Speak: Prof. Tinne Tuytelaars
Time: 10:00-11:00, Nov. 06
Host: Prof. Jingyi Yu
Image-to-image translation has recently received significant attention due to advances in deep learning. Apart from some fun applications like transforming a photograph into a Van Gogh painting, there's also various practical applications like transforming synthetic images into more photorealistic ones, which can help to generate training data for various vision tasks.
In this talk, I will explain two contributions we made in this context:
i) learning to perform many-to-many mappings without the need for paired training data,
ii) learning to transform to different shapes (rather than just changing the appearance).
Tinne Tuytelaars is professor at the Center for Processing Speech and Images at KU Leuven, Belgium. Her research focuses on computer vision, and in particular image representations, object recognition, incremental learning and multi-modal analysis (images and text). She obtained a master in Electrical Engineering from KU Leuven in 1996, followed by a PhD degree in 2000. Since then, she has been mostly affiliated with KU Leuven. In 2009, she was awarded a prestigious ERC Starting Independent Researcher Grant. At ECCV2016, she received the Koenderink Award for the SURF interest point detector(“for fundamental contributions in computer vision that stood the test of time”). She has been area editor and then associate-editor-in-chief of IEEE Transactions of Pattern Analysis and Machine Intelligence from 2014-2018, area editor of Computer Vision and Image Understanding from 2009-2018, and area editor of the International Journal on Computer Vision since 2018. She was one of the program chairs of ECCV2014, one of the general chairs of CVPR2016, and will be one of the program chairs of CVPR2021.
Sist seminar 18215