index.js

const classifier = knnClassifier.create()
const webcamElement = document.getElementById("webcam")

let net

async function app() {
  console.log("Loading mobilnet...")

  net = await mobilenet.load()

  console.log("Loaded model")

  const webcam = await tf.data.webcam(webcamElement)

  const addExample = async (classId) => {
    const img = await webcam.capture()

    const activation = net.infer(img, true)

    classifier.addExample(activation, classId)

    img.dispose()
  }

  document.getElementById("dog").addEventListener("click", () => addExample(0))
  document.getElementById("cat").addEventListener("click", () => addExample(1))
  document
    .getElementById("snake")
    .addEventListener("click", () => addExample(2))

  while (true) {
    if (classifier.getNumClasses() > 0) {
      const img = await webcam.capture()

      const activation = net.infer(img, "conv_preds")

      const result = await classifier.predictClass(activation)

      const classes = ["Dog", "Cat", "Snake"]

      document.getElementById("console").innerText = `
                prediction: ${classes[result.label]}\n
                probabilty: ${result.confidences[result.label]}
            `

      img.dispose()
    }

    await tf.nextFrame()
  }
}

app()