metadata
task_categories:
- image-to-image
tags:
- visual-editing
- reasoning
RISEBench++
This repository contains the dataset for Reasoning-Informed Visual Editing, which benchmarks reasoning-informed visual editing across six reasoning dimensions: temporal, causal, spatial, logical, counterfactual, and hybrid reasoning. It includes 1,000 human-annotated bilingual test cases in English and Chinese, supporting single-image, multi-image, and multi-turn editing.
Code and more information are available at the GitHub repository.