Reproduce “Self-distillation Regularized Connectionist Temporal Classification Loss for Text Recognition: A Simple Yet Effective Approach” 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
Ziyin Zhang, Ning Lu, Minghui Liao, Yongshuai Huang, Cheng Li, Min Wang, Wei Peng
DCTC changes training for CTC-based image text recognition by retaining sequence-level CTC loss and adding framewise character supervision. It estimates frame alignment from the model’s outputs, without changing model structure. On CRNN, it raises aggregate accuracy from 77.3% to 79.9% across six English benchmarks and from 68.7% to 70.8% across four Chinese subsets; these gains do not imply improvement on every dataset.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.