BLISS AI Speaker Series #35: "Beyond Patches: Learning Dense Visual Features"

BLISS is excited to feature Thomas Wimmer, who is currently PhD student at ETH Zürich and will discuss "Beyond Patches: Learning Dense Visual Features", lasting approximately 45 minutes. After the talk, seize the opportunity to connect with fellow AI enthusiasts to share ideas and questions while enjoying free drinks and pizza. Door close by 7.15pm, so please come early! Also, "attend"ing (RSVP) here on Meetup is strictly necessary to be guaranteed entry.
Please note that Meetup has recently been quite keen on promoting its Plus program. However, you are not obligated to purchase it, as both our events and the platform remain free.

Who is this event for?
This event is open to everyone interested in state-of-the-art AI research. We especially design it for students, PhD candidates, academic researchers, and industry professionals with a research focus in machine learning.

Abstract: Modern vision foundation models are highly capable and used as feature extractors in virtually any modern computer vision project. However, their patchified outputs inherently limit performance on dense, pixel-wise tasks. This talk presents strategies to learn and optimize dense visual features beyond these standard patches. We first introduce DIY-SC, a framework that leverages pseudo-labels to significantly improve pretrained models for correspondence tasks, while maintaining the original backbone's generalizability. We then focus on AnyUp, a universal feature upsampler that achieves state-of-the-art upsampling performance across diverse resolutions, domains, and downstream tasks. AnyUp is the first upsampler that is agnostic to the source features at inference time, significantly increasing its utility in practical settings. Finally, the talk briefly discusses recent applications of AnyUp.

Bio: Thomas Wimmer is a doctoral researcher and PhD fellow of the Max Planck ETH Center for Learning Systems, advised by Jan Eric Lenssen, Bernt Schiele (Max Planck Institute for Informatics), and Siyu Tang (ETH Zurich). He is currently a student researcher in the Semantic Perception team at Google Zurich. His research focuses on visual representation learning and 3D computer vision, and his work has been published at major AI conferences including CVPR, ICCV, 3DV, and ICLR. He has served as a reviewer for multiple A* conferences and journals, was awarded an outstanding reviewer token at ICCV '25, and has conducted several research stays during his studies, working with Daniel Cremers (TUM), Maks Ovsjanikov (Ecole Polytechnique), and Federico Tombari (Google/TUM).