Reproduce “Training Protocol Matters: Towards Accurate Scene Text Recognition via Training Protocol Searching” 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
Xiaojie Chu, Yongtao Wang, Chunhua Shen, Jingdong Chen, Wei Chu
The authors search training protocols for existing scene text recognizers using an evolutionary algorithm and a cheaper proxy evaluation, rather than proposing a new architecture. On CRNN, the searched protocol raises accuracy from 84.3% to 88.2% across seven real-world benchmarks. TRBA-Net reaches 92.3%, versus EFIFSTR’s 90.2%, on six benchmarks excluding IC03; inference-time comparisons concern complete systems, not protocol-induced speedups.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.