README.md

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

• The dataset and its challenges
• High level roadmap for the project
• The mentioned solutions and demos for how to tackle the problem
• Technologies Used in this project
• Some useful resources
• For more info and getting in touch
• LICENCSE

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
  
• Some other problems exist in the dataset like occulsion, contrast variation and Intra class variation:

Project Roadmap

Problem solutions and demos

Solutions:

• Realtime emotion recognition using the trained model
• Realtime emotion recognition using DeepFace

Demos:

• Emotion recognition Model training in the fer-2013 dataset
• Realtime face detection demo
• Emotion recognition for multiple faces in a snigle image demo

Technologies

• Tensorflow 2.0
• Keras
• OpenCV
• Python
• DeepFace

Useful resources

• Fer-2013 dataset link in kaggle
• Realtime emotino recognition hypertuning mobilenet model youtube video
• PAZ a GitHub repo which contains different applications on computer vision

Contacts

• LinkedIn
• Email: amshrbo@gmail.com
• Twitter

LICENSE

GNU GPL V3