Keze Wang

Sun Yat-sen University, University of California, Los Angeles

Papers

Total Citations

111

H-Index

About

Keze Wang is a computer vision and machine learning researcher whose work spans human pose estimation, deep learning, and reinforcement learning for robotics. His most prominent contribution, "3D Human Pose Machines with Self-supervised Learning" (2019), has garnered over 100 citations and represents a significant advance in recovering three-dimensional human poses from visual data — a notoriously difficult problem complicated by diverse appearances, varying viewpoints, occlusions, and inherent geometric ambiguities. By incorporating self-supervised learning, Wang and his collaborators reduced reliance on expensive labeled data while maintaining strong predictive performance, making the approach both practical and scalable for real-world computer vision and robotic applications. Beyond pose estimation, Wang has explored the challenges of sample efficiency in deep reinforcement learning through his work on "Continuous Transition," which applies MixUp-based data augmentation to continuous robotic control tasks. This line of research addresses a critical bottleneck in deploying RL systems to real-world environments. Together, his publications reflect a consistent focus on bridging the gap between theoretical machine learning advances and practical robotic perception challenges, establishing him as a thoughtful contributor to the intersection of computer vision, self-supervised learning, and autonomous systems research.