Closed 98mxr closed 1 year ago
Actually, I haven't tried many experiments. I am still trying things out. The reason for using V1 now is simply because V2 is slow and many other swin transformer based models use V1.
I understand, the paper mentions that v2 may adapt better to different size inputs, looking forward to the good news from your side.
Why choose swin_transformer_v1 instead of v2?