Codebase for the publication of the first version of the Stanford Emotional Narratives Dataset (SEND) (for Transactions on Affective Computing paper. Link: on Arxiv)
Citation: If you want to refer our work, please consider cite (citation is required if you want to use our SEND dataset):
@article{ong2019modeling,
title={Modeling emotion in complex stories: the Stanford Emotional Narratives Dataset},
author={Ong, Desmond and Wu, Zhengxuan and Tan, Zhi-Xuan and Reddan, Marianne and Kahhale, Isabella and Mattek, Alison and Zaki, Jamil},
journal={IEEE Transactions on Affective Computing},
year={2019},
publisher={IEEE}
}
Here, We first introduce the first version of the Stanford Emotional Narratives Dataset (SENDv1): a set of rich, multimodal videos of self-paced, unscripted emotional narratives, annotated for emotional valence over time. The complex narratives and naturalistic expressions in this dataset provide a challenging test for contemporary time-series emotion recognition models.
Researchers are welcomed to request the dataset. You can view the dataset here at SEND homepage).
In this section, we present several time-series approaches to model valence ratings on the SENDv1. We implement:
a baseline (non-time-series) discriminative model, a Support Vector Regression (SVR)
a baseline generative model, a Hidden Markov Model (HMM)
a state-of-the-art discriminative Long Short-Term Memory (LSTM) model
a state-of-the-art (deep) generative Variational Recurrent Neural Network (VRNN) model.
Use the package manager pip to install all the required packages mentioned in the requirement.txt.
pip install -r requirements.txtYou will need to go to the model's subdirectory, and run the following command.
For all the supported command,
cd models/lstm
python train.py -HWith the default settings,
cd models/lstm
python train.py