IISc Team Takes Home Top Prize at World‑Renowned Computer Vision Conference

When the curtains fell at the International Conference on Computer Vision (ICCV) this week, a modest banner bearing the IISc logo was swamped by applause. The reason? A research paper, spearheaded by Prof. Srinivas Rao and his Ph.D. scholars, had just been crowned the best paper of the entire event.

It felt almost surreal for the Bangalore‑based team. After months—no, years—of sleepless nights, countless coffee‑fuelled brainstorming sessions, and a few dead‑ends that made them wonder if they were chasing a mirage, their hard‑earned insights finally resonated with a global audience of over 10,000 researchers.

The award‑winning work, titled “Hierarchical Cross‑Modal Fusion for Unsupervised Video Understanding,” tackles a problem that has haunted computer‑vision scientists for decades: how to make machines understand videos without relying on massive labelled datasets. By cleverly marrying contrastive learning with a novel attention‑based hierarchy, the authors demonstrated a leap in accuracy—up to 15 % better than the previous state‑of‑the‑art—on benchmark datasets such as Kinetics‑600.

“We didn’t set out to win anything,” said Dr. Ananya Patel, the lead author, laughing. “Our goal was simply to push the envelope a little further, to see if the ideas we’d been tinkering with in the lab could actually work in the wild. The fact that they did—and that the community recognised it—was a pleasant surprise.”

Beyond the technical triumph, the accolade carries a symbolic weight. India’s AI ecosystem has been burgeoning, yet international recognitions at top‑tier venues have remained scarce. This win, therefore, is not just a feather in IISc’s cap; it’s a beacon for the country’s broader research community, signaling that home‑grown talent can stand shoulder‑to‑shoulder with the world’s best.

University officials were quick to celebrate. The Director of IISc, Prof. M. V. Venkatesh, remarked, “Such achievements are the fruit of sustained investment in fundamental research and the relentless curiosity of our scholars. We hope this inspires a new generation of students to dive deep into AI, machine learning, and beyond.”

Looking ahead, the team is already sketching the next steps. Plans are afoot to open‑source their codebase, collaborate with industry partners for real‑world deployments, and perhaps most excitingly, extend the framework to other modalities like audio and text.

For now, though, the IISc researchers can relish a well‑earned moment of glory, their names etched into the annals of one of the most prestigious gatherings in computer vision. And somewhere in the bustling corridors of the conference centre, a few seasoned reviewers were seen nodding appreciatively, perhaps whispering, “Finally, a piece of work that truly bridges theory and practice.”