# Fitting a neural texture decoder with ES (no backprop)

> Source: <https://richg42.blogspot.com/2026/09/fitting-neural-texture-decoder-with-es.html>
> Published: 2026-09-03 07:06:41+00:00

This is something I prototyped for fun in an evening, and now the idea is Prior Art:

Toy neural texture compression, trained entirely with 

[Evolution Strategies](https://en.wikipedia.org/wiki/Evolution_strategy)
 (no backprop, no derivatives). A 128×128×8 latent texture + a 1035-weight MLP decoder reconstruct a 512×512 image (kodim23 crop) at 32.2 dB PSNR. Latent quantizes to 8 bits with only 0.04 dB loss: ~4 bpp raw, ~3.3 bpp entropy coded. Trained in 150 s on CPU. Single C++ file, ~600 lines, stb_image for I/O.