TensorFlow Lite学习笔记

TensorFlow Lite学习笔记（参考Module: tf.lite  |  TensorFlow Lite (google.cn)）

 

总共分为六块：

Interpreter: 推理

model_content：

num_threads：tflite的卷积推理支持多线程，这里可以指定线程数

tf.lite.Interpreter(
    model_path=None, model_content=None, experimental_delegates=None,
    num_threads=None
)

 

OpsSet: 参数类型

• EXPERIMENTAL_TFLITE_BUILTINS_ACTIVATIONS_INT16_WEIGHTS_INT8
• SELECT_TF_OPS
• TFLITE_BUILTINS
• TFLITE_BUILTINS_INT8

TFLITE_BUILTINS，SELECT_TF_OPS， EXPERIMENTAL_TFLITE_BUILTINS_ACTIVATIONS_INT16_WEIGHTS_INT8 都是在实验阶段，不建议使用

TFLITE_BUILTINS_INT8要用的化需要自定义数据集

 

Optimize: 生成TFlite图的时候优化方法

• DEFAULT：
• OPTIMIZE_FOR_LATENCY：优化算力
• OPTIMIZE_FOR_SIZE：优化模型大小

 

RepresentativeDataset: 重定义数据集

 

TFLiteConverter: tflite转换器

支持

# Converting a SavedModel to a TensorFlow Lite model.
converter = tf.lite.TFLiteConverter.from_saved_model(saved_model_dir)
tflite_model = converter.convert()

#keras model xxx.h5
# Converting a tf.Keras model to a TensorFlow Lite model.
converter = tf.lite.TFLiteConverter.from_keras_model(model)
tflite_model = converter.convert()

# Converting ConcreteFunctions to a TensorFlow Lite model.
converter = tf.lite.TFLiteConverter.from_concrete_functions([func])
tflite_model = converter.convert()

TargetSpec: 对于特定设备的特殊要求指定

supported_ops：比如二进制，正在实验阶段

supported_types：设备上支持的类型

tf.lite.TargetSpec(
    supported_ops=None, supported_types=None
)