Do you consider combine the tokens-to-token (kernel>1 vs vit where kernel=1) with token labeling?

[zhaoguangxiang]

Have you tried this setting?

[yuanli2333]

We have tried it, and token labeling also works for our t2t-vit, we will release such models in this repo and our T2T-ViT repo in the future.

[zihangJiang]

@zhaoguangxiang You can check the models here: https://github.com/yitu-opensource/T2T-ViT#2-t2t-vit-models