Failed to load pb file exported from onnx

[chudegao]

I converted tensorflow pb file to onnx, load onnx to tf and export the graph to a new pb file. But when load the pb file in tensorflow, it failed.

with tf.Session(graph=tf.Graph()) as sess:
... tf.saved_model.load(sess,export_dir='/tmp/pbdir/',tags=['serve'])
...
Traceback (most recent call last):
File "", line 2, in
File "/opt/anaconda3/envs/dlipy3/lib/python3.6/site-packages/tensorflow/python/util/deprecation.py", line 324, in new_func
return func(*args, **kwargs)
File "/opt/anaconda3/envs/dlipy3/lib/python3.6/site-packages/tensorflow/python/saved_model/loader_impl.py", line 269, in load
return loader.load(sess, tags, import_scope, **saver_kwargs)
File "/opt/anaconda3/envs/dlipy3/lib/python3.6/site-packages/tensorflow/python/saved_model/loader_impl.py", line 422, in load
**saver_kwargs)
File "/opt/anaconda3/envs/dlipy3/lib/python3.6/site-packages/tensorflow/python/saved_model/loader_impl.py", line 349, in load_graph
meta_graph_def = self.get_meta_graph_def_from_tags(tags)
File "/opt/anaconda3/envs/dlipy3/lib/python3.6/site-packages/tensorflow/python/saved_model/loader_impl.py", line 327, in get_meta_graph_def_from_tags
"\navailable_tags: " + str(available_tags))
RuntimeError: MetaGraphDef associated with tags 'serve' could not be found in SavedModel. To inspect available tag-sets in the SavedModel, please use the SavedModel CLI: saved_model_cli
available_tags: []

There's no any tags in the new pb file.
Use saved_model_cli to check tag-sets in saved model, there's no any tag.
(dlipy3) [root@haswell01 tmp]# saved_model_cli show --dir /tmp/pbdir/ --all
2019-09-02 04:58:42.332050: I tensorflow/stream_executor/platform/default/dso_loader.cc:42] Successfully opened dynamic library libcudart.so.10.1
(dlipy3) [root@haswell01 tmp]#

The original pb file save by tensorflow model have 'serve' tag.
(dlipy3) [root@haswell01 tmp]# saved_model_cli show --dir /dlfs/chudg/onnx/tutorials/tutorials/assets/output/saved_model --all
2019-09-02 04:46:45.606824: I tensorflow/stream_executor/platform/default/dso_loader.cc:42] Successfully opened dynamic library libcudart.so.10.1
MetaGraphDef with tag-set: 'serve' contains the following SignatureDefs:
signature_def['serving_default']:
The given SavedModel SignatureDef contains the following input(s):
inputs['input/input:0'] tensor_info:
dtype: DT_FLOAT
shape: (-1, 784)
name: input/input:0
The given SavedModel SignatureDef contains the following output(s):
outputs['output/result:0'] tensor_info:
dtype: DT_FLOAT
shape: (-1, 10)
name: output/result:0
Method name is: tensorflow/serving/predict
(dlipy3) [root@haswell01 tmp]#

The purpose I convert tf pb-->onnx--> tf pb is do some pruning of onnx and use tensorflow to re-train the model. Anyone can help?

[ngoanpv]

I have the same problem, did you fix it?

[Wheest]

Onnx-Tensorflow doesn't seem to be setting tags on the pb model. I haven't been able to find a flag in the API that sets this. This is making it tricky to view the model in Tensorboard.

[natedana]

I'm also running into this issue. Has anyone found a workaround?

[sergei-mironov]

I'm also running into this issue. Has anyone found a workaround?

As a workaround I'm trying to load the PB using compat.v1 interface of TF2x, like this:

import tensorflow as tf
def load_pb_v1(path_to_pb):
  with tf.compat.v1.gfile.GFile(path_to_pb, "rb") as f:
    graph_def = tf.compat.v1.GraphDef()
    graph_def.ParseFromString(f.read())
  with tf.Graph().as_default() as graph:
    tf.import_graph_def(graph_def, name='')
    return graph