Seminar Computergraphik WS'25/26
Seminar

Topic: Current Research in Computer Graphics

Content

In the Computer Graphics Seminar we discuss current research results in the field of Computer Graphics. The tasks of the participants are to write up a research report, to review the work of another student in writing, and to later revise and improve their own report to reflect the input gathered from the review. Finally, at the end of the semester and during a block-seminar, each student will give an oral presentation on their respective research reports. This must also be rehearsed beforehand in front of the assigned individual supervisor and their suggestions for improvement must be integrated.

Participants

The course is aimed at Bachelor's and Master's students in Computer Science (Informatik), IST, and Business Informatics (Wirtschaftsinformatik), as well as students pursuing their Master in Data Science.

The registration takes place centrally via StudIP. The number of participants is limited to 5 students.

Important Dates

All dates listed here must be adhered to. In order to successfully complete the module, it is mandatory to fullfill the deadlines, and attend ALL events in person.

• From 04.08.2025 to 08.08.2025: Registration process via Stud.IP
  
• Until 05.10.2025: Submission of topic requests
• 22.10.2025, 15:00: Kick-Off Meeting (G30, ICG)  [Slides]
• 03.11.2025: End of the deregistration period
• 18.11.2025: Submission of the written paper
• 02.12.2025: Submission of the review report
• 16.12.2025: Submission of the revised paper
• Until 16.01.2026: Trial presentation
• 22.01.2026: Submission of the presentation slides
• 23.01.2026, 09:00 - 11:00: Presentations - Block Event
• 30.01.2026, 09:00 - 11:00: Presentations - Block Event Part 2

Registered students have the possibility to deregister until 2 weeks after the start of the lectures (i.e., 03.11.25) at the latest. For a successful deregistration, it is necessary to deregister via e-mail with the seminar supervisor (seminar@cg.cs.tu-bs.de).

Registered students, and students on the waiting list, have the possibility to send their top 3 topic requests in order of preference via email to seminar@cg.cs.tu-bs.de until the 05.10.25, so that they will be considered for the topic assignment.

Once a topic has been assigned to the student, all consequent submissions have to be sent by mail to the respective advisor and additionally to seminar@cg.cs.tu-bs.de. If not communicated otherwise, the deadline for all submissions is at 23:59 on the due day.

The respective drop-offs are done by email to seminar@cg.cs.tu-bs.de and if necessary by email to the respective advisor.

If you have any questions about the course, please contact seminar@cg.cs.tu-bs.de.

Format

• The final assignation of topics will be communicated during the Kick-Off event.
• For each topic, the student needs to prepare a report in latex using the ICG Template.
  The content of the report is a short summary of the work in one's own words and the elaboration of the main points, with a minimum length of 8 pages. The report should clearly reflect that the topic has been understood and was critically assessed.
• Each participant will later write a 1-2 page review on the report of another student (assigned by the seminar's supervisor). For writing the review one should pay particular attention to the comprehensibility and linguistic style of the summary.
• After receiving the review on one's own paper, the student will need to revise and improve their manuscript according to the received feedback.
• For the final presentations, the students can either use their own laptops or one provided by the Institute. If the student needs to use the ICG laptop, they need to contact seminar@cg.cs.tu-bs.de in time, at least two weeks before the presentations.
• The topics will be presented in approximately 20 minute presentations followed by a discussion.
• The language for the presentations can be either German or English.
• The oral presentation, the written paper, and the preparation of the review report, are all mandatory requirements to pass the course successfully.
  

Files and Templates

• Latex-Template Its use is compulsory.
• Slides-Template Using this template is recommended but not mandatory.
• Review-Template (mandatory usage)
• Kick-Off Slides

Tips on How to Give a Talk:

•    Giving a research talk. Short pdf with recommendations by Fredo Durand (MIT CSAIL)

•    How to give a talk (that doesn't put your audience to sleep). Slides by  Ramesh Raskar (MIT Media Lab)

Topics

1. SpotLessSplats: Ignoring Distractors in 3D Gaussian Splatting
   Sara Sabour, Lily Goli, George Kopanas, Mark Matthews, Dmitry Lagun, Leonidas Guibas, Alec Jacobson, David J. Fleet, Andrea Tagliasacchi
   ACM Transactions on Graphics (TOG) 2025
   Advisor: Florian Hahlbohm
   
   

   This paper improves 3D reconstruction by making it robust to real-world scenes with moving objects and inconsistent lighting, which typically violate assumptions of existing methods. By integrating Stable Diffusion features and robust optimization, it enables high-quality results from casual, uncontrolled captures.
   
   
   

2. On-the-fly Reconstruction for Large-Scale Novel View Synthesis from Unposed Images
   Andreas Meuleman, Ishaan Shah, Alexandre Lanvin, Bernhard Kerbl, and George Drettakis
   SIGGRAPH 2025
   Advisor: Junbo Li
   
   3D Gaussian Splatting (3DGS) is a well-known radiance field approach that allow fast scene reconstruction from images, thereby enabling free-viewpoint navigation. This paper proposed a novel incremental method that facilitates the instant production of camera poses and a trained 3DGS model following capture. Nevertheless, it overcomes the constraints of SLAM-based methods and is capable of managing dense and wide-baseline captures of ordered photo sequences and large-scale scenes.
   
   

3. GenFusion: Closing the Loop between Reconstruction and Generation via Videos
   Sibo Wu, Congrong Xu, Binbin Huang, Andreas Geiger, Anpei Chen
   CVPR 2025
   Advisor: Timon Scholz
   
   GenFusion is a method for improving the visual fidelity of 3D scene reconstructions by augmenting sparse-view scene captures with images generated using a diffusion model. This diffusion model is conditioned on the 3D reconstruction itself, enabling multi-view consistent generation of new viewpoints. During optimization, this cyclically improves both the main reconstruction and the quality of diffusion-generated viewpoints used for augmenting the reconstruction.
   
   
   
4. RenderFormer: Transformer-based Neural Rendering of Triangle Meshes with Global Illumination
   Chong Zeng, Yue Dong, Pieter Peers, Hongzhi Wu, and Xin Tong
   SIGGRAPH 2025 (To Appear)
   Advisor: Felix Lehner
   
   This paper introduces a Transformer architecture and training pipeline that enable a Transformer model to predict global illumination — and the associated light transport — on textured triangular geometry. In a second stage, the Transformer network also performs image synthesis as seen from a virtual camera.
   
   
   
5. PoseGuru: Landmarks for Explainable Pose Correction using Exemplar-Guided Algorithmic Recourse
   B. Dittakavi, B. Callepalli, S. Maheshwari, and V. N. Balasubramanian
   CVPR 2025
   Advisor: JP Tauscher
   
   PoseGuru is a system designed to identify and correct misaligned body parts in an individual's pose. It's especially useful in fields like fitness, rehabilitation, and sports, where proper form is crucial. Unlike systems that just tell if a pose is "good" or "bad," PoseGuru provides explainable, personalised, and fine-grained corrections to help users improve their movements.
   
   

Advisors

Florian Hahlbohm

Junbo Li

Timon Scholz

Felix Lehner

JP Tauscher

Susana Castillo