README.md

metadata

license: cc-by-3.0
task_categories:
  - visual-question-answering
language:
  - en
tags:
  - vision-language
  - benchmark
  - domain-specific
pretty_name: DomainBench
size_categories:
  - 10K<n<100K

DomainBench

DomainBench is a comprehensive benchmark for evaluating vision-language models across different domains.

Dataset Structure

The benchmark consists of 7 domain-specific datasets:

• Animals
• Food
• Kitchen
• KITTI
• Person
• Vehicles
• Wildlife

Each domain contains:

• Images folder with visual content
• Question-answer pairs in JSON format

Tasks

The benchmark includes various visual reasoning tasks:

1. Object Detection
2. Spatial Relationships
3. Color Analysis
4. Depth Perception
5. Object Interaction
6. Puzzle Solving
7. Image Variant Analysis
8. Object Condition Analysis