Phys-GS Lite: Material-Aware Dynamic Gaussian Splatting

Tarun Gangadhar Vadaparthi

Demo

Dynamic water ripples induced via physics inspired deformation.

Abstract

Phys-GS Lite augments 3D Gaussian Splatting with semantic separation and physics-inspired dynamics. Using simple geometry heuristics, the method distinguishes rigid and deformable splats (e.g., ship vs water). A sinusoidal wind field introduces wave-like motion, while Laplacian smoothing ensures coherent deformations. This lightweight prototype demonstrates that physical priors can extend Gaussian Splatting toward dynamic, material-aware rendering without heavy neural training.

Method Overview

Pipeline diagram — Pipeline: semantic labeling → sinusoidal wind deformation → Laplacian smoothing.

Results

Semantic ship vs water — Semantic labeling of ship vs water splats.

Heatmap of motion intensity — Heatmap visualization of displacement intensity.