Hafeez Khan, a Ph.D. student in computer science and graduate research assistant in professor Siddhartha Bhattacharyya’s ASSIST Lab, won first place out of 200 competitors in the Computationally Optimal Gaussian Splatting (COGS) 2025 Competition in October.
The competition, held in Honolulu, Hawaii, was hosted by Meta and the University of Toronto at the 2025 International Conference on Computer Vision (ICCV). ICCV is considered one of the most prestigious and selective conferences in the field of computer vision, which is a branch of artificial intelligence that teaches machines to process, analyze and interpret visual inputs, such as images and videos, using machine learning.
This year’s competition focused on 3D Gaussian Splatting (3DGS): a novel, state-of-the-art technique that enables real-time rendering of 3D photorealistic scenes from 2D image samples. It aims to improve the quality of digital 3D landscapes used with virtual tours, augmented reality or interactive maps.
Participants were challenged to build the fastest 3D Gaussian Splatting (3DGS) pipeline using lossless data compression techniques to drastically reduce model size without losing image quality. This makes 3DGS practical on low-power, storage-limited devices such as smartphones and VR headsets.
Khan, who studies computer vision and machine learning, won by achieving the highest compression rates while maintaining high image quality.
“It was a proud moment watching our student grow and shine as he gets into the competitions of the real world,” Bhattacharyya said.
In his winning solution, Khan developed a three-stage method that integrates two pruning strategies, which eliminate redundant data, and quantization, which involves converting large sets of data into smaller values. He had four weeks to develop his solution – a challenge that reminded him how much more there is to learn and achieve.
“I’m only just getting started,” Khan said. “Conferences like these really push us to bring our best work forward and grow as researchers. More than that, they bring together some of the brightest minds in the community, people whose work I’ve followed and learned from so much.”
Khan also presented two papers at the conference: “Adapt, But Don’t Forget: Fine-Tuning and Contrastive Routing for Lane Detection under Distribution Shift” in collaboration with NASA Langley Research, and “Test-time Prompt Refinement for Text-to-Image Models” in collaboration with Microsoft Research.
