README.md

YOLO v5 in Node.JS

Object Detection using YOLO for Node.js backend.

ES6 compatibile

Installation

npm i yolov5

It also installs the required dependencies: tfjs, tfjs-node and regenerator-runtime

For more info on tensorflow backend in node.js:
(https://github.com/tensorflow/tfjs/tree/master/tfjs-node)
(https://github.com/tensorflow/tfjs)

Usage

const tf = require('@tensorflow/tfjs');
const tfnode = require('@tensorflow/tfjs-node');
const path = require('path');
const fs = require('fs');
require('regenerator-runtime');

const yolov5 =  require('./YOLO');


const main = async() => {
	// Reading Image
	const imgName = 'Pic_3_640.jpg'; // RGB images
	let imageBuffer = fs.readFileSync(path.join(__dirname, `${imgName}`));
	let image0 = tfnode.node.decodeImage(imageBuffer, 3);

	// Resize the image, if not 640x640
	// image0 = tf.image.resizeBilinear(image0, size = [imageSize, imageSize]);

	image = tf.cast(image0, 'float32');
	image = image.div(tf.scalar(255));                  // Converting to float32
	image = image.expandDims(0);
	
	// Loading Model
	const yolo = yolov5;
	await yolo.load();
	let predictions;

	if (yolo.model != '') {
		console.log("Model loaded \n");
		const result = await yolo.predict(image);
		predictions = yolo.getDetections(result);
	} else {
		return console.error("No model");
	}

	if (predictions.length > 0) console.log("Predictions:", predictions);
	
	yolo.dispose;
};

main();

View model parameters, methods, functions: console.log(yolo.details())

Output

Prediction format:

{
	"bbox": [409, 1, 639, 169],		// [x0, y0, x1, y1]
	"score": 0.6489,				// probability score
	"class": "car"					// Object class name
}

Notes

• Current backend support only for CPU via tfjs-node.

• Insert regenerator-runtime or its equivalent for async-await based functional implementation.

• tfjs-node is optimized for some specific hardware: (AVX512, MKL-DNN).

Some hardware optimized models run faster in CPU rather than GPU.

Test

• Create a file named yolo.test.js in test folder and test it using mocha.

Model conversion using tfjs-converter Link

tensorflowjs_converter \ --input_format=tf_saved_model \ --output_node_names='MobilenetV1/Predictions/Reshape_1' \ --saved_model_tags=serve \ /mobilenet/saved_model \ /mobilenet/web_model

We try to optimize the model for being served on the web by sharding the weights into 4MB files - that way they can be cached by browsers.
We also attempt to simplify the model graph itself using the open source Grappler project. Graph simplifications include folding together adjacent operations, eliminating common subgraphs, etc. These changes have no effect on the model’s output.
For further optimization, users can pass in an argument that instructs the converter to quantize the model to a certain byte size.

Quantization is a technique for reducing model size by representing weights with fewer bits. Users must be careful to ensure that their model maintains an acceptable degree of accuracy after quantization.