This repository contains the open source code which reproduces the results for the paper: Adversarial learning for modeling human motion. The authors of this paper are: Qi Wang, Thierry Artières, Mickael Chen, and Ludovic Denoyer.
-
For download the EMILYA dataset, you should contact the owner Catherine Pelachaud.
-
Clone this repository code to your computer and rename the root folder as "Seq_AAE_V1" .
-
Install the relevant packages
- Keras
- matlotlib
-
Data Preprocessing: You should save the motions in EmilyaDataset into a npz file according to the readme.md file in the folder 'datasets/EmilyaDaset/'.
-
For training the models in the paper, you should navigate to the "\Training" directory under the root directory of the project and there you can find the following five folders:
Conditional_SAAE
Seq_AAE
Double_GAN_Continuous_Emotion_Representation
Seq_VAE
Double_Gan_Condition_SAAE
These folders are named by the model's name. Enter each of the folders, you can find the training file and evaluation scripts in a subfolder named '\evaluation'. By running the training file in terminal, you can train the model from scratch. If you want to tune the hyperparameters, you can directly modified them in the training file.