IISc Breakthrough Wins Top Honor at Global Computer Vision Conference

When the results were announced at the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) in June 2024, the applause for the Indian Institute of Science (IISc) team was unmistakable. Their paper, titled “Sublime: Self‑Supervised Multimodal Depth Estimation for Adverse Weather Conditions,” took home the coveted Best Paper Award, a first for any Indian institution at this premier global gathering.

The work, led by Professor C. V. Jawahar along with Dr. Karthik Rao and graduate students Ananya Mehta and Rohan Gupta, tackles a problem that has long nagged computer‑vision engineers: how to reliably gauge depth when the scene is shrouded in fog, drenched by rain, or dimmed by night. Their solution weaves together a transformer‑based architecture with self‑supervised learning, while simultaneously ingesting data from LiDAR, radar and RGB cameras. The result is a model that, according to the authors, outperforms existing benchmarks by up to 12 % in the most challenging conditions.

“We wanted to move beyond the ‘clean‑lab’ scenario,” says Dr. Rao, recalling the countless hours of debugging and data‑augmentation experiments. “The real world is messy, and autonomous vehicles or rescue drones can’t wait for perfect lighting.” He adds that the team’s approach also slashes the need for massive labeled datasets—a boon for deployments in low‑resource settings.

Judges at CVPR praised the paper not just for its technical depth but also for its practical relevance. “It’s rare to see a method that is both theoretically elegant and immediately deployable,” noted one reviewer in the official citation. The award, accompanied by a modest cash prize and a spotlight talk, has already sparked interest from industry giants and start‑ups alike.

Back in Bangalore, the response was enthusiastic. “This is a proud moment for IISc and for Indian AI research,” remarked Director of the Department of Computer Science and Automation, Prof. Manish Gupta. Funding agencies, including the Science and Engineering Research Board (SERB) and the Department of Science & Technology, also lauded the achievement, hinting at increased support for interdisciplinary projects that bridge perception and robotics.

While the accolades are fresh, the team is already eyeing the next steps—adapting the algorithm for edge devices and extending it to underwater vision. As Professor Jawahar puts it, “Winning the award is wonderful, but the real win is seeing our work move from conference halls to real‑world impact.”