Reproduce “What Makes a Good Medical Image Tokenizer? Rethinking Reconstruction and Generation in Medical Image Tokenization” 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
Niklas Bubeck, Yundi Zhang, Vasiliki Sideri-Lampretsa, Julian McGinnis, Jiancheng Yang, Daniel Rueckert, Jiazhen Pan
The authors evaluate whether medical-image reconstruction scores can screen tokenizers for generation quality. They compare ten tokenizer families across twelve 2D MedMNIST+ datasets and three compression factors, training one fixed DiT-B/2 diffusion model on continuous latents. Reconstruction-FID and generation-FID correlate (Pearson r = 0.82) across 360 observations, supporting screening, not causation; results do not establish generality beyond this setup.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.