ViT-based image tokenizer

How this idea evolved

Suggested from citations. No curator has recorded what this idea builds on yet. These ideas from the same theme come from papers that ViT-VQGAN cites, directly or one step removed. Citation is a fact; the connection between the ideas is not verified.

• Autoregressive text-to-image TransformerDALL·E · 2021cited directly

  Model text and image tokens as one stream and sample images from text.

• Discrete visual tokens (VQ-VAE)VQ-VAE · 2017cited directly

  Compress images into a vocabulary of discrete codes so Transformers can model them like text.