I am a researcher working at the intersection of computer vision, 3D geometry, and AI. I earned my Master’s in Computer Vision from Carnegie Mellon University's Robotics Institute, where I worked with Shubham Tulsiani, Fernando De la Torre, László A. Jeni, and Jessica Hodgins.
My research focuses on inferring structured representations of the physical world such as geometry, motion, illumination, and material appearance from images and video. I am particularly interested in 4D dynamic scene reconstruction, and hybrid explicit-implicit representations. A central objective of my work is to develop compact, temporally consistent, and interpretable models that can reconstruct complex scenes under challenging conditions such as occlusion and sparse supervision. I also explore the use of diffusion models for fine-grained refinement and uncertainty-aware reasoning in dynamic environments.
SHaDe is a modular framework for MLP-free dynamic 3D reconstruction that introduces structured tri-plane deformation, SH-based radiance attention, and temporally consistent refinement via latent diffusion. It achieves compact and high-fidelity 4D reconstruction.