sanghoon/pytorch_imagenet

The provided model for imagenet is broken

Closed this issue · 3 comments

wangpichao commented

When I downloaded the model, it could not unrar. Could you kindly please upload a new one? Many thanks.

xiequnyi commented

The pretrained model you apply is broken. Could you please check it ? THX

dhkim0225 commented

I had a same problem, and the solution was simple

asdf = torch.load('/downloaded/file/blahblah.pth.tgz')
print(asdf.keys())
dhkim0225 commented

Here are my codes working with torch 0.4.1, cpu-version
I hope this helps.

from collections import OrderedDict

def copyStateDict(state_dict):
    if "module." in list(state_dict.keys())[0]:  # This is when weight file is saved with multi gpu
        new_state_dict = OrderedDict()
        for k, v in state_dict.items():
            name = k.replace('module.', '')
            new_state_dict[name] = v
        return new_state_dict
    else:
        return state_dict

model = PVANet()
asdf = torch.load('weights/pvanet_600epochs.checkpoint.pth.tar', map_location='cpu')
model.load_state_dict(copyStateDict(asdf['state_dict']))
  • 👍1