This the Source code for our adversarial based neural network for affect estimations in the wild.
This source code holds the image augmentations techniques used on our other repositories : https://github.com/deckyal/FADeNN.
Requirements :
- Python 2.7
- Pytorch 0.4
- OpenSmile : https://www.audeering.com/opensmile/
Preparations :
- Extract the LLD sound features using opensmiles.
Trainings :
- chal-main.py : To train the discriminator only using l2 loss
- chal-main_gan_single.py : To train using both Generator and Discriminators. Please observe the required parameters to tweak the configurations.
Tests : The validation code can be seen on the chal-man_gan_single.py, do_test().
Citation : [D. Aspandi, A. Mallol-Ragolta, B. Schuller and X. Binefa, "Latent-Based Adversarial Neural Networks for Facial Affect Estimations," in 2020 15th IEEE International Conference on Automatic Face and Gesture Recognition (FG 2020) (FG), Buenos Aires, undefined, AR, 2020 pp. 348-352.]