​

const checkpointURL = modelURL + "model.json";

const metadataURL = modelURL + "metadata.json";

​

let model;

let totalClasses;

let myCanvas;

​

let classification = "please wait..";

let probability = "?";

let poser;

let video;

​

​

// A function that loads the model from the checkpoint

async function load() {

  model = await tmPose.load(checkpointURL, metadataURL); 

  totalClasses = model.getTotalClasses();

  console.log("Number of classes, ",totalClasses);

}

​

​

async function setup() {

  myCanvas = createCanvas(400, 400);

  videoCanvas = createCanvas(320, 240)

​

  await load();

  video = createCapture(VIDEO, videoReady);

  video.size(320, 240);

  video.hide();

}

​

function draw() {

  background(255);

  if(video) image(video,0,0);

  fill(0)

  textSize(10);

  text("Result : " + classification, 220, 40);

​

  text("Probability : " + probability, 220, 20)

  textSize(8);

  if (poser) { //did we get a skeleton yet;

    for (var i = 0; i < poser.length; i++) {

      let x = poser[i].position.x;

      let y = poser[i].position.y;

      ellipse(x, y, 5, 5);

      //text(poser[i].part, x + 4, y);

    }

  }

​

}

​

function videoReady() {

  console.log("Video is...Ready !");

  predict();

}

​

​

async function predict() {

  // Prediction #1: run input through posenet

  // predict can take in an image, video or canvas html element

  const {pose,posenetOutput} = await model.estimatePose(video.elt );

  // Prediction 2: run input through teachable machine assification model

  const prediction = await model.predict(

    posenetOutput,totalClasses);

  // Sort prediction array by probability

​

  const sortedPrediction = prediction.sort((a, b) => -a.probability + b.probability);

​

  classification = sortedPrediction[0].className;

  probability = sortedPrediction[0].probability.toFixed(2);

  if (pose) poser = pose.keypoints; 

  predict();

}