Multi-View Foul Recognition

We introduce a multi-label video recognition task focused on football referee decisions. This task involves multi-task classification of multi-view videos. You must assign two labels for each multi-view action: the first label determines whether an action is a foul, along with its corresponding severity, and the second label identifies the type of action.

Our task.

We introduce a multi-label video recognition task focused on football referee decisions. This task involves multi-task classification of multi-view videos. You must assign two labels for each multi-view action: the first label determines whether an action is a foul, along with its corresponding severity, and the second label identifies the type of action.

All of our classes.

First label: {No Offence, Offence + No Card, Offence + Yellow Card, Offence + Red Card} Second label: {Standing Tackle, Tackle, Holding, Pushing, Challenge, Dive, High Leg, Elbowing}

Our data.

The data contains over 3,000 multi-view videos for training, validation, and testing and 273 actions for the challenge set.

Our Metric.

The metric for evaluating your model is the mean of the two balanced accuracies for the two tasks.

For more details, check out our development kit on github

Development Kit

How to cite this work ?

@inproceedings{held2023vars,

title={VARS: Video Assistant Referee System for Automated Soccer Decision Making from Multiple Views},

author={Held, Jan and Cioppa, Anthony and Giancola, Silvio and Hamdi, Abdullah and Ghanem, Bernard and Van Droogenbroeck, Marc},

booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},

pages={5085--5096},

year={2023}

}