| license: other | |
| language: | |
| - fa | |
| tags: | |
| - optical-character-recognition | |
| - ocr | |
| - image-to-text | |
| - persian | |
| pretty_name: Persian OCR Evaluation Dataset | |
| size_categories: | |
| - n<1K | |
| configs: | |
| - config_name: default | |
| data_files: | |
| - split: test | |
| path: bench_data/ocr_eval.jsonl | |
| # Persian OCR Evaluation Dataset | |
| This benchmark contains paired Persian document images and UTF-8 transcription | |
| targets for OCR evaluation. Each JSONL row references one image under | |
| `bench_data/images/` and contains its transcription in the `text` field. | |
| ## Schema | |
| - `image`: image path relative to the repository | |
| - `id`: stable image identifier | |
| - `page`: page number, currently `1` | |
| - `type`: evaluation item type, currently `transcription` | |
| - `text`: reference transcription | |
| - `language`: `fa` | |
| - `checked`: local-label status | |
| - `notes`: reviewer notes |