Vansh Garg

I am a robotics and computer vision researcher. I am currently a master’s student at the RRC (Robotics Research Center), IIIT-Hyderabad, where I am extremely fortunate to be advised by Prof. K Madhava Krishna. I have also had the pleasure of being guided by some amazing mentors like Sourav Garg and Siddharth Tourani.

My current research interests lie at the intersection of computer vision and robotics, specifically on improving the spatial and scene understanding of current vision models by grounding them in 3D geometry. I am interested in leveraging geometric priors and 3D-aware representations to improve robustness of these vision models. Recently, these ideas were reflected by our work on improving segment matching and navigation by utilizing the geometric priors of 3D foundation models like MASt3R

When I am not doing research, I usually like to listen to music, go running, or learn guitar. I love receiving emails and messages, so feel free to reach out to me if you have any questions or want to talk about anything related to research or otherwise.

news

Sep 18, 2025	Our work on geometry-grounded segment matching was accepted to NeurIPS 2025 as a spotlight! 🎉 See you in San Diego this December!

publications

NeurIPS

SegMASt3R: Geometry Grounded Segment Matching

Rohit Jayanti^*, Swayam Agrawal^*, Vansh Garg^*, and 4 more authors

In The Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025

* denotes equal contribution

Abs HTML

Segment matching is an important intermediate task in computer vision that establishes correspondences between semantically or geometrically coherent regions across images. Unlike keypoint matching, which focuses on localized features, segment matching captures structured regions, offering greater robustness to occlusions, lighting variations, and viewpoint changes. In this paper, we leverage the spatial understanding of 3D foundation models to tackle wide-baseline segment matching, a challenging setting involving extreme viewpoint shifts. We propose an architecture that uses the inductive bias of these 3D foundation models to match segments across image pairs with up to 180 degree view-point change rotation. Extensive experiments show that our approach outperforms state-of-the-art methods, including the SAM2 video propagator and local feature matching methods, by up to 30% on the AUPRC metric, on ScanNet++ and Replica datasets. We further demonstrate benefits of the proposed model on relevant downstream tasks, including 3D instance mapping and object-relative navigation.