TL;DR of the paper, generated automatically.

A research paper proposes dNaViT, a discrete visual tokenizer for images.

• What it is: a native-resolution vision transformer that turns images into discrete codes, like language tokens.
• Results: the paper says it matches or beats task-specific baselines on 12 understanding, 4 generation and 1 editing benchmark.
• Code: the authors say code and a project page are available.