< Back to previous page


Self-learning Algorithms for the Computation of Global Illumination in Computer Graphics

The global light transport problem is a fundamental problem in the field of computer graphics. Given the description of a 3D scene, a photorealistic image is computed. The current most advanced algorithms are all based on Monte Carlo integration. Recent developments in image synthesis use machine learning to reduce the calculations per pixel to form the image. These techniques are often aimed at reducing the stochastic variance, as well as the 2D domain of the image to be generated. However, the question arises whether such self-learning algorithms can also be applied directly to the 3D scene, by learning the 3D distribution of light energy rather than its projection onto the image plane. Such an approach could be camera-independent and depend only on the 3D scene itself, thus obtaining further computational advantages.

Date:3 Mar 2021 →  1 Sep 2022
Keywords:Computer graphics, Machine learning, Global illumination, Ray tracing
Disciplines:Computer graphics, Machine learning and decision making
Project type:PhD project