//

// This software is released under the MIT License.

// https://opensource.org/licenses/MIT

​

/* ===

ml5 Example

Image classification using MobileNet and p5.js

This example uses a callback pattern to create the classifier

=== */

let nn;

const IMAGE_WIDTH = 64;

const IMAGE_HEIGHT = 64;

const IMAGE_CHANNELS = 4;

​

let video;

let trainBtn;

let addDataBtn;

let labelInput;

let resultLabel;

​

function setup() {

  // load the pixels for each image to get a flat pixel array  

  createCanvas(400, 400);

​

  video = createCapture(VIDEO);

  video.size(64, 64)

​

  trainBtn = createButton('train')

  trainBtn.mousePressed(train)

  addDataBtn = createButton('addData')

  addDataBtn.mousePressed(addData)

  resultLabel = createDiv('I see:')

​

  labelInput = createInput();

​

  const options = {

    task: 'imageClassification',

    debug: true,

    inputs:[IMAGE_WIDTH, IMAGE_HEIGHT, IMAGE_CHANNELS],

    layers: [{

        type: 'conv2d',

        kernelSize: 5,

        filters: 8,

        strides: 1,

        activation: 'relu',

        kernelInitializer: 'varianceScaling'

      },

      {

        type: 'maxPooling2d',

        poolSize: [2, 2],

        strides: [2, 2]

      },

      {

        type: 'conv2d',

        filters: 16,

        kernelSize: 5,

        filters: 8,

        strides: 1,

        activation: 'relu',

        kernelInitializer: 'varianceScaling'

      },

      {

        type: 'maxPooling2d',

        poolSize: [2, 2],

        strides: [2, 2]

      },

      {

        type: 'flatten'

      },

      {

        type: 'dense',

        kernelInitializer: 'varianceScaling',

        activation: 'softmax'

      }

    ]

  }

​

  // construct the neural network

  nn = ml5.neuralNetwork(options);

  nn.loadData('daytime_nightime.json', train)

​

​

}

​

function draw(){

  image(video, 0,0, width, height)

}

​

​

function addData(){

  console.log('adding data', labelInput.value())

  nn.addData({image:video}, {label: labelInput.value()})

}

​

function train(){

  // nn.normalizeData();

​

  const TRAINING_OPTIONS = {

    batchSize: 16,

    epochs: 4,

  }

​

  nn.normalizeData();

  nn.train(TRAINING_OPTIONS, finishedTraining)

}

​

​

function finishedTraining() {

​

  console.log("finished training");

  nn.classify([video], gotResults)

​

}

​

function gotResults(err, result) {

  if (err) {

    console.log(err);

    return

  }

​

  // image(video, 0,0, width, height)

  resultLabel.elt.textContent = `I see ${result[0].label}`

​

  nn.classify([video], gotResults)

}