can't train on resnet 50
Closed this issue · 4 comments
deBrian07 commented
IFICL commented
Do you use my provided checkpoints? They are only for ResNet18. If you are retraining the models, pretrain from ImageNet pretrain weights.
deBrian07 commented
Yes, I'm trying to use slfm-pretext.py to retrain the models with resnet50, and the errors occurred.
IFICL commented
Have you had the opportunity to troubleshoot by yourself in order to identify the specific line of code causing this issue? I suspect it's the imagenet pretraining loading issue. Please modify it to meet the torchvision package.
IFICL commented
I have tested it on my end, I don't have this issue. I assume it's likely because of the pytorch and torchvision environment update. Please try to fix this by yourself or use the provided environment yml file.