/Facial-Expression-Recognition-using-Deep-Learning

Primary LanguageJupyter NotebookMIT LicenseMIT

Kaggle Python 3.7TensorflowstackoverflowGithub

Facial Expressions Recognition using Convolutional Neural Networks

Branches:

  • master : Contains files related to training the model and realtime testing.
  • deployment : Contains files related to deployment of the deep learning model on web using flask micro framework.
  • Either clone the repo or download the Zip file based on your need, and start working.

Description:

  • fer2013.csv.zip - This is the dataset I've used. Do extract that on to your local system.
  • Facial Expression Detection (csv) - This notebook contains the implementation of Convolutional Neural Networks using fer2013.csv dataset directly.
  • Facial Expression Detection (dir) - This notebook contains the implementation of Convolutional Neural Networks using fer2013.csv dataset using Datagenerator Class of Keras after splitting the csv dataset to directories. The code for converting the fer2013.csv file to directories is in convertodir.py
  • model_weights.h5 & model.json - These weights file and architecture file as json are generated after training the CNN model in Facial Expression Detection (dir) notebook.
  • model_filter.h5 & fer.json - These weights file and architecture file as json are generated after training the CNN model in Facial Expression Detection (csv) notebook.

Dependencies:

  • Python=3.7
  • Tensorflow>=2.1
  • opencv>=4.0.1
  • keras>=2.3.1

Installation of dependencies using Anaconda distribution

  • Create a new virtual environment using conda

    conda create -n faceexpressions python=3.7 -y

  • Activating our new virtual environment

    conda activate faceexpressions

  • Installing necessary libraries

    pip install ipykernel

  • I'd recommend you to use the display name same as environment name

    python -m ipykernel install --user --name faceexpressions --display-name "faceexpressions"

  • If you have want to install Tensorflow GPU then use below command

    conda install tensorflow-gpu==2.0.0

  • If you have want to install Tensorflow CPU then use below command

    conda install tensorflow==2.0.0

  • The below pip command installs keras, opencv, pandas, matplotlib, scikit-learn

    pip install keras opencv-contrib-python pandas numpy matplotlib scikit-learn

  • The below command install jupyter notebook on Anaconda Distribution

    conda install jupyter

  • Once your dependencies installation is done, you can run the notebook files on jupyter notebook(if you are using Anaconda) using below command.

    jupyter notebook

Real-Time Testing of our model:

  • This can be done just by running RealtimePredictions.py

    python RealtimePredictions.py

  • You'll see a window pop up and you can find the predictions on your screen(I've cropped the window). (See the demo file below)

GIF

Result :

  • The architechture that I've built is very pretty simple, and I was able to get an validation accuracy of around 60.25 and training accuracy around 85.25.
  • The accuracy can still be improved by tuning the hyperparameters of the model. We can even make this more accurate by also using Transfer Learning.
Badges Copyright-MIT © aleen42