Emu (in progress)

Emotion Mini Utility - simple API for recognizing human emotions on picture.

This project's goal is to quickly classify 7 basic human emotions (anger, contempt, fear, disgust, happiness, sadness, surprise) by picture of a human face. Our main goal is to help in research and treatment of autism, but possible applications of Emu in industry are endless.

Project Structure

|-- app
|-- model
|-- data
|   |-- cascades
|   |-- dataset
|   |-- logs
|-- emu.py
|-- preprocess.py
|-- train.py

app - package for GUI app
model - package for building and training model
data/cascades - directory for OpenCV cascade needed for face detection
data/dataset - directory where the you should unpack FER-2013 dataset
data/logs - directory for all outputs of training
emu.py - main GUI application script
preprocess.py - main data preprocessing script
train.py - main training script

Setup

First make sure you have Python 3.5 or newer.

Dependencies

Install required packages.

pip3 install tensorflow-gpu numpy PyQt5

If your machine does not support CUDA.

pip3 install tensorflow numpy PyQt5

In case of trouble with installation of Tensorflow go here.

The application also requires cv2 module. Refer to these tutorials for guidelines regarding the OpenCV installation.

Application

If you want to quickly launch gui application.

python3 emu.py

Training

If you want to train your own model.

Download and unpack FER-2013 database to /data/dataset directory.
Before you run the training for the first time, remember to preprocess the data.

python3 preprocess.py

Run the script to train the model.

python3 train.py

Status

Project in progress. Feedback welcome. Still improving neural net. Don't mind the labels in GUI yet, they are in a wrong order.