Urban Vision Hackathon

Date: 21–22 June 2025

As part of the Ministry of Education (MOE), Government of India (GoI)’s Centre of Excellence (CoE) in Artificial Intelligence for Sustainable Cities activity, a two-phase hackathon [https://airawat-mobility.github.io/hack/] was organised by the Indian Institute of Science (IISc), Bengaluru, with the prizes sponsored by Capital One. This was a national-level AI-driven initiative aimed at addressing India’s urban traffic congestion and mobility challenges. Conducted in two major phases, the hackathon combined the power of scalable India-centric AI models on annotated crowdsourced data collected from Bengaluru’s safe city camera. 

Goals for the hackathon

India faces some of the highest traffic congestion and poorest road safety statistics globally. Most existing AI models are trained on Western data and fail to adapt to the complexities of Indian roads. The hackathon aimed to:

  • Develop AI-powered solutions tailored to Indian traffic conditions.
  • Build the largest annotated vehicle detection dataset for Indian urban traffic using top-down CCTV footage.
  • Enable smart city planning, improved traffic flow prediction, optimised signal controls, and enhanced road safety.
  • Empower students, researchers, and developers to contribute to real-world societal issues.

Phase 1: The Traffic Quest Game (9 May–15 June 2025, online)

During Phase 1, 630+ teams across 120 cities (over 1500 students) participated in a gamified vehicle annotation challenge to label vehicle types across 15 Indian traffic vehicle classes in real urban scenes. They reviewed and corrected baseline annotations of vehicles from CCTV images from Bengaluru, added missing annotations and removed incorrect labels, and accurately labelled objects such as hatchback, truck, bus, and more from a list of 15 Indian traffic vehicle types.

Impact and outcomes:

  • 6.51 million bounding boxes for vehicle class added or confirmed.
  • 454,000 annotations made.
  • 70,000 unique images annotated.
  • Top 20 teams selected for advancement to Phase 2, based on overall accuracy and contributions.

Phase 2: AI Model Building (21–22 June 2025, IISc)

After the successful crowdsourced data annotation campaign in Phase 1, the top 20 teams (76 participants from across India) were invited to the IISc campus to compete in the next phase of the hackathon — training deep learning models on the newly annotated dataset. This in-person, hands-on event was the culmination of weeks of effort, designed to test not only the quality of the annotations but also the participants’ ability to build practical AI systems. The participants competed to train the best-performing AI models for multi-class vehicle detection on a subset of phase 1 data. The event included:

  • Hands-on model development using top-down traffic images.
  • Mentorship from experts from IISc and Capital One.
  • Provision of dedicated GPU computing infrastructure for training and testing, supported by the Kotak IISc AI–ML Centre.

Benefits and societal impact

  • Urban traffic solutions: data and models developed can assist in traffic prediction, congestion management, and signal optimisation.
  • Safer roads: better detection of all vehicles helps improve surveillance and emergency responses.
  • India-specific AI models: tailored models ensure higher accuracy in real-world Indian conditions.
  • Academic and industry collaboration: fosters collaboration between academia, government (Bangalore Traffic Police), and industry.
  • Skill building: encourages AI and data science skill development among Indian youth.