Allow arbitrary image sizes and upstream changes from Swin-Transformer-Object-Detection

[vadimkantorov]

It is useful in object detection context to allow arbitrary sizes by doing dynamic mask computation (probably possible only with relative position encoding).

These kinds of edits were done in https://github.com/SwinTransformer/Swin-Transformer-Object-Detection and in https://github.com/megvii-research/SOLQ/. It would be nice if you upstreamed these changes. This will simplify trying out ESviT checkpoints as pretraining for object detection.

Also, fyi I created a similar issue in SimMIM: microsoft/SimMIM#13. Overall, having some stable version of swin_transformer.py somewhere (maybe even in main SwinTransformer/Swin-Transformer repo?) supporting dynamic masking would help a lot :)

Thanks!

[sym0926]

Hi,do you have ckpt and train logs , can you share with me ? I got an error ,when I download them.