Gang Wei

Papers

Total Citations

H-Index

About

Gang Wei is a researcher advancing the frontiers of 3D computer vision, with a focus on point cloud learning and zero-shot perception. His work addresses critical challenges in unstructured 3D data analysis, which has direct applications in robotics, autonomous driving, and spatial intelligence. In his influential paper "PointAGCN: Adaptive Spectral Graph CNN for Point Cloud Feature Learning" (2018), Wei introduced a novel graph-based convolutional architecture that overcomes limitations of earlier methods like PointNet by better capturing local geometric structures in point clouds. Although early in its citation trajectory, this work laid important groundwork for adaptive spectral filtering on irregular 3D data. More recently, in "RE0: Recognize Everything with 3D Zero-Shot Instance Segmentation" (2025), Wei tackles the fundamental challenge of generalizing 3D segmentation to unseen object categories—a critical bottleneck for real-world robotics. By leveraging vision foundation models, his approach circumvents the scarcity of high-quality 3D training data, enabling recognition of arbitrary objects without task-specific fine-tuning. Wei’s research bridges the gap between 2D foundation models and 3D perception, pushing toward more generalizable and practical computer vision systems.