Beyond Patches: Learning Dense Visual Features | Thomas Wimmer, ETH Zurich | BLISS e.V.

We are excited to feature *Thomas Wimmer**, PhD Fellow at the Max Planck ETH Center for Learning Systems and ETH Zürich, discussing **"Beyond Patches: Learning Dense Visual Features"* and advanced strategies for optimizing pixel-wise representations in modern vision foundation models. 🎥 *WATCH TO DISCOVER:* 🔹 *The Patchification Problem:* Why modern vision foundation models face inherent limitations on dense, pixel-wise tasks due to their patchified outputs. 🔹 *The DIY-SC Framework:* How leveraging pseudo-labels can significantly improve pretrained models for correspondence tasks while fully preserving the original backbone's generalizability. 🔹 *AnyUp Upsampler:* A deep dive into a universal, source-agnostic feature upsampler that achieves state-of-the-art performance across diverse resolutions, domains, and downstream tasks. 🔹 *Agnostic Inference Engineering:* How to design feature upsamplers that operate independently of source features at inference time to dramatically increase utility in practical settings. 🔹 *Next-Gen Visual Representations:* Actionable insights into applying dense feature learning to real-world computer vision pipelines and 3D vision frameworks. This session is open to everyone interested in state-of-the-art AI research. Whether you are a student, PhD candidate, academic researcher, or industry professional with a focus on machine learning, computer vision, and dense visual representations, this talk offers deep technical insights from the front lines of visual representation learning. 📅 *VIDEO TIMESTAMPS:* 00:00 - Introduction & The Limits of Patchified Vision Models 08:30 - Overcoming Patch Constraints in Vision Transformers 14:15 - Deep Dive: The DIY-SC Framework & Pseudo-Labels 23:45 - Introducing AnyUp: The Universal Feature Upsampler 34:10 - Real-World Applications & Benchmark Results 42:00 - Summary & Key Takeaways 👥 *EVENT LOGISTICS & RSVP:* 🔹 *Schedule:* Please arrive early! Doors close strictly by **7:15 PM**. 🔹 *Networking:* Stay after the talk to connect with fellow AI enthusiasts over free pizza and drinks! 🔹 *Entry Requirement:* An RSVP ("Attend") on Meetup is *strictly necessary* to be guaranteed entry. 🔹 *Note on Meetup Plus:* You are not obligated to purchase Meetup Plus to attend; BLISS events and the platform remain 100% free. 📷 Disclaimer: By attending this event, you agree to be photographed for community highlights. ✨ *ABOUT THE SPEAKER:* 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, with work published at major AI venues including CVPR, ICCV, 3DV, and ICLR. He was awarded an outstanding reviewer token at ICCV '25 and has conducted research stays at TUM, Ecole Polytechnique, and Google. ✨ *ABOUT BLISS e.V.* This talk was recorded in Summer 2026 at TU Berlin as part of the BLISS AI Speaker Series. BLISS e.V. is the premier AI organization in Berlin, Germany. We connect machine learning engineers, researchers, and data science enthusiasts through high-impact technical events. Our community initiatives include: *Weekly ML Reading Groups:* In-depth, paper-by-paper discussions on the latest AI breakthroughs. *Tech Networking Events:* Connecting local Berlin talent with global AI industry leaders from top institutions like Cohere, ETH Zürich, Oxford, Hugging Face, and Stanford. 🔗 *JOIN THE COMMUNITY:* Follow our Meetup page to stay updated on upcoming machine learning workshops and speaker sessions! 👉 [https://www.meetup.com/bliss-speaker-...](https://www.meetup.com/bliss-speaker-...) 🌐 *EXPLORE MORE:* Website: [https://bliss.berlin](https://bliss.berlin) YouTube: [   / @bliss.ev.berlin  ](   / @bliss.ev.berlin  ) #MachineLearning #ComputerVision #VisualRepresentation #ETHZurich #AnyUp #AIResearch #TUBerlin #BLISSAI #3DComputerVision