Reproduce “CLEval: Character-Level Evaluation for Text Detection and Recognition Tasks” 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
Youngmin Baek, Daehyun Nam, Sungrae Park, Junyeop Lee, Seung Shin, Jeonghun Baek, Chae Young Lee, Hwalsuk Lee
CLEval introduces a new character-level evaluation metric for text detection, recognition, and end-to-end Optical Character Recognition (OCR) tasks. This framework provides a more fair, reliable, and fine-grained assessment by explicitly handling granularity and correctness issues that traditional metrics fail to capture.
Key insights
- Existing metrics often provide binary scores (0 or 1) that fail to reflect partially correct detections or recognitions, leading to a mismatch with human perception.
- Character-level evaluation, facilitated by Pseudo-Character Centers, enables a fine-grained, proportional scoring system that accurately credits partial successes and penalizes specific error types.
- Quantifying specific error attributes like splits, merges, and missing characters provides actionable diagnostic feedback, aiding in targeted model improvements.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.