yitu-opensource / T2T-ViT

ICCV2021, Tokens-to-Token ViT: Training Vision Transformers from Scratch on ImageNet
Other
1.15k stars 176 forks source link

Input size is not a square, what should I do in this line? #76

Open JerryKingQAQ opened 1 year ago

JerryKingQAQ commented 1 year ago

Thanks to your excellent work! There has a tensor which size is (3, 56, 112), What should I do in this line to modify ? self.num_patches = (img_size // (4 * 2 * 2)) * (img_size // (4 * 2 * 2)) # there are 3 sfot split, stride are 4,2,2 seperately