three-ntc-trainer

GPU trainer that bakes a three.js material (MaterialX or plain baked textures) into a compact .ntc Neural Texture Compression model — a shared multiresolution latent grid plus MLP decoder, jointly fit against the material's PBR channels — and exports it for the three-ntc runtime to load.

An implementation of NVIDIA's 2023 Neural Texture Compression paper: rather than baking separate albedo/normal/roughness/metalness textures, this trainer fits all of a material's channels jointly into one small grid + MLP model. Every material pictured above is about 93KB total — often smaller than a single texture from the original material — so scenes trained here ship far richer materials without the usual memory and download cost.

Try the trainer live at three-ntc.ben3d.ca.

Install

npm install three-ntc-trainer

Usage

import { fitNTCMaterial, NTCExporter } from 'three-ntc-trainer';

// `material` is any three.js node material (e.g. loaded via MaterialXLoader).
const trainedMaterial = await fitNTCMaterial(renderer, material, {
  onProgress: ({ step, loss }) => console.log(step, loss),
});

const exporter = new NTCExporter();
const manifest = exporter.parse(trainedMaterial, { name: 'Gold' });
const blob = new Blob([JSON.stringify(manifest)], { type: 'application/json' });

The resulting manifest is the .ntc file NTCLoader (from three-ntc) reads back.

For finer control over training (grid resolution, MLP shape, step count, ...), construct NTCTrainer directly instead of using the fitNTCMaterial convenience wrapper.