VisionTransformer (ViT) Attention Maps using MNIST

Code walkthrough (YouTube Video)

An attention map for a test image (code)

An attention map for query and key images (code)

---

Videos you might be interested in

Vision Transformer (ViT) in JAX

Swin Transformer Code

Swin Transformer Attention

References

@misc{dosovitskiy2021image,
  title         = {An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale},
  author        = {Alexey Dosovitskiy and Lucas Beyer and Alexander Kolesnikov and Dirk Weissenborn and Xiaohua Zhai and Thomas Unterthiner and Mostafa Dehghani and Matthias Minderer and Georg Heigold and Sylvain Gelly and Jakob Uszkoreit and Neil Houlsby},
  year          ={2021},
  eprint        = {2010.11929},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CV}
}