Nasser4963/jamo-gold

JAMO Gold — Human-Validated Hangul Rendering Judgments A small, carefully labelled dataset for a specific purpose: testing whether an OCR engine can tell that a glyph is not a real character. It is not a Hangul OCR training set, and not a model leaderboard. Why this exists Benchmarks that measure visual text rendering score generated images with OCR. But OCR engines are closed-set classifiers — given an image, they must return some character from their… See the full description on the dataset page: https://huggingface.co/datasets/Nasser4963/jamo-gold.