Tensorflow：如何将冻结模型转换为已保存模型

Question

I'm trying to convert this Frozen Model ( extension .pb) from here:

https://www.dropbox.com/s/bb7n78hq5owyjr0/tf_model_pb.zip?dl=0

To a Saved Model .

Is it possible? Can you share the exact code to convert this pb file to saved model?

Answer 1

Answer first, it's possible but quite messy to do so, because all variables required for SavedModel is converted to constant or other type during freezing ckpt to pb model. In tf, revert 'Constant' to 'Variable' is a little bit complicated, like:

# node is a constant
node = tf.Constant([1, 2])
output_node = tf.NodeDef()
output_node.op = "Variable"
output_node.name = node.name
dtype = node.attr["dtype"].type
data = node.attr["value"].tensor
output_node.attr["dtype"].type = dtype
output_node.attr["value"].CopyFrom(tf.AttrValue(tensor=data))

See this answer for detailed info to convert constant node to some other types.

So, freezing SavedModel from ckpt file is the easiest and most direct way.