Use pre-trained ResNet with frozen layers

Question

Use pre-trained ResNet with frozen layers

Tato14 opened this issue 5 years ago · 2 comments

Hi!
As transfer learning seem to be quite a big issue on these days, I was wondering if you tried to train ResNet 18 using ImageNet pretrained model but freezing all the layers except the last one.

Also, I found the documentenation in PyTorch here but I am not sure where should I implement this lines of code. Should be in the create_model function?
Thanks!

Answer 1 · 2019-04-09T10:13:37.000Z

Also, it would be great if you could share the weights of your ResNet 18 (and 34?) to give them a try as a pretrained model.

Answer 2 · 2019-04-10T16:22:31.000Z

You can set pretrain=True in code/config.py to instantiate the ImageNet pretrained model. Currently, the weights are not available, but they will be posted in this repo once they are.