MixTok: Adaptive Visual Tokenization with Global 1D Tokens and Sparse 2D Refinement
-
Updated
Oct 14, 2026 - Python
MixTok: Adaptive Visual Tokenization with Global 1D Tokens and Sparse 2D Refinement
Generative modelling of tokenized medical images: autoregressive, MaskGIT, diffusion, flow matching, and Bayesian flow.
MixTok: Adaptive Visual Tokenization with Global 1D Tokens and Sparse 2D Refinement
First-Divergence Consequence Analysis for masked visual generators: theory, audited coupling/instrumentation, and reproducible quantization-shock experiments.
Patch-inpainting for Factorio blueprints (native 2.0 + Space Age): mask a 16x16 region, predict the entities. Leak-free 42x dataset, U-Net vs transformer vs MaskGIT, game-accurate FBSR renders.
Twelve modern text-to-image architectures (Stable Diffusion U-Net, DiT, MMDiT, FLUX single-stream, SANA, JiT, MeanFlow, Transfusion, AR, MaskGIT, VAR, MAR) in PyTorch, trained and benchmarked side by side on a CPU
To associate your repository with the maskgit topic, visit your repo's landing page and select "manage topics."