Reproduce “Ancient-Bench: A Comprehensive Multi-millennial, Multi-medium, and Multi-script Benchmark for Ancient Chinese Artifact Text Recognition” 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

Hiuyi Cheng, Nuo Xu, Yuyi Zhang, Xuhan Zheng, Wei Pan, Jing Zhang, Dezhi Peng, Minghui Liao and 4 more

The paper introduces Ancient-Bench, a benchmark for recognizing ancient Chinese artifact text across 2,700 images from 14 institutions. It applies rules for symbols, character forms, and layout, then evaluates systems with NED and F1. Doubao-seed-2-0-lite-260428 reaches 52.08% NED and 57.86% F1 overall. Results make comparison possible, but recognition remains difficult; the benchmark does not show that standards improve performance.

Summary by alphaXiv.

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