Modern Computer Vision Methods


This seminar course addresses recent methods in the domain of computer vision and their broader impact and application. A list of selected papers is proposed and will be studied and discussed. The advantages and drawback of modern computer vision methods are discussed and put in historic perspective in the field. Application domains include 2D and 3D object detection, image/video/spatial content segmentation, 3D reconstruction, multi-view imaging, sensor fusion, multi-modal imaging, vision for autonomous driving, camera and object pose estimation, SLAM, as well as deformable and rigid registration.


  • Presenter Matching finalized (see below)
  • Paper preference to be indicated until Tuesday, October 25, 2022 via this form
  • Sessions: Mondays, at 12.00 in room MI 03.13.010

Dates and Topics




15.07.2022 Preliminary Course Meeting Slides
17.10.2022 - no session due to Welcome@TUM event -  
24.10.2022 Introduction Session Intro Slides Projects
31.10.2022 Presentation Training Slides
07.11.2022 CVPR Break (Individual Preparation)  
14.11.2022 Slot to meet with Supervisor  
21.11.2022 Slot to meet with Supervisor  
28.11.2022 Object Poses + Robotic Grasping Gen6D TransGrasp
05.12.2022 Neural Fields: NeRF Enhancements + Semantics Depth Priors Panoptic Fields
12.12.2022 Invited Talk by Steven McDonagh, Huawei on Tackling modality, data distribution challenges for real world Computer Vision Google Scholar
19.12.2022 Polarization Sensing Polarization
09.01.2023 3D Reconstruction + Manipulation ShapeFormer 3D Recon CLIP-NeRF
16.01.2023 Scene Mapping & High-Level Understanding NICE-SLAM Commonsense
23.01.2023 Point Cloud Registration GeoTransformer Lepard
30.01.2023 Deformable Shapes KeyTR Neural Flow
06.02.2023 Invited Talk by Fabian Manhardt, Google on 3D Object Understanding Google Scholar

List of Topics and Material

Session Topic Supervisor Presenter
28.11. Gen6D Pengyuan Rabia
TransGrasp Guangyao Bengisu
05.12 Depth Priors Patrick Abhinav
Panoptic Fields Stefano Omar
19.12 Polarization Hyun Michael
09.01. ShapeFormer Lennart Ibrahim
3D Reconstruction Shun-Cheng Tarlan
  CLIP-NeRF Hannah Caghan
16.01 NICE-SLAM Patrick Weihang
Commonsense Localisation Shun-Cheng Haoran
23.01 GeoTransformer Hao Azad
Lepard Mert Florian
30.01 KeyTR Mahdi Christoph
Neural Flow Lennart Jim


For the application, please sent us your motivation (+ previous experience / CV / transcript in the field to increase your chances for matching) via e-mail to: (Application Deadline: July 26, 2022)

Please also register in the TUM matching system for the course. Keep in mind that your chances to be assigned to the course increase if you give it a higher rank in your choices. For further details about how the matching system works and its schedule please check this website.
We select appropriate candidates based on their background, interests, and motivation.