search

mariodoebler / test-time-adaptation

A repository and benchmark for online test-time adaptation.
MIT License
188 stars 16 forks source link

Unable to run ViT experiments #13

Closed goirik-chakrabarty closed 6 months ago

goirik-chakrabarty commented 6 months ago

Hi,

Firstly, thanks a lot for the excellent repo! I was trying to use the transformer architectures to reproduce the numbers however I am not able to understand which ARCH_NAME to use. The split_head function also gives an error.

This may be due to a mismatch in the Timm version, but I am unsure.

RobMarsden commented 6 months ago

Hi,

all our experiments use the models provided by torchvision. If you want to reproduce some of the experiments, you should have a look at the run.sh. Alternatively, you can manually set MODEL.ARCH to vit_b_16 or swin_b. If you encounter any errors after updating the environment, please provide the details.

manogna-s commented 6 months ago

Hi,

I had the same issue. I was able to use vit_b_16 architecture. However, I get this error on using swin_b :

raise AttributeError(f"'{type(self).__name__}' object has no attribute '{name}'") AttributeError: 'SwinTransformer' object has no attribute 'patch_embed'

Any suggestions to resolve this?

RobMarsden commented 6 months ago

Hi, can you provide the configuration you are using?

goirik-chakrabarty commented 6 months ago

Thank you! @RobMarsden it worked for me.