How to train 224*224 images

[yongZheng1723]

If I want to run 224*224 images, what parameters should I change in config.py when I choose vitb16 or SWIN_W7_224?

[chou141253]

About vitb16, you should change

FGVC-PIM/config.py

Line 27 in 3ecf29c

parser.add_argument("--use_layers",

and

FGVC-PIM/config.py

Line 30 in 3ecf29c

default=[True, True, True, True], type=list)

, because of the difference of the number of blocks.
Swin-T should be fine.
if you have any questions, please let me know, thanks.

[yongZheng1723]

Thanks for your answer, but when I run swin-t_win7_224，I changed line 234 of Swinvit12.py to self.extractor = timm.create_model('swin_large_patch4_window7_224_in22k', pretrained=True)，

The following problems were encountered:RuntimeError: Given groups=1, weight of size [576, 144, 1], expected input[4, 49, 1536] to have 144 channels, but got 49 channels instead
What do I need to do next？

[chou141253]

Hi, do you replace folder timm/ to our timm/ folder (for ViT or Swin-T)?