---
dataset_info:
  features:
  - name: image
    dtype: image
  - name: prompt
    dtype: string
  - name: category
    dtype: string
  - name: question_id
    dtype: string
  - name: ground_truth
    dtype: string
  splits:
  - name: test
    num_bytes: 281204341.0
    num_examples: 191
  download_size: 281191178
  dataset_size: 281204341.0
configs:
- config_name: default
  data_files:
  - split: test
    path: data/test-*
---

This is a mirror of craigwu/vstar_bench with images saved via PIL instead of filepaths, for ease of usage.