BLISS AI Speaker Series #35: Beyond Patches: Learning Dense Visual Features - T. Wimmer

Join us for an enlightening session featuring Thomas Wimmer, a PhD student at ETH Zürich, as he delves into his talk titled "Beyond Patches: Learning Dense Visual Features".

🕒 Duration: Approximately 45 minutes of insightful discussion followed by a networking opportunity!
🗓️ Connect with fellow AI enthusiasts, share ideas, and ask questions while enjoying FREE drinks and pizza!
🚪 Note: Doors close by 7:15 PM - arrive early to secure your spot!
✨ Important: RSVP on Luma to guarantee your entry! RSVP Here

🎯 Who Should Attend?

This event is open to anyone passionate about cutting-edge AI research. It's particularly tailored for students, PhD candidates, academic researchers, and industry professionals focusing on machine learning.

🔍 Abstract:

Modern vision foundation models excel in feature extraction for computer vision, yet their reliance on patch outputs can hinder performance in dense, pixel-wise tasks. This talk will cover innovative strategies for learning and optimizing dense visual features beyond traditional patches. Highlights include:

• DIY-SC: A framework utilizing pseudo-labels to enhance pretrained models for correspondence tasks while retaining generalizability.
• AnyUp: A groundbreaking universal feature upsampler that delivers state-of-the-art performance across various resolutions and tasks. Its unique aspect? It's agnostic to source features at inference time, greatly enhancing practicality.
• Additionally, recent applications of AnyUp will be discussed briefly.

👤 About the Speaker:

Thomas Wimmer is a doctoral researcher and PhD fellow at the Max Planck ETH Center for Learning Systems, under the guidance of notable advisors like Jan Eric Lenssen and Bernt Schiele. As a student researcher in the Semantic Perception team at Google Zurich, his expertise lies in visual representation learning and 3D computer vision. Thomas has made significant contributions to leading AI conferences, served as a reviewer for prominent journals, and received an outstanding reviewer token at ICCV '25.

We are BLISS e.V., the go-to AI organization in Berlin dedicated to connecting passionate individuals in the machine learning field.
This summer 2026, we will host another exciting speaker series featuring esteemed researchers from Cohere, ETH Zürich, University of Oxford, HuggingFace, and Stanford University!

🌐 Website: bliss.berlin
📽️ YouTube: bliss.ev.berlin

Disclaimer: By attending this event, you consent to being photographed.