Yash

Yash Thube

I work on making machines recover metric, physically valid world states from unconstrained visual observations. Broadly, my interests revolve around 3D/4D Computer Vision, Representation Learning, and Spatial intelligence for science and sports.

I believe true intelligence must intuitively understand space, time, and causality in ways that go beyond a human brain, neural architectures to respect newtonian mechanics, projective geometry, and strict physical constraints…..apart from work, I ramble here !

Currently building the core at Future Sportler (India). My day-to-day research and implementation spans multi-view geometry, inverse biomechanics/kinematics, and 4D reconstruction (HMR, 3DGS, NeRFs) to handle complex motion and physics across diverse sports and camera angles, tackling temporal segmentation under heavy occlusion, MOT, action recognition, and instruction-tuning VLMs, depth estimators, and diffusion models. Always open to collaborations!