MixFormer: Mixing Features Across Windows and Dimensions

Qiang Chen, Qiman Wu, Jian Wang, Qinghao Hu, Tao Hu, Errui Ding, Jian Cheng, Jingdong Wang; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 5249-5259


While local-window self-attention performs notably in vision tasks, it suffers from limited receptive field and weak modeling capability issues. This is mainly because it performs self-attention within non-overlapped windows and shares weights on the channel dimension. We propose MixFormer to find a solution. First, we combine local-window self-attention with depth-wise convolution in a parallel design, modeling cross-window connections to enlarge the receptive fields. Second, we propose bi-directional interactions across branches to provide complementary clues in the channel and spatial dimensions. These two designs are integrated to achieve efficient feature mixing among windows and dimensions. Our MixFormer provides competitive results on image classification with EfficientNet and shows better results than RegNet and Swin Transformer. Performance in downstream tasks outperforms its alternatives by significant margins with less computational costs in 5 dense prediction tasks on MS COCO, ADE20k, and LVIS. Code is available at https://github.com/PaddlePaddle/PaddleClas.

Related Material

[pdf] [supp] [arXiv]
@InProceedings{Chen_2022_CVPR, author = {Chen, Qiang and Wu, Qiman and Wang, Jian and Hu, Qinghao and Hu, Tao and Ding, Errui and Cheng, Jian and Wang, Jingdong}, title = {MixFormer: Mixing Features Across Windows and Dimensions}, booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)}, month = {June}, year = {2022}, pages = {5249-5259} }