This is the repo for the paper Video-based emotion estimation using deep neural networks: a comparative study. This repo contains the code used in the paper to perform training of DL models on the OMG Emotion Challenge dataset. Here we also give to researchers the pre-trained weights of the backbones, as used in the paper. Some are taken from other sources (see the paper for details), while some others we trained ourselves.
To run the training of the models, you should use the train.py script. The file config_train.yaml contains the necessary configurations. Specifically, it is used to specify the filepath for the dataset and the pre-trained weights of the backbone, as well as the choice for the models.
In order to run the training, the file prepare_filelist.py should be used.
The code has been inspired by https://github.com/ewrfcas/OMGEmotionChallengeCode.
2023 Leonardo Alchieri, Luigi Celona, Simone Bianco
If you use our code, please consider cite the following:
- Leonardo Alchieri, Luigi Celona, and Simone Bianco. Video-based emotion estimation using deep neural networks: a comparative study. In ICIAP 2023 Workshops, Springer International Publishing, 2023.
@inproceedings{alchieri2023video,
author = {Alchieri, Leonardo and Celona, Luigi and Bianco},
year = {2023},
title = {Video-based emotion estimation using deep neural networks: a comparative study},
booktitle = {ICIAP 2023 Workshops},
publisher="Springer International Publishing",
}