Node.js quickstart

The same built package runs in Node 22 or later as ESM. With an installed tarball and an unbundled Node entry point, the default loader finds the adjacent WASM automatically.

edges.ts

import {
  Canny,
  COLOR_BGR2GRAY,
  COLOR_BGRA2GRAY,
  CV_16U,
  CV_8U,
  cvtColor,
  decodeImage,
  encodeImage,
  IMREAD_UNCHANGED,
  initOpenCV,
  Mat
} from '@banou/opencv-wasm'
import { readFile, writeFile } from 'node:fs/promises'
await initOpenCV()
using image = decodeImage(await readFile('input.png'), IMREAD_UNCHANGED)
using gray = new Mat()
using edges = new Mat()

if (image.channels() === 4) cvtColor(image, gray, COLOR_BGRA2GRAY)
else if (image.channels() === 3) cvtColor(image, gray, COLOR_BGR2GRAY)
else image.copyTo(gray)

// Canny expects an 8-bit input. Convert explicitly if the decoded image is 16-bit.
using gray8 = new Mat()
gray.convertTo(gray8, CV_8U, image.depth() === CV_16U ? 1 / 257 : 1)
Canny(gray8, edges, 50, 150)
await writeFile('edges.png', encodeImage('.png', edges))

Run TypeScript with your existing build or a runner such as tsx. Use module: "NodeNext", moduleResolution: "NodeNext", strict checking and a library set containing ESNext and DOM for the package’s shared browser types. The Node processing path does not need a DOM implementation.

Native files and host files

Node’s filesystem and FS are separate. readFile returns host file bytes. Native model loaders expect paths inside FS:

import { dnn_readNetFromONNX, FS } from '@banou/opencv-wasm'
FS.writeFile('/model.onnx', await readFile('model.onnx'))
using net = dnn_readNetFromONNX('/model.onnx')
FS.unlink('/model.onnx')

See files and codecs for format support, and matrices and ownership before retaining arrays across native calls.