If you want to use our model in your project, we suggest you to visit this repository. This repository is dedicated to the preparation of the data and the training on Affect-Net.
This is a deep learning model to classify facial expressions in five categories:
- Neutral
- Happy
- Sad
- Angry
- Surprised
We implement different architectures:
To train and evaluate our networks, we used the database Affect-Net, which is composed of 1M of images.
Our best accuracy is 74% on a validation set using Inception-resnet v2.
Here you can see a demo.
