Xinge Zhu
Papers
4
Total Citations
25
H-Index
3
About
Xinge Zhu is a rising star in computer vision, whose research focuses on the robustness and real-world applicability of 3D scene understanding. His work critically examines the vulnerability of deep neural networks, as demonstrated in his highly cited 2020 paper on adversarial attacks against monocular depth estimation—a foundational contribution that highlights the security risks in deploying these models. Moving from attacking to building, Zhu has pioneered methods for grounding natural language in complex, dynamic 3D environments. His WildRefer framework (2023, 2024) is a landmark achievement, enabling 3D object localization in large-scale, multi-modal scenes by fusing 2D images and LiDAR point clouds with linguistic descriptions. This work tackles the critical challenge of human-centric robotics. Furthering this, his HUNTER system (2024) introduces an unsupervised approach for detecting humans in real scenes by transferring knowledge from synthetic data, directly addressing the scarcity of labeled real-world data. With a growing citation count and a clear trajectory from foundational security analysis to cutting-edge multi-modal perception, Zhu is establishing himself as a key innovator in making 3D vision systems both safer and more capable.
Research Focus
Key Achievements
Top Papers
- 1Adversarial Attacks on Monocular Depth Estimation14 citations · 2020
- 2
- 3
- 4