Reproduce “TinyChart: Efficient Chart Understanding with Visual Token Merging and Program-of-Thoughts Learning” 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
Liang Zhang, Anwen Hu, Haiyang Xu, Ming Yan, Yichen Xu, Qin Jin, Ji Zhang, Fei Huang
TinyChart answers chart questions by preserving 768×768 input while merging similar visual patches and training a 3B model to generate executable Python for arithmetic. On ChartQA, its average relaxed accuracy is 83.60, versus 79.90 for 13B ChartAst at 448×448, with throughput of 3.14 versus 1.47 it/s. The lead is not present on every subset.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.