Reproduce “MixTok: Adaptive Visual Tokenization with Global 1D Tokens and Sparse 2D Refinement” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Chenge Li, Yefei He, Xinyu Liu, Feng Chen, Hong Zhou, Bohan Zhuang

MixTok combines a compact 1D global representation with selectively added 2D features to reconstruct images while allocating local detail by image content. It expands the 1D features into a spatial grid and replaces selected locations with 2D features. On ImageNet at 256 × 256, it achieves rFID 1.1950 at about 128 average tokens, but trails two methods at that budget.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.