Talk Abstract: The field of computer vision is in the midst of a paradigm shift, moving from task-specific models to “foundation models” – huge neural networks trained on massive amounts of unlabeled data. These models are capable of learning powerful and rich representations of our visual world, as evident for example by the groundbreaking results in text-to-image generation. Nevertheless, current foundation models are still largely treated as black boxes – we do not fully understand the priors they encode, how they are internally represented, which limits their usage for downstream tasks. In this talk, I’ll dive into the internal representations learned by prominent models and unveil new striking properties about the information they encode, using simple empirical analysis. I’ll demonstrate how to harness their power and unique properties through novel visual descriptors and perceptual losses, for a variety of visual tasks, including co-part segmentation, image correspondences, semantic appearance transfer and text-guided image and video editing. In all cases, the developed methodologies are lightweight, and require no additional training data other than the test example itself. 

Speaker Biography: Tali Dekel is an Assistant Professor at the Mathematics and Computer Science Department at the Weizmann Institute, Israel. She is also a Staff Research Scientist at Google, developing algorithms at the intersection of computer vision, computer graphics, and machine learning. Before Google, she was a Postdoctoral Associate at the Computer Science and Artificial Intelligence Lab (CSAIL) at MIT. Tali completed her Ph.D. studies at the school of electrical engineering, Tel-Aviv University, Israel. Her research interests include computational photography, image/video synthesis, geometry and 3D reconstruction. Her awards and honors include the National Postdoctoral Award for Advancing Women in Science (2014), the Rothschild Postdoctoral Fellowship (2015), the SAMSON – Prime Minister’s Researcher Recruitment Prize (2019), Best Paper Honorable Mention in CVPR 2019, and Best Paper Award (Marr Prize) in ICCV 2019. She often serves as program committee member and area chair of major vision and graphics conferences More information in: https://www.weizmann.ac.il/math/dekel/home

We hope you can join us in person, but if this is not possible, you can join online by registering in advance for this meeting.