I’m using resnet to do feature extraction. I’m assuming the current resnet provided in model zoo is converted from fb.resnet.torch. Are you planning to convert the caffe model into pytorch version? (From my own experience, it seems the caffe one is better.)
1 Like
the current model is trained from scratch and matches accuracy to fb.resnet.torch model
I realize that. The features just don’t work as well as the features extracted from the caffe model.
Anyway, I wrote some code which could convert the caffe resnet to pytorch model.
3 Likes