392155 Generative Models for Visual Computing (S) (SoSe 2025)
Contents, comment
-
Generative models have revolutionized computer vision and computer graphics, enabling new applications such as generating images and 3D models from a simple text description. The foundation for this are the topics covered in this course:
- variational inference and the evidence lower bound (recap)
- the relation between score matching and denoising diffusion models
- the use of iterative sampling and differential equations for inference
- denoising implicit models and their inversion
- unconditional and conditional models using classifier-based and classifier-free guidance
- applications in image generation, video generation, editing, and 3D modeling
Central will be the mathematical modeling behind the most recent diffusion models.
The course is based on courses given at KAIST:
https://mhsung.github.io/kaist-cs492d-fall-2024/List of relevant papers (a subset will be presented):
https://docs.google.com/spreadsheets/d/1j7amDru9bRQsQgp2pfm1a8GrZ6K0HWwCDORGq-sj7dQ/edit?gid=0#gid=0The following slides provide additional details and context to the existing visual computing modules at Bielefeld University given by the Visual AI group:
https://docs.google.com/presentation/d/e/2PACX-1vRGyE0j_cU324qp-vOEE4hEBPn3HdwliZQqeg_vT0w7Yt9EUXmgFB6NamL_ynqeBMYRgD-cdvfvXobA/pub?start=false&loop=false&delayms=3000
Requirements for participation, required level
-
Required skills: Solid theoretical machine learning knowledge (introductory + advanced course), good python programming skills (>1 year), and fundamental mathematics (linear algebra).
Optional skills: PyTorch experience and knowledge in solving differential equations and computer vision (lectures or projects) helps but is not required.These skills are typically acquired through lectures or by private projects, hackathons, and coding competitions.
Bibliography
-
SIGGRAPH 2024 Course: Diffusion Models for Visual Content Generation: https://geometry.cs.ucl.ac.uk/courses/diffusion4ContentCreation_sigg24/
CVPR 2023 Tutorial: Denoising Diffusion Models: A Generative Learning Big Bang: https://cvpr2023-tutorial-diffusion-models.github.io/
External comments page
Teaching staff
Dates ( Calendar view )
| Frequency | Weekday | Time | Format / Place | Period |
|---|
Subject assignments
| Module | Course | Requirements | |
|---|---|---|---|
| 31-M-ASM2 Advanced Statistical Methods II | Veranstaltungen aus dem Bereich Statistik und/oder in (einem) methodisch verbundenen Gebiet(en) (I.) | Graded examination
|
Student information |
| Veranstaltungen aus dem Bereich Statistik und/oder in (einem) methodisch verbundenen Gebiet(en) (II.) | Graded examination
|
Student information | |
| 39-M-Inf-AI-app Applied Artificial Intelligence | Applied Artificial Intelligence: Seminar | - | Student information |
| - | Graded examination | Student information | |
| 39-M-Inf-AI-app-foc Applied Artificial Intelligence (focus) | Applied Artificial Intelligence (focus): Seminar | - | Student information |
| - | Ungraded examination | Student information | |
| 39-M-Inf-INT-app Applied Interaction Technology | Applied Interaction Technology: Seminar | - | Student information |
| - | Graded examination | Student information | |
| 39-M-Inf-INT-app-foc Applied Interaction Technology (focus) | Applied Interaction Technology (focus): Seminar | - | Student information |
| - | Ungraded examination | Student information |
The binding module descriptions contain further information, including specifications on the "types of assignments" students need to complete. In cases where a module description mentions more than one kind of assignment, the respective member of the teaching staff will decide which task(s) they assign the students.
The focus of this seminar is on reading, presenting, and discussing recent papers on generative models and diffusion models in particular.
Lectures during the first half of the term prepare students by giving a deep dive into diffusion models, in more detail than covered in existing courses. It requires basic knowledge of machine learning, statistics, and math, including topics such as gradient descent, log-likelihood, distributions, and differentiation. These are the basis for the advanced topics discussed, such as, score matching and inference by solving differential equations.
Students present one paper, lead the discussion of one paper, and engage in the discussion of the papers presented by others, including the writing style, strengths, limitations, and ethical implications.
A report and small assignments written in PyTorch support the practical understanding of probabilistic models, as typical for seminars+tutorial.
- Moodle Courses
- Moodle Courses
This course has a video conference. Details will be displayed to you as a participant of this course. For an event with participant management, you must also be registered as attending by the teaching staff.
- Address:
- SS2025_392155@ekvv.uni-bielefeld.de
- This address can be used by teaching staff, their secretary's offices as well as the individuals in charge of course data maintenance to send emails to the course participants. IMPORTANT: All sent emails must be activated. Wait for the activation email and follow the instructions given there.
- If the reference number is used for several courses in the course of the semester, use the following alternative address to reach the participants of exactly this: VST_515738727@ekvv.uni-bielefeld.de
- Notes:
- Additional notes on the electronic mailing lists
- Last update basic details/teaching staff:
- Thursday, April 10, 2025
- Last update times:
- Wednesday, February 5, 2025
- Last update rooms:
- Wednesday, February 5, 2025
- Type(s) / SWS (hours per week per semester)
- seminar (S) / 2
- Language
- This lecture is taught in english
- Department
- Faculty of Technology
- Questions or corrections?
- Questions or correction requests for this course?
- Planning support
- Clashing dates for this course
- Links to this course
- If you want to set links to this course page, please use one of the following links. Do not use the link shown in your browser!
- The following link includes the course ID and is always unique:
- https://ekvv.uni-bielefeld.de/kvv_publ/publ/vd?id=515738727
- Send page to mobile
-
Click to open QR code
Scan QR code:
- ID
- 515738727