A research paper from the Indian Institute of Science (IISc) has been selected as the best paper finalist at the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2026.

The CVPR is one of the world’s premier conferences in computer vision and artificial intelligence (AI), and this year’s conference was held at Colorado in the U.S.

The IISc team, comprising R. Venkatesh Babu, Priyam Dey, Aditya Sahdev, Sunny Bhati, Konda Reddy Mopuri from the Department of Computational and Data Sciences (CDS), was selected as the best paper finalists for the paper on “Rethinking Dataset Distillation: Hard Truths about Soft Labels”.

Prof. Babu, who is the chair of the CDS, said that the paper focuses on the very important subject of dataset distillation.

“This (paper) is about dataset distillation. There is a huge amount of data, and we use it for training various models. We don’t know whether all the data we use for training are useful or not. We don’t need millions of samples, only thousands of samples are enough to train the model,” Prof. Babu told The Hindu.

He added that apart from helping in data efficient learning and improving scaling performance of the models, their research also helps bring down the carbon emission.

“Dataset distillation has emerged as a promising direction for efficient learning, and this work revisits some of its core assumptions, uncovering important insights about the role of soft labels and their limitations,” the CDS said.

The IISc paper was recognised among the top 15 papers out of more than 16,000 submissions made at the CVPR 2026.

“From 16,000 submissions, around 4,000 papers were selected and from this we made it to the top 15, which is a significant achievement for the team,” Prof. Babu added.