Tensorflow Lite 笔记(二)
tensorflow lite model maker
metadata which provides a standard for model descriptions
The default model is EfficientNet-Lite0.
Currently, we support several models such as EfficientNet-Lite* models, MobileNetV2, ResNet50 as pre-trained models for image classification
此处的模型为 EfficientNet-Lite0
model.tflite 大小为 4.0MB
model_fp16.tflite 大小为 6.8MB
此处的模型为 MobileNetV2
model.tflite 大小为 2.8MB
model_fp16.tflite 大小为 4.6MB
此处的模型为 InceptionV3
model.tflite 大小为 22.4MB
model_fp16.tflite 大小为 43.8MB
Post-training quantization is a conversion technique that can reduce model size and inference latency, while also improving CPU and hardware accelerator inference speed, with a little degradation in model accuracy. Thus, it’s widely used to optimize the model
Image classification with TensorFlow Lite Model Maker 总结:
export formats 的几种形式
The allowed export formats can be one or a list of the following:
- ExportFormat.TFLITE
- ExportFormat.LABEL
- ExportFormat.SAVED_MODEL
By default, it just exports TensorFlow Lite model with metadata. You can also selectively export different files. For instance, exporting only the label file as follows:
model.export(export_dir='.', export_format=ExportFormat.LABEL)
Customize Post-training quantization on the TensorFLow Lite model
config = QuantizationConfig.for_float16()
model.export(export_dir='.', tflite_filename='model_fp16.tflite', quantization_config=config)
Change to the model that’s supported in this library
model = image_classifier.create(train_data, model_spec=model_spec.get('mobilenet_v2'), validation_data=validation_data)
Change to the model in TensorFlow Hub
inception_v3_spec = image_classifier.ModelSpec(
uri='https://tfhub.dev/google/imagenet/inception_v3/feature_vector/1')
inception_v3_spec.input_image_shape = [299, 299]
Change your own custom model