Pou-Chun Kung

Papers

Total Citations

H-Index

About

Pou-Chun Kung is a rising researcher at the forefront of 3D scene representation and novel view synthesis, with a focus on bridging the gap between visual and acoustic sensing. His primary research areas include neural rendering, Gaussian splatting, and depth-supervised 3D reconstruction. Kung’s most significant contribution is **SAD-GS (Shape-aligned Depth-supervised Gaussian Splatting)**, a 2024 work that introduces a shape-aligned depth supervision strategy to dramatically improve the accuracy of 3D geometry reconstruction from Gaussian splatting. This method has already garnered 4 citations, signaling its importance for applications in dynamic scene reconstruction and real-time simulation. In a bold extension of his work, Kung authored **SonarSplat** (2025), the first Gaussian splatting framework designed specifically for imaging sonar. This pioneering method achieves realistic novel view synthesis of underwater environments while modeling complex acoustic phenomena like streaking, representing a major leap in autonomous underwater vehicle perception. By representing scenes with 3D Gaussians that encode acoustic reflectance and saturation, Kung has opened a new frontier in sensor-agnostic scene understanding. His work is notable for its technical rigor and its potential to transform how machines perceive both visual and acoustic worlds.