How machines perceive, reconstruct, and render the 3D world: from classical geometry and image formation, through depth estimation, Structure-from-Motion and SLAM, to modern neural rendering with NeRF and 3D Gaussian Splatting.
This module follows a single thread: turning 2D observations into 3D structure you can reconstruct from real images and render back into new views. We begin with how images are formed and the classical ways of representing 3D geometry, move through depth estimation and multi-view reconstruction and state estimation (Structure-from-Motion and SLAM), and then into the neural rendering methods: volumetric fields, NeRF, and 3D Gaussian Splatting, that now define the field.
Teaching combines lectures with hands-on practicals that run inside the class hours, on a laptop, using Python and PyTorch with COLMAP and NeRFStudio. The module is assessed through a short individual quiz and a group project or group presentation. The last two sessions are seminars that look past the exam program, at where this field is heading. It is one module of the 12 CFU Elective in Artificial Intelligence.
A glimpse of what you build during the module.
The same office scene reconstructed two ways: a classical mesh and rendering (3D Gaussian Splatting). Drag the slider to wipe between them.
This module follows a single chain from end to end: how an image is formed, how a set of images becomes 3D geometry, and how that geometry is rendered back into new views. The program stops at rendering, which leaves each step the time it needs. By the end you should understand why each link in that chain is built the way it is.
Treat the side list as a comfort check. You do not need to arrive knowing everything: it is enough to recognize these ideas and not freeze when they appear on a slide. Everything past that we cover together, and we explain what we use as we use it.
Epipolar geometry, multi-view stereo, Structure-from-Motion and SLAM, volume rendering, NeRF, and Gaussian Splatting are all introduced from scratch during the module. No prior exposure is assumed.
You will not be examined on linear algebra. With thirteen sessions on one chain, we can go deep enough for you to build something and understand why it works. Several of these topics are entire courses, or entire PhDs, on their own. If one of them grabs you, the suggested readings in the schedule are the door in (this will be updated during the course), and we're happy to point you further at office hours.
Students taking this module are strongly encouraged to fill in the enrollment form. It is how we reach you about slides, practicals, room changes and anything else, so it makes communication with the teachers much easier.
Thirteen sessions over six weeks, twice a week: Monday 08:00–10:00 and Wednesday 08:00–11:00 in room B2. Lessons start on Wednesday 23 September and end on Wednesday 4 November 2026. Each row below is a block of the course, so the order of topics inside a block can shift as we go. Slides are shared through Google Classroom with enrolled students only. Readings are added under each topic during the lessons. SCHEDULE CAN BE SLIGHTLY CHANGED.
Assessment has two main parts: an individual quiz (13 questions, max 10 points), and either a group project or a group presentation (max 20 points). Both are done in groups of two or three students. Two small homeworks during the course are worth up to +3 points, and a further +3 go to outstanding projects and outstanding presentations.
This grade counts toward the overall mark of the 12 CFU Elective in Artificial Intelligence.
