search

lukemelas / PyTorch-Pretrained-ViT

Vision Transformer (ViT) in PyTorch
770 stars 124 forks source link

TypeError: 'tuple' object is not callable #1

Closed MC-E closed 3 years ago

MC-E commented 3 years ago

Thanks for your respectable work! But there seem to be some mistakes when I run the training code:
File "main.py", line 427, in main() File "main.py", line 111, in main main_worker(args.gpu, ngpus_per_node, args) File "main.py", line 134, in main_worker model = ViT(args.arch, pretrained=args.pretrained,image_size=args.image_size) File "/userhome/Transformer/PyTorch-Pretrained-ViT/pytorch_pretrained_vit/model.py", line 135, in init resize_positional_embedding=(image_size != pretrained_image_size), File "/userhome/Transformer/PyTorch-Pretrained-ViT/pytorch_pretrained_vit/utils.py", line 62, in load_pretrained_weights has_class_token=hasattr(model, 'class_token')) File "/userhome/Transformer/PyTorch-Pretrained-ViT/pytorch_pretrained_vit/utils.py", line 101, in resize_positionalembedding posemb_grid = zoom(posemb_grid, zoom, order=1) TypeError: 'tuple' object is not callable

This mistake appears when I want to finetune the pretained model (B_16, 224*224) on the ImageNet-1K (384*384). Thank you again for this work. Looking forward to your reply!

lukemelas commented 3 years ago

Thanks for the issue! Indeed, there was a bug when reloading and resizing the embeddings. It's fixed and a new version has been pushed to pip. Upgrade and let me know if you have any more issues!

I'm going to close this -- reopen if you continue to have issues.