Frames per patch

[angadkalra]

Hi,

Thanks a lot for your repo! I have 224x224x320 3D data that I want to finetune on, and I'm wondering if your inflation code allows for a smaller depth patch_size? I've only been able to successfully load and inflate a pre-trained ViT using patch_size (16, 16, 320), because anything other than 320 results in tensor shape errors. Could you please advise me on this? I would like to have a patch_size of (16,16,16) for my input data.

Thanks!