Made by RITIK PARIDA
CSE,NITR-22
(This is NOT a Coursera-Guided Project)

Facial Expression Recogntion (FER)

This Project is a real time facial expression recogniser built using OpenCV with the help of Machine Learning and deep Nets.It shows the most likely expression of a face like Happy,Angry,Calm, Surprised etc.

This Repository contains two models

Using KNN (K Nearest Neighbours) Algorithm.

• This model has achieved a very high accuarcy on faces like mine and that of my friends.The reason being simple, it's trained on a dataset consisting of faces generated through openCV locally on my Computer.

• However this model fails to give good results on unseen data ( New faces).

Using CNN ( Convolutional Neural Networks) Algorithm

• This Model has been trained on the Kaggle dataset FER2013 which had around 38k images.Trained on 24k images , validated on 6k images. Finally tested on 8k faces.

• This Model works well on New Faces too.

• Model in figures

  • Val accuracy : 70.5 %
  • Test accuracy ~ 70 %
  • Model that won the Kaggle contest had an acc of 71%