About this Event
5200 N Lake Rd, Merced, CA 95343
Electrical Engineering and Computer Science (EECS)
Ph.D. Dissertation Defense
"Learning Visual Correspondence across Instances and Video Frames"
Electrical Engineering and Computer Science
University of California, Merced
Visual correspondence learning aims to link different visual elements such as pixels or image patches across images. It is an important element in various vision tasks such as 3D reconstruction or video segmentation. However, acquiring ground truth correspondence annotations in images or video frames requires large amounts of human effort, while synthetic data often presents a large domain gap compared to natural images. In this dissertation, I will introduce our work on learning visual correspondences in a self-supervised manner by exploring the massive redundant information in videos. I will also demonstrate its applications in video label propagation and video mesh reconstruction. Lastly, I will show how to utilize learned correspondences across instances to facilitate single-view reconstruction with only foreground masks as supervision.
Xueting Li is a Ph.D. candidate working with Professor Ming-Hsuan Yang at the Electrical Engineering and Computer Science, University of California, Merced. She received her B.Sc. and M.Sc. from Beijing University of Posts and Telecommunications and Tsinghua University in 2013 and 2016. Xueting’s research interest includes self-supervised learning, low-level vision, and 3D vision.
1 person is interested in this event