Comparison of approaches to neural model execution in computer graphics

Comparison of approaches to neural model execution in computer graphics

Milin N.A. (KIAM RAS, Moscow, Russia)
Rodionov R.O. (KIAM RAS, Moscow, Russia)
Frolov V.A. (KIAM RAS, Moscow, Russia)

Abstract

This work provides a comparative performance study for different approaches of neural-based methods in computer graphics. Two directions are considered: representation of surfaces via neural models of the signed distance function (using SIREN as an example) and neural models of materials represented by a bidirectional reflectance function (using NBRDF as an example). This work includes analysis of both cross-platform and hardware accelerated MLP implementations. Experimental results reveal significant performance differences between investigated approaches. Based on that, the paper formulates practical recommendations for developers who plan to integrate neural methods in computer graphics applications.

Keywords

computer graphics; real-time rendering; neural networks; signed distance functions.

Edition

Proceedings of the Institute for System Programming, vol. 38, issue 3, part 3, 2026, pp. 39-48

ISSN 2220-6426 (Online), ISSN 2079-8156 (Print).

DOI: 10.15514/ISPRAS-2026-38(3)-34

For citation

Milin N.A., Rodionov R.O., Frolov V.A. Comparison of approaches to neural model execution in computer graphics. Proceedings of the Institute for System Programming, vol. 38, issue 3, part 3, 2026, pp. 39-48 DOI: 10.15514/ISPRAS-2026-38(3)-34.

Full text of the paper in pdf (in Russian) Back to the contents of the volume