The Urban Vision Hackathon Dataset and Models: Towards Image Annotations and Accurate Vision Models for Indian Traffic

The Urban Vision Hackathon (UVH-26) dataset, created by AIM@IISc, represents a significant advancement in the field of computer vision for Indian traffic scenarios. Comprising over 26,000 annotated images captured from CCTV cameras across Bengaluru, this dataset was developed through a collaborative hackathon involving 565 students. The primary objective of UVH-26 is to enhance image annotation quality and to improve the accuracy of vision models tailored to the complexities of Indian traffic conditions. By focusing on real-world traffic footage from a major Indian city, the dataset addresses a critical gap in existing vision datasets that often lack regional specificity. The hackathon format facilitated large-scale participation and diverse input, contributing to the dataset’s comprehensiveness and robustness. This initiative underscores the growing emphasis on localized data to train AI models that can better interpret and manage urban traffic environments in India. Overall, UVH-26 is recognized as a groundbreaking resource that could drive future innovations in traffic monitoring and management systems.