Tensorflow 2 Caffe error

[gasgallo]

I'm using the provided docker image to convert a TF frozen model (.pb) into a Caffe model.

The mmconvert command results in:

tensorflow.python.framework.errors_impl.DataLossError: Unable to open table file /model_frozen.pb: Data loss: not an sstable (bad magic number): perhaps your file is in a different file format and you need to use a different restore operator?

[linmajia]

@gasgallo , thank you very much for your feedback:

1. "a TF frozen model (.pb)" is it a serialized frozen graph?
2. What is your command line?

[gasgallo]

@linmajia

1. yes
2. mmconvert -sf tensorflow -iw /model_frozen.pb --inNodeName data --inputShape 1,112,112,3 --dstNodeName fc1/add_1 -df caffe -om tf_model

Thanks for the help!!

[linmajia]

@gasgallo, please "-in" instead of "-iw" for the model file.

[gasgallo]

@linmajia

I get a different error:
Info: Trying to parse file [/model_frozen.pb] with binary format but failed with error [Error parsing message].
UnicodeDecodeError: 'utf-8' codec can't decode byte 0xac in position 39: invalid start byte

btw, my model is a custom version of ResNet50

[NikolaMr]

I have the same problem over here when converting from tensorflow to caffe.

[gasgallo]

No clue so far?

[NikolaMr]

@gasgallo Absolutely not. However, if you can somehow export to Pytorch, this one works https://github.com/xxradon/PytorchToCaffe

[yongchaoding]

Same problem, can anyone help?

[zli2014]

i meet the same problem , but i found the pb graph is diiffrent ckpt graph;