Zihui Xue
Papers
2
Total Citations
8
H-Index
1
About
Zihui Xue is a rising researcher whose work sits at the intersection of computer vision, robotics, and augmented reality, with a core focus on bridging the extreme perceptual gap between egocentric (first-person) and exocentric (third-person) video perspectives. Her major contribution lies in developing novel representation learning techniques that enable view-invariant understanding of human actions without requiring expensive, synchronized multi-view datasets. In her highly cited 2023 paper, "Learning Fine-grained View-Invariant Representations from Unpaired Ego-Exo Videos via Temporal Alignment" (7 citations), Xue pioneered a method to align unpaired ego and exo videos temporally, learning robust features that generalize across dramatically different viewpoints—a critical capability for applications like robot imitation learning and AR assistance. Building on this foundation, her 2025 work "Viewpoint Rosetta Stone: Unlocking Unpaired Ego-Exo Videos for View-invariant Representation Learning" (1 citation) introduces VIEWPOINTROSETTA, a scalable framework that leverages large-scale unpaired data to learn clip-level viewpoint-invariant representations. By eliminating the need for paired recordings, Xue’s research significantly reduces data collection costs while advancing the fundamental challenge of cross-view action understanding, positioning her work as essential reading for students and researchers tackling view-invariant learning in embodied AI.
Research Focus
Key Achievements
Top Papers
- 1
- 2