search

SHI-Labs / Compact-Transformers

Escaping the Big Data Paradigm with Compact Transformers, 2021 (Train your Vision Transformers in 30 mins on CIFAR-10 with a single GPU!)
https://arxiv.org/abs/2104.05704
Apache License 2.0
495 stars 77 forks source link

x += self.positional_emb mismatch #74

Closed githubjqh closed 1 year ago

githubjqh commented 1 year ago

When I run the model vit_2_4_32, dataset is flowers102, this error will occur: File "C:\pyprj\Compact-Transformers\src\vit.py", line 62, in forward return self.classifier(x) File "C:\Users\asp\.conda\envs\detpy\lib\site-packages\torch\nn\modules\module.py", line 1102, in _call_impl return forward_call(*input, **kwargs) File "C:\pyprj\Compact-Transformers\src\utils\transformers.py", line 199, in forward x += self.positional_emb RuntimeError: The size of tensor a (3137) must match the size of tensor b (65) at non-singleton dimension 1 size of x is (2,3137,65), size of positional_emb is (1,65,65) When I turn to other models, this error always occur!
I use python 3.8, and pytorch=1.10.1. Thank you!

githubjqh commented 1 year ago

for dataset flowers102, models of image size 224 such as cct_7_7x2_224 should be used.

stevenwalton commented 1 year ago

Did you set the img_size parameter for the models? src/cct.py has all the pre-defined models and shows you how you can create your own. Our registered models are of the format cct num_layers kernel_size x num_convs img_size. A full list can be viewed here