Han Hu

Probably you can plot the attention maps within the neighboring window for each query point

> Thanks for your great repo. I would like to ask when the swin-v2 pre-trained model will be released? The swin-v2 models have been released

> the swin transformer can't deal with segmentation works! the output have one color can you help me resolve this trouble? Please try this codebase: https://github.com/SwinTransformer/Swin-Transformer-Semantic-Segmentation

> 我只使用了swin transformer的模型架构，win size是4，输入大小是32，其他都没变，也没有使用任何数据增强，resnet可以达到80%的acc，但是swin trans只有52%，请问，这样的结果，是什么引起的？ You may need to first design the full architecture, and change also many other hyper-parameters accordingly

> Swin-MLP code is right for Swin-Mixer

We use bi-cubic interpolation on relative_position_bias_table to deal with larger window size. Will provide related code soon.

Swin V2 has a better approach dealing with different window sizes.

Thanks @xizi

> Dear authors, > > Sorry for repeating [this issue](https://github.com/microsoft/Swin-Transformer/issues/55). It will be great if you could release an ImageNet-22K pretrained Swin-T model (if possible). I believe this will greatly...

> I train the `swin_tiny_patch4_window7_224` with one million classes and 100 million images with softmax loss and adamw, the batch size is 600 and train for 400,000 iterations but the...