Deemotions

This repository contains the implementation of the method illustrated in:

Emotion Recognition in the Wild using Deep Neural Networks and Bayesian Classifiers

Preamble

This repository contains the code to perform emotion recognition both for the following datasets:

1. Emotion Recognition based on CK+ dataset
2. GAF (Group Affect Dataset) associated with EmotiW Challenge 2017 and for the GAF dataset.

The source files for this work are contained into the "wild" folder (it is so funny, isn't it?). This method has been published in the paper "Emotion Recognition in the Wild Using Deep Neural Networks and Bayesian Classifiers". Please cite us if you are using our code, or if you are finding it useful for your research.

Cite:

@inproceedings{surace2017emotion, title={Emotion recognition in the wild using deep neural networks and Bayesian classifiers}, author={Surace, Luca and Patacchiola, Massimiliano and Battini S{\"o}nmez, Elena and Spataro, William and Cangelosi, Angelo}, booktitle={Proceedings of the 19th ACM International Conference on Multimodal Interaction}, pages={593--597}, year={2017}, organization={ACM} }

The rest of the readme is referring to the EmotiW code

Contributors

• Luca Surace

• Massimiliano Patacchiola

Instructions

Dependences

• Numpy

sudo pip install numpy

• OpenCV

sudo apt-get install libopencv-dev python-opencv

• TensorFlow

sudo pip install tensorflow

• Google Cloud Vision API - Please note you need a Google Cloud account to import the python modules and run the code

• Pgmpy

Installation

Download the repository from [here] or clone it using git:

git clone https://github.com/lukeoverride/deemotions.git
cd deemotions/wild

Test with pre-trained weights

To run the code, please type on a terminal:

python ClassifyImage.py <test-images-path> <real-label (Positive|Negative|Neutral)>

Example:

python ClassifyImage.py /home/yourUsername/images/positiveInstances/ Positive

python ClassifyImage.py /home/yourUsername/images/negativeInstances/ Negative

python ClassifyImage.py /home/yourUsername/images/neutralInstances/ Neutral

Training

The code used to train the model is available in ./wild/64net_emotion_detection_training.py for demonstration purposes only.

To launch it, you would need the dataset and additionals .csv files. Unless you are interested in increasing global warming, I do not think it is a good idea.