Hongxun Yao

Papers

Total Citations

425

H-Index

About

Hongxun Yao is a leading researcher in 3D computer vision, with a primary focus on point cloud completion, 3D object reconstruction, and stereo imaging. Her most impactful contribution is the development of GRNet (Gridding Residual Network), a pioneering framework for dense point cloud completion that addresses the critical challenge of reconstructing complete 3D shapes from partial, incomplete scans. Unlike conventional MLP-based methods that often lose fine structural details, GRNet introduces a novel gridding operation and residual learning to preserve geometric fidelity, achieving state-of-the-art performance. This work has garnered over 375 citations, underscoring its significance in enabling robust 3D perception for robotics, autonomous navigation, and augmented reality. Yao has also advanced 3D object reconstruction from stereo images, tackling the generalization limitations of template-matching approaches by proposing learning-based methods that infer shape directly from RGB pairs. Her research bridges the gap between sparse sensor data and dense, high-quality 3D models, making her a key figure in the push toward reliable, real-world 3D understanding.