Skip to content

Load your YOLO v5 or v8 model in browser using tensorflow.js

License

Notifications You must be signed in to change notification settings

asouqi/yolo-tfjs

Repository files navigation

⚡️ Load your YOLO v5 or v8 model in browser

Run object detection models trained with YOLOv5 YOLOv8 in browser using tensorflow.js

Demo

check out a demo of Aquarium Dataset object detection

Install

Yarn

yarn add yolo-tfjs

Or NPM

npm install yolo-tfjs

Usage Example

import YOLOTf from "yolo-tfjs";

const CLASSES = ["fish", "jellyfish"]
const COLORS = ["#00C2FF", "#FF9D97"]
const imageRef = useRef<HTMLImageElement>(null)

// load model files
const yoloTf = await YOLOTf.loadYoloModel(`model_path/model.json`, CLASSES, {
    yoloVersion: 'v8', onProgress(fraction: number){
        console.log('loading model...')
    }})

// return dection results with detected boxes
const results = await yoloTf.predict(imageRef.current)

// draw boxes in the canvas element
yoloTf.renderBox(canvasRef.current, {
    ...results, ratio: [results["xRatio"],results["yRatio"]]
}, COLORS)

API Docs

loadYoloModel(model, classes, config): YOLOTf

Args

Param Type Description
model string path to model.json file
classes string[] classes of the trained model
config Object see below model configuration
Config Type Default Description
[options.scoreThreshold] Number 0.5
[options.iouThreshold] Number 0.45
[options.maxOutputSize] Number 500
[options.onProgress] Callback (fraction: number) => void
[options.yoloVersion] YoloVersion _ selected version v5 or v8

YOLOTf

PredictionData: {boxes, classes, scores, xRatio, yRatio}

predict(image, preprocessImage): PredictionData

Param Type Description
image HTMLImageElement
preprocessImage (image: HTMLImageElement) => PreprocessResult this optional param to use custom image preprocessing

renderBox(canvas, predictionData, colors): {boxes, classes, scores, xRatio, yRatio}

About

Load your YOLO v5 or v8 model in browser using tensorflow.js

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published