Notes for the master course Hands-on AI based 3D Vision (Prof. Dr.-Ing. Gerard Pons-Moll, University of Tübingen / MPI for Informatics, Summer Semester 2026).

The course is about understanding and reconstructing the 3D world from images and video. It starts with the fundamentals of multiple view geometry and goes up to recent methods: NeRF, Gaussian Splatting, 3D diffusion models and token-based reconstruction.

Goal of the course

After this course, students should be able to:

  • understand research papers related to 3D vision and assess how they fit within the state of the art
  • develop classical and modern algorithms: camera parameter estimation, Structure-from-Motion, point cloud processing, neural rendering techniques, generative models

Lectures

#LectureKey concepts
1Introduction to 3D Computer Vision3D vision vs. computer graphics, applications, history of 3D vision
2.1Image Formationpinhole camera, lens camera, intrinsics, extrinsics, projection
2.2Rotationsrotation matrices, Euler angles, gimbal lock, quaternions, axis-angle, exponential map
3Classical Reconstruction Methods”inverting” image formation, epipolar geometry, essential matrix, 8-point algorithm, triangulation, Structure-from-Motion, bundle adjustment
4Stereo Vision and Depth Estimationepipolar geometry, rectification, disparity, stereo matching, multi-view stereo, monocular depth
5Surface Reconstructionimplicit vs. explicit geometry, SDF, marching cubes, point clouds, Poisson reconstruction, Procrustes, ICP
6.1Neural Fieldsvoxel grids, octrees, neural fields, triplanes, hash encoding, neural implicits
6.2Point Cloud ProcessingPointNet, permutation invariance, PointNet++, point transformers
7Neural Radiance Fields (NeRF)volume rendering, transmittance, positional encoding, NeRF variants
83D Gaussian Splatting3D Gaussians, covariance, splatting, alpha blending, adaptive density control
9Learning-Based 3D ReconstructionCOLMAP, bundle adjustment, DUSt3R, point maps, 3D foundation models
10Diffusion ModelsGAN, VAE, DDPM, DDIM, Stable Diffusion, multi-view diffusion
11Flow Matching and 3D Generationflow matching, score distillation sampling, point cloud diffusion, TRELLIS, Hunyuan3D

Lecture pages are linked here as soon as they are written.

Terms that appear in several lectures have their own short page under Concepts.

How to read these notes

Every lecture page has the same structure: a short TL;DR, the content in the order of the slides, a self-test and links to related lectures. The table of contents on the right follows the slide sections. Small labels like Slides 12-15 say where a topic is in the original slides.

Callouts mark what matters most:

Definition

The exact meaning of a term.

Key formula

Formulas you should know by heart.

Intuition

Why something works, in plain words.

Common mistake

Typical misunderstandings and wrong exam answers.

Exam relevance

What was asked in the exam or is likely to be asked.

Self-test question

Course organization

LecturerProf. Dr.-Ing. Gerard Pons-Moll
Teaching assistantsNiklas Berndt, Pradyumna YM, Yannan He
Workload6 ECTS, one lecture + one tutorial per week
Grading50% exam, 20% exercises, 30% project
Course pagehttps://virtualhumans.mpi-inf.mpg.de/3DVision26/
  • Exercises consist of theoretical questions and programming exercises, done in teams of 2 or 3 people.
  • Project: a mini research project (~6 weeks). It is graded on the report (motivation, prior work, methodology, evaluation, limitations), on how substantial the method is, on the robustness of the result and on the presentation.
  • To pass the course, you need to pass the exercises, the project and the exam.

Requirements: linear algebra, probability theory, machine learning, optimization and programming skills (Python). PyTorch and computer graphics background are a plus.