---

toc: true title: Swin Transformer

tags: [‘temp’]

---

Swin Transformer

• Swin Transformer: Hierarchical Vision Transformer Using Shifted Windows
  • Vision Transformer
  • general-purpose backbone for computer vision
  • hierarchical feature representation
  • linear computational complexity with respect to input image size
  • shifted window based Self Attention
  • address the challenges in adapting Transformer from language to vision
  • limiting self-Attention computation to non-overlapping local windows while also allowing for cross-window connection
  • flexibility to model at various scales
  • linear computational complexity with respect to image size
  • ImageNet
  • COCO
  • ADE20K
  • The hierarchical design and the shifted window approach also prove beneficial for all Perception Architectures.
  • Ratio of 1:1:3:1