Neural BRDF Representation and Importance Sampling

Abstract

Controlled capture of real-world material appearance yields tabulated sets of highly realistic reflectance data. In practice, however, its high memory footprint requires compressing into a representation that can be used efficiently in rendering while remaining faithful to the original. Previous works in appearance encoding often prioritized one of these requirements at the expense of the other, by either applying high-fidelity array compression strategies not suited for efficient queries during rendering, or by fitting a compact analytic model that lacks expressiveness. We present a compact neural network-based representation of BRDF data that combines high-accuracy reconstruction with efficient practical rendering via built-in interpolation of reflectance. We encode BRDFs as lightweight networks, and propose a training scheme with adaptive angular sampling, critical for the accurate reconstruction of specular highlights. Additionally, we propose a novel approach to make our representation amenable to importance sampling: rather than inverting the trained networks, we learn to encode them in a more compact embedding that can be mapped to parameters of an analytic BRDF for which importance sampling is known. We evaluate encoding results on isotropic and anisotropic BRDFs from multiple real-world datasets, and importance sampling performance for isotropic BRDFs mapped to two different analytic models.

Citation Style:    Display Alpha ACM SIGGRAPH IEEEtran Publication

Neural BRDF Representation and Importance Sampling. Alejandro Sztrajman, Gilles Rainer, Tobias Ritschel, Tim Weyrich. Computer Graphics Forum, 40(6), pp. 332–346, September 2021.Alejandro Sztrajman, Gilles Rainer, Tobias Ritschel, and Tim Weyrich. Neural BRDF representation and importance sampling. Computer Graphics Forum, 40(6):332–346, September 2021.Sztrajman, A., Rainer, G., Ritschel, T., and Weyrich, T. 2021. Neural BRDF representation and importance sampling.Computer Graphics Forum 40, 6 (Sept.), 332–346.A. Sztrajman, G. Rainer, T. Ritschel, and T. Weyrich, “Neural BRDF representation and importance sampling,” Computer Graphics Forum, vol. 40, no. 6, pp. 332–346, Sep. 2021. [Online]. Available: https://onlinelibrary.wiley.com/doi/abs/10.1111/cgf.14335 PDF (16 MB) Low-res PDF (3.0 MB) Suppl. MERL (55 MB) Slides (5.9 MB PDF) BibTeX DOI

WebGL Demo

For a real-time demo inside your web browser, see Alejandro Sztrajman’s personal paper page. That WebGL widget uses our neural encodings (NBRDF) to render a scene with measured materials from the MERL database.

Code and Pre-trained Models Data

We provide pre-trained NBRDFs for measured materials from multiple databases.

• code.zip (3.0 MB) Python code to train NBRDF networks, and Mitsuba plugin to render using pre-trained NBRDFs.
• merl_nbrdf.zip (0.45 MB) All materials from the MERL database.
• rgl-isotropic_nbrdf.zip (0.23 MB) Isotropic materials from the RGL database.
• nielsen_nbrdf.zip (0.04 MB) Materials from Nielsen et al. (2015).
• pytorch_code.zip (6 KB) NBRDF training script (alternative implementation in PyTorch by Michael Fischer).

Acknowledgments

This work was supported in part by the Marie Skłodowska-Curie grant agreement No. 642841 (DISTRO).