let facemeshModel;

let face;

let brain;

let faceLabel = "N/A"; // Default label when no face is detected or classified

​

function setup() {

  createCanvas(640, 480);

  video = createCapture(VIDEO);

  video.hide();

​

  // Load the Facemesh model

  facemeshModel = ml5.facemesh(video, modelReady);

  facemeshModel.on('predict', gotFaces);

​

  // Setup neural network

  const options = {

    inputs: 468 * 2, // Using x and y coordinates for each of the 468 facemesh points

    outputs: 4, // Assuming 4 classes for classification

    task: 'classification',

    debug: true

  };

  brain = ml5.neuralNetwork(options);

​

  // Load the trained model

  const modelInfo = {

    model: 'model2/model.json',

    metadata: 'model2/model_meta.json',

    weights: 'model2/model.weights.bin',

  };

  brain.load(modelInfo, brainLoaded);

}

​

function modelReady() {

  console.log('Facemesh model ready');

}

​

function gotFaces(results) {

  if (results.length > 0) {

    face = results[0];

    classifyFace();

  }

}

​

function classifyFace() {

  if (face) {

    let inputs = [];

    for (let i = 0; i < face.scaledMesh.length; i++) {

      inputs.push(face.scaledMesh[i][0]); // x coordinate

      inputs.push(face.scaledMesh[i][1]); // y coordinate

      // Optionally, include z coordinate

      // inputs.push(face.scaledMesh[i][2]);

    }

    brain.classify(inputs, gotResult);

  }

}

​

function gotResult(error, results) {

  if (results && results[0].confidence > 0.75) {

    faceLabel = results[0].label.toUpperCase();

  }

  // Continue classification with the next frame

  classifyFace();

}

​

function brainLoaded() {

  console.log('Facemesh classification ready!');

}

​

function draw() {

  push();

  translate(video.width, 0);

  scale(-1, 1); // Mirror the video feed

  image(video, 0, 0, width, height);

​

  fill(255, 0, 255);

  noStroke();

  textSize(64);

  textAlign(CENTER, CENTER);

  text(faceLabel, width / 2, height / 2);

  pop();

}

​