- Title
- Seminar [09/04] Learning Weakly-supervised Semantic Correspondence and Its Applications
- Date
- 2019.08.30
- Writer
- 전기전자공학부
- 게시글 내용
-
< BK21+ BEST Seminar Series Announcement>
Time and Date : 16:00 ~ 17:00 Wednesday 09/04/2019
Place : C716, Engineering Building #3
Title : Learning Weakly-supervised Semantic Correspondence and Its Applications
Abstract:
Establishing dense semantic correspondences across semantically similar images is essential for numerous computer vision, machine learning, and computation photography applications. Unlike traditional dense correspondence for estimating depth or optical flow, semantic correspondence poses additional challenges due to intra-class appearance and shape variations among different instances within the same object category. In addition, the lack of an appropriate benchmark with dense ground-truth correspondences make supervised learning less feasible for this task. In this talk, we first investigate state-of-the-art research on semantic correspondence that leverages deep convolutional neural networks (CNNs) in a weakly supervised fashion, called recurrent transformer networks (RTNs). We then investigate interesting its applications such as photorealistic style transfer and landmark detection that can be jointly solved with semantic correspondence. Finally, other potential applications and further directions will be presented.Presenter: Kim Seung Ryong & Post-Doctoral Researcher / École Polytechnique Fédérale de Lausanne (EPFL)
Host: Prof. Sohn, Kwanghoon, Yonsei EEE