Reproduce “Query Independent Variable Rate Visual Token Coding” 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
Hongbo Zhang, Zihao Yang, Liuyang Song, Daqian Yang, Haoyang Yao, Yan Wen, Zhengtao Yao
RDTok compresses visual representations by keeping every token and allocating variable bit rates from image-only distortion–rate measurements under a fixed budget. On ChartQA with Qwen3-VL-8B-Instruct at 256 coefficient bits per token, it reports output KL of 0.166, versus 0.327 for uniform-rate coding. This compresses stored features, not the token sequence or model computation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.