转换 Keras H5 模型 tflite 的问题答案

【问题标题】：Issue in converting Keras H5 model tflite转换 Keras H5 模型 tflite 的问题
【发布时间】：2021-12-09 17:54:23
【问题描述】：

converter = tf.lite.TFLiteConverter.from_keras_model( './signet.h5')
tflite_model = converter.convert()
with open('model.tflite', 'wb') as f:
  f.write(tflite_model)

AttributeError                            Traceback (most recent call last)
~\AppData\Local\Temp/ipykernel_2464/3288577076.py in <module>
      1 converter = tf.lite.TFLiteConverter.from_keras_model( './signet.h5')
----> 2 tflite_model = converter.convert()
      3 # Save the model.
      4 with open('model.tflite', 'wb') as f:
      5   f.write(tflite_model)

~\anaconda3\envs\tensorflow\lib\site-packages\tensorflow\lite\python\lite.py in convert(self)
    795     # to None.
    796     # Once we have better support for dynamic shapes, we can remove this.
--> 797     if not isinstance(self._keras_model.call, _def_function.Function):
    798       # Pass `keep_original_batch_size=True` will ensure that we get an input
    799       # signature including the batch dimension specified by the user.

AttributeError: 'str' object has no attribute 'call'

我正在尝试将 h5 keras 模型转换为 tflite 以用于反应本机应用程序，但显示此错误。我在论坛上尝试了一些解决方案，其中一个在 github 上，但没有奏效。有没有办法将我的 h5 模型转换为 tflite。

【问题讨论】：

标签： python tensorflow keras deep-learning tensorflow-lite

【解决方案1】：

您应该将实际的 Keras 模型对象传递给转换器的 `from_keras_model 方法 :-)。见例子：

import tensorflow as tf

# Create a model using high-level tf.keras.* APIs
model = tf.keras.models.Sequential([
    tf.keras.layers.Dense(units=1, input_shape=[1]),
    tf.keras.layers.Dense(units=16, activation='relu'),
    tf.keras.layers.Dense(units=1)
])
model.compile(optimizer='sgd', loss='mean_squared_error') # compile the model
model.fit(x=[-1, 0, 1], y=[-3, -1, 1], epochs=5) # train the model
# (to generate a SavedModel) tf.saved_model.save(model, "saved_model_keras_dir")

# Convert the model.
converter = tf.lite.TFLiteConverter.from_keras_model(model)
tflite_model = converter.convert()

# Save the model.
with open('model.tflite', 'wb') as f:
  f.write(tflite_model)

所以你首先需要从 H5 文件中加载。

【讨论】：