Sam20202/UVH-26

ArXiv | Models Models trained on UVH-26 deliver up to 31.5% higher mAP than COCO-pretrained baselines, demonstrating significant gains in real-world performance for Indian traffic scenarios. Dataset Card for UVH-26 (Urban Vision Hackathon Dataset) Dataset Summary UVH-26 is a large-scale, India-specific traffic-camera image dataset released by AIM @ IISc for research in intelligent transportation systems and… See the full description on the dataset page: https://huggingface.co/datasets/Sam20202/UVH-26.