5fa1a76
1
2
We train a sequence Transformer to auto-regressively predict pixels, without incorporating knowledge of the 2D input structure.
