Winter Semester 2025/2026, TU München

Organisers: Viktoria Ehm, Maolin Gao, Dr. Riccardo Marin, Thomas Dages

Contact : savha-ss25@vision.in.tum.de

• 14.7.2025 Preliminary meeting at 3PM; Room: 02.09.023, Seminarraum (5609.02.023); Zoom Link: https://tum-conf.zoom-x.de/j/65773324345?pwd=b99EYbOIts8UbGtbGrmVlFE1ZDbITM.1

Apply to the matching system, and send us an email, attaching your CV and transcripts (Bachelor+Master) and with the following structure:

In the body, please give at least the following details: Subject: Application [Your Matriculation Number]

Matriculation #:
Name:
Name of Degree:
Masters Semester #:
Average Grade:
○ Bachelor:
○ Master (For the previous semester, if available)
List of Relevant courses taken with grade

Share any additional documents, information (eg. link to git, past research projects) that could support your application.

Geometry surrounds our lives, and we recognize ourselves as part of the 3D world. While we intuitively interact with it, developing computer algorithms to analyze, manipulate, and generate such structured data is far from trivial and a vivid research area. The development of this field and recent deep learning development opened to dramatic opportunities like the simulation of surgeons, videogames creations, assisted architecture design, or procedural generation of limitless universes. It should also be no surprise that, among the many possible classes of objects, 3D Virtual Humans are of particular interest. Acquiring and generating avatars in different identities and poses, clothes, and maybe in interaction with physical objects has caused a dramatic explosion of interest, both in academia and industry. In this course, we aim to motivate our exploration by a fundamental question: considering two 3D objects, how are they similar? How are they different? Or, namely, is it possible to find correspondence between them?

Students will work on specific projects on topics related to:

• Shape matching
• Point-Cloud registration
• Geometric Deep Learning
• Virtual Humans reconstruction and animation
• 3D Human-Object interaction
• Generative AI for 3D

Previous knowledge expected It is expected that all participants have basic programming skills in a scientific programming language (eg Python). Having experience or knowledge of the following is also a plus: continuous/discrete optimization, 3D geometry, computer vision, computer graphics, and Deep learning libraries (e.g., PyTorch).

Expected results of study Upon completion of this module, students will acquire a deep understanding of correspondence problems, recent trends and tools to work with 3D virtual humans, and they will gain practical experience in solving challenging problems. Furthermore, they will learn how to work in small teams and to present their results.

Teaching and learning method At the beginning of the course, the basics of 3D shape analysis, shape matching, and virtual humans will be covered, both theoretically and practically. In the second part of the work, students will work on projects in groups of 2-3. Project will run with regular communication with tutors and mid-term presentations.

Recommended reading:

• Bronstein et al. Numerical Geometry of Non-Rigid Shapes, 2009
• Numerical Optimization, 2006 - Solomon.
• SMPL made Simple (https://smpl-made-simple.is.tue.mpg.de/)
• Further relevant references will be provided during the course.