Datasets:
Download eval.yaml from walled120/VDEBench: direct link, hf CLI and curl.
- Browser
- Download file 689 Bytes
-
https://huggingface.co/datasets/walled120/VDEBench/resolve/main/eval.yaml
- Command line
-
hf download hf://datasets/walled120/VDEBench/eval.yaml
-
curl -L -o eval.yaml https://huggingface.co/datasets/walled120/VDEBench/resolve/main/eval.yaml
689 Bytes
| name: VDEBench | |
| description: > | |
| VDEBench (Visual Document Editing Benchmark) evaluates instruction-guided | |
| image editing on 3,506 dense English and Chinese document pages: text | |
| replacement, addition and deletion, plus table content and structure edits. | |
| Edited pages are parsed with PaddleOCR-VL and compared with the reference | |
| page. Each task reports the equally weighted mean of four OCR metrics, IoU, | |
| CDM, BLEU-4 and TEDS (0-1, higher is better). The local task scores the | |
| annotated editing region and is the headline setting. The global task scores | |
| the full page. | |
| tasks: | |
| - id: local | |
| config: default | |
| split: test | |
| - id: global | |
| config: default | |
| split: test | |