Human Emotion Recognition by a CNN (TensorFlow)

Introduction

In this project, the aim is to implement a CNN to achieve the task of human emotion recognition from facial images. The architecture of the implemented Neural Network is as follows:

Layer	Kernel	Activation	Output Shape
Input	-	-	48x48x1
Conv2D	3x3x32	ReLU	46x46x32
BN	-	-	46x46x32
Conv2D	3x3x32	ReLU	44x44x32
BN	-	-	44x44x32
Conv2D	3x3x32	ReLU	42x42x32
BN	-	-	42x42x32
Conv2D	3x3x64	ReLU	40x40x64
BN	-	-	40x40x64
MaxPooling	2x2	ReLU	20x20x64
Conv2D	3x3x32	ReLU	18x18x64
BN	-	-	18x18x64
Conv2D	3x3x32	ReLU	16x16x64
BN	-	-	16x16x64
MaxPooling	2x2	ReLU	8x8x64
Conv2D	3x3x32	ReLU	6x6x128
BN	-	-	6x6x128
MaxPooling	2x2	ReLU	3x3x128
Conv2D	3x3x32	ReLU	1x1x128
Flatten	-	-	128
Dense	-	ReLU	7

Dataset: FER-2013

This dataset contains 35887 images and their corresponding labels. The images are in bytes, rather a conventional image file. Therefore, images are first read using the BytesIO method of the io library and then converted to an image using PIL. Then, they are turned into numpy arrays for easier manipulation and convenience.

There 7 classes in this dataset, each represented by a number (0-6): 0=Angry, 1=Disgust, 2=Fear, 3=Happy, 4=Sad, 5=Surprise, 6=Neutral.

All the data points are first split into 2 segments of training and test data, where test data contains 15% of the whole. Also, during the training (fitting) phase, 5% of the training data is considered as the validation data.

Installation

First, you'll need to clone this repository to your local drive via the following command:

$ git clone https://github.com/aylinghsr/Human_Emotion_Recognition.git
$ cd Human_Emotion_Recognition

Alternatively, if git package is not installed, you can download the zip file for this repository and then extract it.

Requirements

This project is written in Python3 and requires Pillow, IO, Numpy, Scikit-learn, and TensorFlow.

All the required libraries can be installed by running the following command:

$ pip install -r requirements.txt

If the command above results in an error, you can also try:

$ python -m pip install -r requirements.txt

Also, the code will download the dataset automaticly on your computer.

Results

The designed and implemented CNN gets trained on the training data on 15 epochs. After this, it achieves an accuracy of 0.9313 on the training data, 0.5773 on the validation data, and 0.5804 on the test data.