powder\nfeather boa\nfile\nfireboat\nfire engine\nfire screen\nflagpole\nflute\nfolding chair\nfootball helmet\nforklift\nfountain\nfountain pen\nfour-poster\nfreight car\nFrench horn\nfrying pan\nfur coat\ngarbage truck\ngasmask\ngas pump\ngoblet\ngo-kart\ngolf ball\ngolfcart\ngondola\ngong\ngown\ngrand piano\ngreenhouse\ngrille\ngrocery store\nguillotine\nhair slide\nhair spray\nhalf track\nhammer\nhamper\nhand blower\nhand-held computer\nhandkerchief\nhard disc\nharmonica\nharp\nharvester\nhatchet\nholster\nhome theater\nhoneycomb\nhook\nhoopskirt\nhorizontal bar\nhorse cart\nhourglass\niPod\niron\njack-o'-lantern\njean\njeep\njersey\njigsaw puzzle\njinrikisha\njoystick\nkimono\nknee pad\nknot\nlab coat\nladle\nlampshade\nlaptop\nlawn mower\nlens cap\nletter opener\nlibrary\nlifeboat\nlighter\nlimousine\nliner\nlipstick\nLoafer\nlotion\nloudspeaker\nloupe\nlumbermill\nmagnetic compass\nmailbag\nmailbox\nmaillot\nmaillot\nmanhole cover\nmaraca\nmarimba\nmask\nmatchstick\nmaypole\nmaze\nmeasuring cup\nmedicine chest\nmegalith\nmicrophone\nmicrowave\nmilitary uniform\nmilk can\nminibus\nminiskirt\nminivan\nmissile\nmitten\nmixing bowl\nmobile home\nModel T\nmodem\nmonastery\nmonitor\nmoped\nmortar\nmortarboard\nmosque\nmosquito net\nmotor scooter\nmountain bike\nmountain tent\nmouse\nmousetrap\nmoving van\nmuzzle\nnail\nneck brace\nnecklace\nnipple\nnotebook\nobelisk\noboe\nocarina\nodometer\noil filter\norgan\noscilloscope\noverskirt\noxcart\noxygen mask\npacket\npaddle\npaddlewheel\npadlock\npaintbrush\npajama\npalace\npanpipe\npaper towel\nparachute\nparallel bars\npark bench\nparking meter\npassenger car\npatio\npay-phone\npedestal\npencil box\npencil sharpener\nperfume\nPetri dish\nphotocopier\npick\npickelhaube\npicket fence\npickup\npier\npiggy bank\npill bottle\npillow\nping-pong ball\npinwheel\npirate\npitcher\nplane\nplanetarium\nplastic bag\nplate rack\nplow\nplunger\nPolaroid camera\npole\npolice van\nponcho\npool table\npop bottle\npot\npotter's wheel\npower drill\nprayer rug\nprinter\nprison\nprojectile\nprojector\npuck\npunching bag\npurse\nquill\nquilt\nracer\nracket\nradiator\nradio\nradio telescope\nrain barrel\nrecreational vehicle\nreel\nreflex camera\nrefrigerator\nremote control\nrestaurant\nrevolver\nrifle\nrocking chair\nrotisserie\nrubber eraser\nrugby ball\nrule\nrunning shoe\nsafe\nsafety pin\nsaltshaker\nsandal\nsarong\nsax\nscabbard\nscale\nschool bus\nschooner\nscoreboard\nscreen\nscrew\nscrewdriver\nseat belt\nsewing machine\nshield\nshoe shop\nshoji\nshopping basket\nshopping cart\nshovel\nshower cap\nshower curtain\nski\nski mask\nsleeping bag\nslide rule\nsliding door\nslot\nsnorkel\nsnowmobile\nsnowplow\nsoap dispenser\nsoccer ball\nsock\nsolar dish\nsombrero\nsoup bowl\nspace bar\nspace heater\nspace shuttle\nspatula\nspeedboat\nspider web\nspindle\nsports car\nspotlight\nstage\nsteam locomotive\nsteel arch bridge\nsteel drum\nstethoscope\nstole\nstone wall\nstopwatch\nstove\nstrainer\nstreetcar\nstretcher\nstudio couch\nstupa\nsubmarine\nsuit\nsundial\nsunglass\nsunglasses\nsunscreen\nsuspension bridge\nswab\nsweatshirt\nswimming trunks\nswing\nswitch\nsyringe\ntable lamp\ntank\ntape player\nteapot\nteddy\ntelevision\ntennis ball\nthatch\ntheater curtain\nthimble\nthresher\nthrone\ntile roof\ntoaster\ntobacco shop\ntoilet seat\ntorch\ntotem pole\ntow truck\ntoyshop\ntractor\ntrailer truck\ntray\ntrench coat\ntricycle\ntrimaran\ntripod\ntriumphal arch\ntrolleybus\ntrombone\ntub\nturnstile\ntypewriter keyboard\numbrella\nunicycle\nupright\nvacuum\nvase\nvault\nvelvet\nvending machine\nvestment\nviaduct\nviolin\nvolleyball\nwaffle iron\nwall clock\nwallet\nwardrobe\nwarplane\nwashbasin\nwasher\nwater bottle\nwater jug\nwater tower\nwhiskey jug\nwhistle\nwig\nwindow screen\nwindow shade\nWindsor tie\nwine bottle\nwing\nwok\nwooden spoon\nwool\nworm fence\nwreck\nyawl\nyurt\nweb site\ncomic book\ncrossword puzzle\nstreet sign\ntraffic light\nbook jacket\nmenu\nplate\nguacamole\nconsomme\nhot pot\ntrifle\nice cream\nice lolly\nFrench loaf\nbagel\npretzel\ncheeseburger\nhotdog\nmashed potato\nhead cabbage\nbroccoli\ncauliflower\nzucchini\nspaghetti squash\nacorn squash\nbutternut squash\ncucumber\nartichoke\nbell pepper\ncardoon\nmushroom\nGranny Smith\nstrawberry\norange\nlemon\nfig\npineapple\nbanana\njackfruit\ncustard apple\npomegranate\nhay\ncarbonara\nchocolate sauce\ndough\nmeat loaf\npizza\npotpie\nburrito\nred wine\nespresso\ncup\neggnog\nalp\nbubble\ncliff\ncoral reef\ngeyser\nlakeside\npromontory\nsandbar\nseashore\nvalley\nvolcano\nballplayer\ngroom\nscuba diver\nrapeseed\ndaisy\nyellow lady's slipper\ncorn\nacorn\nhip\nbuckeye\ncoral fungus\nagaric\ngyromitra\nstinkhorn\nearthstar\nhen-of-the-woods\nbolete\near\ntoilet tissue\n"