Skip to content

Latest commit

 

History

History
60 lines (49 loc) · 2.97 KB

File metadata and controls

60 lines (49 loc) · 2.97 KB

realtime-face-emotion-recognition

Done into two main steps:

  1. Face-detection: from the video source using OpenCV and haarcascade algorithm.
  2. Emotion recognition:
    - First solution: Using a model trained on FER-2013 dataset with Tensorflow.
    - Second solution: I've used DeepFace package as a prefabricated solution.

Table of contents

Dataset

FER 2013 Dataset:

The data consists of 48x48 pixel grayscale images of faces. The dataset consists of 7 unblanced classes

Note that the data has lots of pitfalls:

So don't expect a high accuracy on training I got about 70% on the validation set

  • Imbalanced classes: you can notice from the below charts that the happy class represents 25% of the data
    imbalanced
  • Some other problems exist in the dataset like occulsion, contrast variation and Intra class variation: imbalanced

Project Roadmap

roadmap

Problem solutions and demos

Solutions:

Demos:

Technologies

Useful resources

Contacts

LICENSE

GNU GPL V3