My team has spent several months training a Caffe model but now we're looking at converting to Tensorflow and we don't want to spend several months retraining.
How can I convert our caffemodel weight files to TensorFlow weight files?
I've tried caffe-tensorflow but it won't work for Caffe models that have layers with more than one top.
I am not familiar with tensorflow, but you can use this python code to "dump" the trained caffe weights to python-readable numpy arrays. I would expect TF to have a functionality to init its weights from numpy arrays - this method might not be automatic or trivial, but you only need to do it once.
What is the error you get from caffe-tensorflow? If this is just a validation check, you can disable it. You can also try to enforce the tool to load the weights through load_using_pb function. (transformers.py / DataInjector / load function)
