Culture and Social Signals In the Wild

Many Faces of Anger in the Wild dataset (MFA-Wild )

Cultural differences in emotion expression is considered a barrier in affective computing systems, since most research and datasets are biased toward Western subjects. Moreover, most datasets, only have basic emotion labels, while a single label like anger may have wide range of arousal levels. In this dataset, we collected more than 200 in-the-wild videos from North American and Persian cultures with fine-grained labels of: 'annoyed', 'anger', 'disgust', 'hatred' and 'furious' and 13 related emojis.

Data Description

Type of data:

  • Video data, in mp4 file format.

  • Data is separated into training and testing. Training data can be used for k-fold cross-validation.

  • Emotion/emoji labels are in the csv files.

  • Extracted features from OpenFace are in files na_dataset.csv and persian_dataset.csv.


The data collection approach and baseline models are published in a paper here. You may also be interested in the related paper here.


The MFA-Wild

Our data is publicly available for research purposes. Please cite our paper:


author={Javadi, Roya and Lim, Angelica},

booktitle={2021 16th IEEE International Conference on Automatic Face and Gesture Recognition (FG 2021)},

title={The Many Faces of Anger: A Multicultural Video Dataset of Negative Emotions},