Winter Semester 2025/26, TU München

Organisers: Shenhan Qian, Linus Härenstam-Nielsen, Ganlin Zhang, Weirong Chen

This seminar examines the rise of foundation models for 3D vision, focusing on their architectures, training paradigms, and downstream applications. Core topics include models like DUST3R and VGGT, as well as related approaches that aim for generalizable 3D understanding across tasks such as matching, pose estimation, and reconstruction. Through readings and projects, students will critically analyze the strengths and limitations of these models and explore their potential as building blocks for next-generation 3D perception systems.

The seminar is conducted in English. Dates and schedule of the seminar can be found in TUMOnline.

The seminar will be arranged as a 1-hour weekly course (Wednesday 1pm-2pm), where each student will present a paper assigned after matching.

The preliminary meeting will take place on Zoom, on Tuesday 8th July at 2:00 pm: zoom link. The slides of the preliminary meeting can be found here as slides. If you have any questions do not hesitate to contact us.

Please read the preliminary meeting slides and fill in the google form for application. Applicants who did not fill the form will be ignored.