prototxt file ignored?

[leongatys]

Hi, I am a little confused, it appears to me that the prototxt argument to loadcaffe.load(prototxt, model, backend) is ignored and the converted network in torch is completely determined by the .caffemodel file. Is that correct? If yes, one might want to add a quick note on that to the documentation and/or remove the prototxt argument from the function. Right now it might be a little confusing since in caffe things like the padding or pooling operations are specified in the .prototxt file and one can load the same .caffemodel file with different .prototxt files to get a network with the same weights but different padding/pooling etc. operations.

Best Leon

[szagoruyko]

yes you're right. I have a fix for that, will push soon.

[axeldavy]

The problem is still there.

Yours.