Sebastian Merzbach

In this project we take image-based reflectance measurements, which are subsequently processed into high quality spatially varying BRDFs (SVBRDFs). We develop alternative approaches to the SVBRDF fitting problem with the help of deep learning.

To correctly simulate materials under arbitrary illumination, the light simulation in a virtual scene must be calculated on a pure spectral basis. This is already done in modern rendering systems. For a few classes of materials spectral reflectance data is already acquired for a few light and view directions using spectrometers and gonioreflectometer setups. This is sometimes enough to fit analytical models to the measured data. But for anisotropic materials or for materials with strong variations in angular or spatial domain there are currently no measurement setups at hand. Similar setups like the ones based on RGB CCD cameras are impractical for spectral measurements because of the high costs of cameras and light sources needed for spectral measurements. In this project we plan to combine RGB and spectral measurement methods to come up with an efficient and pratical measurement setup for spectral BTFs. Furthermore, algorithm for analysis, compression and efficient rendering for such RGB-spectral-combined data will be investigated.