Feixiang Xu

University of Electronic Science and Technology of China

Papers

Total Citations

112

H-Index

About

Feixiang Xu is a computer vision and human-computer interaction researcher whose work centers on human action recognition, particularly from arbitrary and varying viewpoints — a critical challenge in real-world robotics and surveillance applications. His most influential contribution is the creation of large-scale RGB-D databases designed specifically to support arbitrary-view human action recognition, addressing a significant gap in available research resources. His 2018 dataset paper has garnered 77 citations, reflecting the community's urgent need for more versatile and comprehensive benchmarks that move beyond the constraints of single-view and fixed multi-view setups. Building on this foundation, Xu extended his dataset work in 2019 with a varying-view RGB-D action dataset, further enabling progress in human-robot interaction (HRI) scenarios. Complementing his dataset contributions, he developed the Attention Transfer (ANT) Network, a novel deep learning approach that tackles view-invariant action recognition by addressing action occlusion and information loss caused by perspective changes — moving beyond conventional common feature space methods. Together, these contributions establish Xu as a key figure in bridging the gap between controlled laboratory action recognition and the demands of dynamic, real-world deployment environments.