我需要将模型转换为处理B64图像而不是JSON，我正在使用tf.keras.estimator.model_to_estimator
tf-nightly version:
Version: 1.15.0-dev20190714
Eager mode: True
Hub version: 0.5.0
GPU is available
基于this教程
我使用tf.Hub图层创建了Keras模型：
https://tfhub.dev/google/tf2-preview/mobilenet_v2/feature_vector/4
转换为估算器：
estimator = tf.keras.estimator.model_to_estimator(
keras_model_path=h5_model_path,
custom_objects={'KerasLayer':hub.KerasLayer},
model_dir=tf_model_path)
我得到：
ValueError: Unknown layer: KerasLayer
模型摘要：
model.summary()
Model: "sequential"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
keras_layer (KerasLayer) (None, 1280) 2257984
_________________________________________________________________
dense (Dense) (None, 5) 6405
=================================================================
Total params: 2,264,389
Trainable params: 6,405
Non-trainable params: 2,257,984
如果我通过Keras模型对象：
estimator = tf.keras.estimator.model_to_estimator(
keras_model=model,
custom_objects={'KerasLayer':hub.KerasLayer},
model_dir=tf_model_path)
我得到：
FailedPreconditionError: 2 root error(s) found.
(0) Failed precondition: Error while reading resource variable save_counter from Container: localhost. This could mean that the variable was uninitialized. Not found: Resource localhost/save_counter/N10tensorflow3VarE does not exist.
[[{{node save_counter/Read/ReadVariableOp}}]]
(1) Failed precondition: Error while reading resource variable save_counter from Container: localhost. This could mean that the variable was uninitialized. Not found: Resource localhost/save_counter/N10tensorflow3VarE does not exist.
[[{{node save_counter/Read/ReadVariableOp}}]]
[[MobilenetV2/expanded_conv_15/project/weights/Read/ReadVariableOp/_2363]]
0 successful operations.
0 derived errors ignored.