LibreYOLO Web

Object detection in the browser. 100% MIT Licensed.

The web companion to libreyolo. Same models, same license, no AGPL. YOLOX, YOLO9, RF-DETR, and YOLOv8/v11/v26 running in the browser on WebGPU or WASM.

Install

npm install libreyolo-web onnxruntime-web

Quick Start

import { loadModel } from 'libreyolo-web';

const model = await loadModel('LibreYOLOXn');
const result = await model.predict(imageElement);

console.log(`Found ${result.numDetections} objects`);

That's it. The model auto-downloads from HuggingFace and handles its own preprocessing.

Drawing Boxes

import { loadModel, BoxOverlay } from 'libreyolo-web';

const model = await loadModel('LibreYOLO9t');
const result = await model.predict(imageElement);

new BoxOverlay({ canvas: myCanvas }).draw(result.detections);

Model Zoo

14 pre-trained models, ready to go: LibreYOLOXn, LibreYOLO9s, LibreRFDETRm, and friends. Full list and benchmarks at libreyolo.com/docs.

import { listModels } from 'libreyolo-web';
listModels().forEach(({ name }) => console.log(name));

Your Own Model

const model = await loadModel('./my_model.onnx', {
  modelFamily: 'yolox',  // 'yolo' | 'yolox' | 'yolo9' | 'rfdetr'
  inputSize: 640,
});

Export from the Python sister project:

from libreyolo import LibreYOLO
LibreYOLO('LibreYOLOXs.pt').export(format='onnx', simplify=True)

Docs

Everything else (full API reference, bundler config, backend tuning) lives at libreyolo.com/docs.

License

MIT. Truly MIT. No AGPL.