README.md

metadata

language:
  - en
  - zh
task_categories:
  - visual-question-answering
pretty_name: OmniMapBench
size_categories:
  - 1K<n<10K

OmniMapBench

OmniMapBench is a map-centric visual question answering benchmark. This release contains 2096 QA samples over 1603 unique map images.

Dataset Structure

The repository is organized as:

README.md
dataset_infos.json
metadata.jsonl
data/
images/

Each row in metadata.jsonl contains:

• image_path: relative image path inside this dataset repo.
• id: original QA id.
• question: question text.
• choices: answer candidates.
• answer: ground-truth answer.
• qa_type: single_choice, multi_choice, or sort.
• capability_level: level_1, level_2, or level_3.
• language: en or zh.
• type: map category.