Neural Luminaire Light Field Compression

Speaker: David Köppl

Abstract

This master’s thesis investigates how measured near-field luminaire emission can be transformed into a compact representation suitable for real-time rendering. The goal is to model the emitted appearance of a luminaire as a surface-parameterized, direction-dependent radiance field and to compress it efficiently using neural methods. To this end, an end-to-end pipeline will be developed that generates training data from calibrated measurement images, learns a compact representation from these samples, and makes it deployable for GPU-based evaluation in a renderer. The approach will be evaluated on both synthetic test cases and real measured data.