Japan OCR Mini Benchmark is a compact benchmark for evaluating whether OCR/VLM models can read Japanese receipts and return structured JSON. It covers receipt-level fields, item rows, tax summaries, totals, payment methods, and clean/noisy image variants.
The current repository entry point includes the frozen v0.2.0 dataset payload, the v0.4.0 Clean/Noisy LM Studio leaderboard, the v0.4.2 receipt-generation design-candidate… See the full description on the dataset page: https://huggingface.co/datasets/K10124/japan-ocr-mini-benchmark.
