Winter Semester 2022, TU München

Organisers: Maolin Gao, Tarun Yenamandra

Please direct questions to 3dsga-ws22@vision.in.tum.de

2022-09-09: The topic assignment and seminar schedule are out, please contact your supervisor to discuss your paper ~4 weeks prior to your presentation date.

2022-08-24: The list of papers has been published here (password protected), please send us your top 4 preference by 07.09.2022 EOD .

2022-07-21: The slide of the pre-meeting has been published here (password protected).

2022-06-29: [preliminary meeting] will take place at 10:00 - 11:00 on 21.07.2022 online via zoom. The slides will be published afterwards. You are encouraged to participate and ask questions directly in the meeting, since it will positively affect the matching process. Please find the zoom link in TUMonline.

Recently advances in computer vision and graphics underline the importance of 3D generative models and shape analysis. A generative model, as the name suggests, generates data from latent parameters. 3D generative models generate 3D data from a learned distribution of 3D shapes. The distribution is often learned from a 3D dataset. This area of research is now highly active due to the explosion of research in neural generative models such as variational auto-encoders, and generative adversarial networks. On the other hand, shape analysis alone is already an interesting research topic, which has many relevant applications, such as animation and reasoning. Recent studies also show a great potential, that combining the synergy of 3D generative model and shape analysis is beneficial for both tasks.

In this seminar, we will review the classic and recent advances in 3D generate models and shape analysis, both optimization-based and machine-learning-based approaches. Students will read a list of selected research papers, and each student will deeply study the problem setting and methods described in one existing paper under our supervision, and report the final outcome in terms of open presentations followed with a Q&A session and reports. There will be no additional written or oral exam.

After attending the seminar, we expect the participants should have a good overview of current approaches to tackle 3D shape generation and analysis problems, and a deeper understanding about one particular method, which should lay a good foundation of future hands-on research projects.

All participants should have a solid working knowledge of linear algebra and calculus. In addition, it is useful (but not required), that students have a background in one of the following topics: continuous/discrete optimisation, 3D geometry, computer vision, image processing, computer graphics or deep learning.