6 months ago
2
The AI for Integrated Mobility (AIM@IISc) inaugural astatine the Indian Institute of Science (IISc) has announced the nationalist merchandise of UVH-26, a large-scale postulation representation dataset on with a suite of fine-tuned entity detection models tailored for Indian municipality postulation conditions.
UVH-26 contains 26,646 high-resolution postulation images sourced from Bengaluru’s Safe City task and features 1.8 cardinal bounding-box annotations crossed 14 India-specific conveyance classes, including two-wheelers, autorickshaws, airy commercialized vehicles, and buses.
According to a release, the dataset was curated done a nationwide Urban Vision Hackathon that progressive much than 560 pupil volunteers, who contributed to a ample crowdsourced annotation effort. High-quality statement crushed information was generated utilizing Majority Voting and STAPLE algorithms.
Alongside the dataset, AIM@IISc has released six state-of-the-art entity detection models based connected YOLOv11, DAMO-YOLO, and RT-DETRv2 architectures, fine-tuned specifically for Indian postulation scenes. These models amusement up to a 31.5% betterment successful mAP@50:95 show compared to COCO-trained baselines, demonstrating the value of region-specific information for intelligent proscription systems.
The dataset and models person been made disposable connected HuggingFace nether open-source licences. Since their merchandise connected November 5, 2025, the UVH-26 resources person already recorded much than 1,500 downloads and are presently ranked arsenic the apical trending entity detection models connected the platform, indicating beardown planetary involvement successful mobility-focused AI probe from the Global South, said the release.
AIM@IISc has invited researchers, developers, students, and policymakers to research and physique upon UVH-26 to beforehand AI-enabled proscription solutions successful emerging economies. The inaugural focuses connected the usage of AI, data-driven methods, and integer duplicate platforms to tackle mobility challenges specified arsenic congestion, safety, and municipality planning.
The merchandise was made imaginable with enactment from the Bengaluru City Police and Bengaluru Traffic Police, who provided entree to imagery from the Safe City project. Additional enactment came from IISc’s ARTPARK, the Kotak IISc AI-ML Centre, Capital One, and outreach partners including ACM India Council and IEEE India Council.
“India has ever been data-poor for AI applications, particularly successful mobility, fixed the standard and complexity of Indian traffic. UVH-26 is simply a testament to however crowdsourced efforts by hundreds of engaged students tin make high-quality datasets that rival planetary benchmarks. This shows the powerfulness of community-driven AI for solving section challenges. By making these datasets and exemplary openly available, we purpose to catalyse research, improvement and deployment of AI-driven postulation analytics for safer and smarter cities,” said Yogesh Simmhan, Associate Professor astatine CDS, IISc.
