Neural block texture compression with CUDA

A developer built a neural block texture compression system that runs entirely on the GPU using CUDA 13.1, combining per-pixel selectors, per-4×4-block latents, and a tiny 1,767-weight MLP decoder trained with evolution strategies rather than backpropagation. The method reaches 40.78 dB in fp32 and 40.71 dB with the block latent quantized to 8 bits, at 6.03 bpp raw and 5.54 bpp entropy-coded, completing 3,000 iterations in 55 seconds on an RTX 5090.

The 2 latent layers containing the per-block parameters are sampled with nearest sampling. - Format: per-pixel selectors + one latent per 4×4 block + tiny MLP block decoder - Selectors: 1024×1024×2, 3 bits + 1 bit per pixel, picked by exact exhaustive search each iteration - Block latent: 256×256×4, 8 bits/value, trained by ES with footprint attribution - Decoder: 8→36→36→3 MLP, 1767 weights; inputs = 2 selectors + 4 block values + 2 block-local coords - Decoder training: antithetic ES 64 pairs, 4096-pixel minibatches then central finite differences for the last quarter - Latent training: 4 antithetic pairs/iteration on the full image, no backprop anywhere - 3000 iterations, lr annealed 1× → 0.05× over the second half - 40.78 dB fp32, 40.71 dB with the block latent quantized to 8 bits - 6.03 bpp raw 4 bpp selectors + 2 bpp block latent + 0.05 bpp MLP , 5.54 bpp entropy-coded - 96 bits per 4×4 block: 64 selector bits + 32 latent bits - 55 s total on an RTX 5090, everything on the GPU CUDA 13.1 Left: original, right: reconstruction The block artifacts are there if you look closely: