// for this feature to work

​

// put the model URL from https://teachablemachine.withgoogle.com/train/pose here

let poseModelURL = "https://teachablemachine.withgoogle.com/models/a-c04xzVV/";

​

let video;

let tmPoseModel;

let prediction = [];

​

async function setup() {

  createCanvas(640, 480);

​

  video = createCapture(VIDEO, { flipped: true });

  video.size(width, height);

  video.hide();

​

  // this loads the custom model from Teachable Machine and schedules the

  // first prediction to be in 33ms

  tmPoseModel = await tmPose.load(poseModelURL + "model.json", poseModelURL + "metadata.json");

  console.log("Model loaded");

  setTimeout(predictTmPose, 33);

}

​

function draw() {

  image(video, 0, 0, width, height);

​

  if (prediction.length > 0) {

    fill(255, 0, 0);

    text(prediction[0].className + ' (' + prediction[0].probability + ')', 20, 20);

  }

}

​

​

​

// please ignore the details of this piece of code

// it runs the custom model, and updates the prediction

// variable with the result (we use the variable in draw)

async function predictTmPose() {

  try {

    let { pose, posenetOutput } = await tmPoseModel.estimatePose(video.elt);

    prediction = await tmPoseModel.predict(posenetOutput);

    // sort result by probability

    prediction = prediction.sort((a,b) => (a.probability < b.probability) ? 1 : ((b.probability < a.probability) ? -1 : 0))

  } catch (e) {

    console.warn(e);

  }

  //console.log(prediction);

  setTimeout(predictTmPose, 33);

}

​