Dataset · hongzhuyi
Document and Table Pointwise Editing Benchmark
hongzhuyi/VDEBench
A set of 3,506 examples for evaluating models that edit text and tables on document scans via instructions.
What's inside
It contains 3,506 pairs of images and text instructions in English and Chinese. The task is to pointwise replace, add, or delete text, as well as fix the content or structure of tables on densely formatted pages. Meanwhile, the surrounding text, fonts, and layout must remain untouched.
What you can do with it
- Evaluating the quality of image-to-image document editing models
- Testing the preservation of typography and table structure during generative edits
- Validating pipelines for automated editing of scans and forms
- Tasks
- Image → image
- Languages
- English, Chinese
- Size
- 1K–10K rows
- Downloads this month
- 16.2K
- Likes
- 62
How to load it
from datasets import load_dataset
ds = load_dataset("hongzhuyi/VDEBench")