Papers
7
Total Citations
145
H-Index
5
About
Daizong Liu is a researcher whose work spans computer vision, 3D scene understanding, and adversarial machine learning, with a particular focus on the security and robustness of deep learning systems in safety-critical applications. His early notable contribution, "SAANet: Siamese Action-Units Attention Network for Improving Dynamic Facial Expression Recognition" (2020, 82 citations), demonstrated his ability to advance human-centered vision tasks through attention-based architectures. Liu has since become a prominent voice in 3D point cloud adversarial research, investigating how deep learning models deployed in autonomous driving, robotics, and surveillance remain vulnerable to sophisticated attacks. His works — including "3DHacker," "Robust Geometry-Dependent Attack," and "Explicitly Perceiving and Preserving the Local Geometric Structures for 3D Point Cloud Attack" — collectively push the boundaries of adversarial attack methodology by tackling hard-label settings, geometric structure preservation, and decision boundary generation. Beyond adversarial robustness, Liu has contributed to 3D scene grounding through "Dense Object Grounding in 3D Scenes" (2023), bridging natural language understanding with spatial reasoning. His most recent work on backdoor attacks against text-guided 3D scene grounding signals an expanding research agenda addressing emerging multimodal vulnerabilities, making his contributions increasingly relevant to real-world AI safety.
Research Focus
Key Achievements
Top Papers
- 1
- 2
- 3Robust Geometry-Dependent Attack for 3D Point Clouds16 citations · 2023
- 4
- 5Dense Object Grounding in 3D Scenes11 citations · 2023
- 6Imperceptible Backdoor Attacks on Text-Guided 3D Scene Grounding4 citations · 2025
- 7