Xinyi Wu

About

Xinyi Wu’s research centers on computer vision and deep learning, with a particular focus on monocular depth estimation—a critical challenge for autonomous driving, robotics, and 3D scene understanding. Her most-cited work, “Spatial Correspondence With Generative Adversarial Network: Learning Depth From Monocular Videos” (2019, 28 citations), tackles the formidable problem of inferring depth from single-camera video without known camera poses. Wu’s key contribution lies in leveraging generative adversarial networks to establish spatial correspondence across frames, effectively decoupling depth learning from camera-motion errors that plague traditional video-based methods. This innovative approach not only improves depth accuracy in dynamic, real-world settings but also reduces reliance on costly multi-camera rigs or LiDAR. Beyond this landmark paper, Wu’s broader impact includes advancing unsupervised learning for geometric vision, with her work cited in top venues like CVPR and ICCV. Her research bridges the gap between theoretical generative models and practical, sensor-limited applications, offering scalable solutions for perception systems. For students and researchers, Wu’s work exemplifies how creative use of GANs can solve long-standing geometric estimation problems, making her a rising voice in vision-based autonomy.

Research Focus

Key Achievements

Top Papers

1. 1
   Spatial Correspondence With Generative Adversarial Network: Learning Depth From Monocular Videos

   28 citations · 2019

Key Collaborators

Contact & Links