run tensorflow error


I want use tx2 to do face recognition with facenet.
Copy the trianed model and classifier from the pc to the tx2.
And install tensorflow(1.5),and others stuff
Run the contributed/
Then report errors:
Model directory: /home/nvidia/facenet/facenet/contributed/…/src/models/20170512-110547
Metagraph file: model-20170512-110547.meta
Checkpoint file: model-20170512-110547.ckpt-250000

WARNING:tensorflow:The saved meta_graph is possibly from an older release:
‘model_variables’ collection should be of type ‘byte_list’, but instead is of type ‘node_list’.
Traceback (most recent call last):
File “contributed/”, line 107, in
File “contributed/”, line 63, in main
face_recognition = face.Recognition()
File “/home/nvidia/facenet/facenet/contributed/”, line 65, in init
self.identifier = Identifier()
File “/home/nvidia/facenet/facenet/contributed/”, line 91, in init
self.model, self.class_names = pickle.load(infile)
File “/usr/lib/python2.7/”, line 1384, in load
return Unpickler(file).load()
File “/usr/lib/python2.7/”, line 864, in load
File “/usr/lib/python2.7/”, line 892, in load_proto
raise ValueError, “unsupported pickle protocol: %d” % proto
ValueError: unsupported pickle protocol: 3

I tested it on the PC,and it’s ok. the tensorflow version on pc is 1.6, and the tx2 tensorflow version is 1.5.


Based on your log, this error is caused from pickle rather than TensorFlow.

Please remember to write pickle data with the lower number.

pickle.dump(object, file, protocol=2)


Hi AstaLLL:

Thanks for your reply very much.

I trained new model and classifier with python2.7 environment, and then it’s ok.

But when i run face recognition, the speed of recognition is too slow. and the fan didn’t work.

Dose the GPU don’t work? And how to check the status of GPU?

Note: I have not use tensorrt now, just do face recogonition on tensorflow.but my installed tensorflow is gpu version.



Guess that you didn’t maximize the CPU/GPU clock.
Please remember to maximize the hardware frequency to have the best performance.

sudo ./


I tried what you said, but it didn’t work.

Dose the GPU don’t work if i do not use tenserRT.


Depends on the op you used in TensorFlow.

Some op doesn’t have GPU implementation or has poor performance on GPU.
It’s recommended to check it with TensorFlower to get further information.