MM-Hallu/Common-O-Bench
收藏Hugging Face2026-04-25 更新2026-05-03 收录
下载链接：
https://hf-mirror.com/datasets/MM-Hallu/Common-O-Bench
下载链接
链接失效反馈官方服务：
资源简介：
---
dataset_info:
features:
- name: image_1
dtype: image
- name: image_2
dtype: image
- name: question
dtype: string
- name: answer
dtype: string
- name: objects_1
dtype: string
- name: objects_2
dtype: string
- name: num_objects_image_1
dtype: int64
- name: num_objects_image_2
dtype: int64
- name: question_template
dtype: string
- name: answer_type
dtype: string
- name: choices
dtype: string
- name: num_choices
dtype: int64
- name: num_ground_truth_objects
dtype: int64
- name: real_or_synthetic
dtype: string
- name: ground_truth_objects
dtype: string
configs:
- config_name: default
data_files:
- split: train
path: data/main-*.parquet
license: cc-by-4.0
task_categories:
- visual-question-answering
language:
- en
tags:
- cross-scene-reasoning
- hallucination
- multi-image
- benchmark
size_categories:
- 10K<n<100K
---
# Common-O-Bench
Benchmark for evaluating cross-scene reasoning hallucinations in VLMs. 10,426 question pairs asking "what's in common?" across two images.
## Fields
| Field | Description |
|-------|-------------|
| image_1 | First input image |
| image_2 | Second input image |
| question | Cross-scene reasoning question |
| answer | Ground truth answer |
| objects_1/2 | Objects in each image |
| num_objects_image_1/2 | Object counts |
| question_template | Question template |
| answer_type | Answer type |
| choices | JSON-encoded choices |
| real_or_synthetic | Image source type |
| ground_truth_objects | JSON-encoded ground truth |
Source: [facebook/Common-O](https://huggingface.co/datasets/facebook/Common-O)
提供机构：
MM-Hallu
