GeoDrive Bench A multi country driving scene benchmark for evaluating vision language models on culture and region specific traffic knowledge. Statistics 5,053 multiple choice questions 6 countries: China (cn), USA (us), UK (uk), Japan (jp), Singapore (sg), India (ind) 4 task categories: perception, prediction, planning, region 7,160 unique driving images from 6 source datasets: nuScenes (Singapore) ONCE (China) IDD (India) CoVLA (Japan) LingoQA (UK) Waymo (USA) Schema Field Type Description id int Unique question id country str Country code (cn / us / uk / jp / sg / ind) image path list[str] Paths to driving images (relative to repo root) question str Question text options list[str] Four answer options (A / B / C / D) answer str Ground truth letter (A / B / C / D) question type str Always multiple choice question category str One of: perception, prediction, planning, region rule reference list[str] Relevant traffic rule IDs (e.g., S3 , S8 ) explanation str Human written explanation of the answer Usage datasets library Croissant (mlcroissant) Manual JSON Citation
Runs entirely in your browser via DuckDB-Wasm — this dataset's real data file is loaded once, then queried locally. Nothing is sent to a server.
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy