FrontEnd-EmotionDetection

This is a real-time Emotion detection using TensorFlow.js to load a pretrained model into the browser .

Abstract

Detect faces

• We use two methods to detect faces:

  1. Chrome Shape Detection API
  2. face-api.js

Dataset

• We combine two datasets:
  1. Microsoft FERPlus as one of the dataset to train the emotion-detect model.
  2. Real-world Affective Faces Database (RAF-DB) as another dataset.

Convert Model

• Use TensorFlow.js converter convert Keras model to .json file for loading and running Javascript inference.

Requirements

• For MobileNetImage.html and MobileNetWebcam.html:

  Chrome Shape Detection API : FaceDetector: Chrome on Android, macOS, Windows 10 platfrom. Go to chrome browser chrome://flags/#enable-experimental-web-platform-features and enable the feature

• For TinyFaceDetectWebcam.html you can run on Chrome, Safari and Firefox.

Demo

Note, that wearing glasses might decrease the accuracy of the prediction results.

All demos are on our LAB server

• MobileNetImage.html ( face detection via Chrome Shape Detection API )

• TinyFaceDetectWebcam.html ( face detection via face-api tiny-face-detector )

• MobileNetWebcam.html ( face detection via Chrome Shape Detection API )