README.md

metadata

language:
  - en
license: mit
size_categories:
  - n<1K
tags:
  - vision-language
  - reward-modeling
  - benchmark
task_categories:
  - image-text-to-text

Benchmark Data for ViLBench: A Suite for Vision-Language Process Reward Modeling

arXiv | Project Page

There are 600 data collected from 5 existing vision-language tasks