如何在不使用Bazel的情况下冻结Tensorflow 1.5中的图形？

Question

in tools directory in tensorflow, we have freeze graph python file, that we can use directly, so why do we need bazel, also I tried using that script, and got the following error :

    226   elif input_graph:
--> 227     input_graph_def = _parse_input_graph_proto(input_graph, input_binary)
    228   input_meta_graph_def = None
    229   if input_meta_graph:

/usr/local/lib/python3.5/dist-packages/tensorflow/python/tools/freeze_graph.py in _parse_input_graph_proto(input_graph, input_binary)
    167   with gfile.FastGFile(input_graph, mode) as f:
    168     if input_binary:
--> 169       input_graph_def.ParseFromString(f.read())
    170     else:
    171       text_format.Merge(f.read(), input_graph_def)

DecodeError: Error parsing message

Answer 1

You don't need bazel, you can simply execute python freeze_graph.py with the appropriate arguments.

About your error, it looks like you're passing a .pbtxt file without setting --input_binary=False . Sinceit defaults to True , the script expects you to pass a .pb file. When it tries to decode a .pbtxt as binary, errors are raised.