← All resources

Adventures in Neural Rendering

Auto-discovered

Blog Advanced

Kostas Anagnostou

Visit resource ↗

Description

Adventures in Neural Rendering is a February 2026 blog post by Kostas Anagnostou on Interplay of Light about using small multilayer perceptrons as signal encoders in real-time rendering.

It summarizes MLP structure, LeakyReLU activation, weight storage, and GPU forward plus Adam training. Experiments compare MLPs to L2 spherical harmonics for cubemap radiance and irradiance, try depth and RTAO caches, and fit Cook-Torrance BRDFs including Rusinkiewicz reparameterization. Results stress quality versus storage and high inference cost for larger nets.

It is aimed at intermediate to advanced rendering engineers who know shading and GPU buffers but may be new to neural methods.

The article is a free HTML post with diagrams, shader-style snippets, and visual comparisons rather than a full ML tutorial.

It is useful for judging when tiny MLPs help or hurt versus classical encodings before larger neural rendering work.