Reproduce “Texo: Formula Recognition within 20M Parameters” 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

Sicheng Mao

Texo converts formula images into LaTeX using a LaTeX-specific tokenizer and transferred decoder embeddings, reducing its vocabulary from 50,000 to 687 tokens and its model to 20M parameters. On UniMER-Test, it scores higher than PPFormulaNet-S across all four CDM categories, but below UniMERNet-T; its measured GPU inference time is also slower than PPFormulaNet-S.

Summary by alphaXiv.

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