Instantly share code, notes, and snippets.

👋

Matt DesLauriers mattdesl

👋

artist & creative coder

7.3k followers · 28 following

View GitHub Profile

Recently created

Least recently created

Recently updated

Least recently updated

@mattdesl

mattdesl / gmm.py

Last active March 30, 2026 14:37

color quantization using gaussian mixture model in OKLCH

import numpy as np

from PIL import Image

from sklearn.mixture import GaussianMixture

import matplotlib.pyplot as plt

import matplotlib.patches as patches

import ColorMath

IMAGE_PATH = "data/adirondack_chairs.png"

OUTPUT_PATH = "output/quant_oklch.png"

K = 10

@mattdesl

mattdesl / gaussian-random.js

Created January 19, 2026 18:14

function nextGaussianBoxMuller(mean = 0, standardDerivation = 1, rng = Math.random) {

return (

mean +

standardDerivation *

(Math.sqrt(-2.0 * Math.log(rng())) * Math.cos(TWO_PI * rng()))

);

}

@mattdesl

mattdesl / main.js

Last active August 5, 2025 03:09

uform3 perf test

import {

Tensor,

TextEncoder,

TextProcessor,

ImageEncoder,

env,

} from "./uform-encoder.js";

import imageUrl from "../assets/images/monalisa.png";

const model = "fp32"; // "v3", "fp16" or "fp32"

@mattdesl

mattdesl / decoder.js

Last active December 18, 2024 18:22

Filigree file encoder and decoder. MIT License, free to use.

/**

* A decoder for the Filigree file format, which are assets stored on IPFS.

* Since this code ended up on-chain, it is written in such a way as to compress well.

* @license MIT

* @author Matt DesLauriers (@mattdesl)

**/

export function decodeStream(buffer, headers, layer, cell) {

let dv = new DataView(buffer);

let byteOffset = 0;

@mattdesl

mattdesl / README.md

Created July 8, 2024 08:59

Fast OKLab Gamut Mapping in JavaScript

Using OKLCH and sRGB. This also works for Display P3, however you'll have to pass the correct coefficients to the cusp & intersection utilities, and instead of using linear sRGB, you would convert to linear Display P3.

If you wish to support both sRGB and Display P3 gamut mapping, it is possible with a single function and a matrix parameter, as they both use the same coordinate range (0..1) and gamma transfer function.

import { OKLCH_to_linear_sRGB } from './convert.js';
import { rgb_in_gamut, linear_to_gamma } from './util.js';
import { find_cusp, find_gamut_intersection } from './oklab-cusp.js';