ICCV '21 | Swin Transformer: Hierarchical Vision Transformer Using Shifted Windows

[mental2008]

Presented in ICCV '21. [ Paper | Supplement | arXiv | Code ] Awarded Best Paper (Marr Prize)!

Authors: Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, Baining Guo Microsoft Research Asia, University of Science and Technology of China, Xian Jiaotong University, Tsinghua University

[mental2008]

As mentioned in the paper, Swin Transformer is a new vision transformer (ViT), which serves as a general-purpose backbone for computer vision.

"Interesting" points:

1. Excellent results in many tasks. (Is really good?)
2. The authors argue that the hierarchical design and the shifted window approach can be applied for all-MLP architectures efficiently.
3. The code is open-sourced in GitHub.

The model architecture is as follows:

Not read the details of the paper.