$\mu$NCA: Texture Generation with Ultra-Compact Neural Cellular Automata

TL;DR

This paper introduces $dNCA, a highly compact neural cellular automata model for procedural texture synthesis, capable of generating complex textures with only a few hundred parameters, enabling efficient implementation.

Contribution

The paper presents a novel, ultra-compact neural cellular automata model for texture generation, demonstrating that complex textures can be represented with significantly fewer parameters than traditional neural networks.

Findings

Models as small as 68 parameters can generate textures.

Quantized models can be as small as 68 bytes.

Texture generators can be implemented with minimal code.

Abstract

We study the problem of example-based procedural texture synthesis using highly compact models. Given a sample image, we use differentiable programming to train a generative process, parameterised by a recurrent Neural Cellular Automata (NCA) rule. Contrary to the common belief that neural networks should be significantly over-parameterised, we demonstrate that our model architecture and training procedure allows for representing complex texture patterns using just a few hundred learned parameters, making their expressivity comparable to hand-engineered procedural texture generating programs. The smallest models from the proposed $\mu$NCA family scale down to 68 parameters. When using quantisation to one byte per parameter, proposed models can be shrunk to a size range between 588 and 68 bytes. Implementation of a texture generator that uses these parameters to produce images is possible with just a few lines of GLSL or C code.