Tensorflow Nodejs example. Read an image from file, detect objects and save a new image with bounding boxes drawn. 100% server side no browser code.

tensorflow_node.js

const tf = require('@tensorflow/tfjs-node')
//const tf = require('@tensorflow/tfjs-node-gpu')

const cocossd = require('@tensorflow-models/coco-ssd');

const fs = require('fs');
const gm = require('gm');


var path = 'image.jpg';

//https://js.tensorflow.org/api_node/latest/
const buf = fs.readFileSync(path);
const input = tf.node.decodeJpeg(buf);

load(input);

async function load(img){
  const model = await cocossd.load({ 'base': 'mobilenet_v2' });
  const predictions = await model.detect(img);

  console.log('Predictions: ');
  console.log(predictions);
  drawPredictions(predictions);
}

async function drawPredictions(predictions) {
  var pic = gm(path);

  for(let i = 0; i < predictions.length; i++) {
    var p = predictions[i].bbox;

    pic.fill('transparent');
    pic.stroke('#ff0000', 5);
    pic.drawRectangle(p[0], p[1], p[0]+p[2], p[1]+p[3]);

    pic.fill('red');
    pic.stroke('#ff0000', 1);
    pic.fontSize(25);
    pic.drawText(p[0]+10, p[1]+25, predictions[i].class);
  }

  pic.write('predicted-' + path, function (err) {
    if (!err) console.log('Done');
  });
}