search

Talented-Q / POSTER_V2

MIT License
112 stars 20 forks source link

Problem about the evaluation with the checkpoint. #28

Open LuoXubo opened 4 months ago

LuoXubo commented 4 months ago

Hi!

Tanks for your great work. But there is a bug when I run this in terminal: python main.py --data [MY_data_path] --evaluate checkpoint/affectnet-7-model_best.pth

The erros information is: RuntimeError: Error(s) in loading state_dict for pyramid_trans_exper2: Missing key(s) in state_dict

It seems like the .pth file does not corresponds to the model used in main.py. How can I fix this bug?

Traceymaimai commented 2 months ago

i solve this problem, If you're running this on a single GPU or CPU, you might encounter this. It seems the problem is due to a naming discrepancy between the model's keys (e.g., "face_landback.conv1.conv.weight") and the checkpoint's keys (which have "module." prefixed). Here's a quick snippet to help you out:

checkpoint = torch.load('path_to_your_checkpoint.pth') state_dict = checkpoint['state_dict'] for key in list(state_dict.keys()): if 'module.' in key: state_dict[key.replace('module.', '')] = state_dict.pop(key) model.load_state_dict(state_dict)

you can run this model with pre-train weights on mutil gpus or delete the 'model.' in the static_dict.