GeoDrive-Bench

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)… See the full description on the dataset page:

Source: Hugging Face Hub (GeoDriveBench/GeoDrive-Bench). Metadata imported from the dataset’s Hub tags.