VisE-O
The VisE datasets plus additional documentation can be found on GitHub.
VisE-O: Visual Event Ontology
Based on a set of real-world events from EventKG, the Visual Event Ontology (VisE-O) contains 409 nodes describing 148 unique event types such as different kinds of sports, disasters, and social events with high news potential that can be created with little supervision. It covers the largest number of event types for image classification to date.
A snapshot of the refined Visual Event Ontology.
Cite as
@inproceedings{muellerbudack2021vise,
title={{Ontology-driven Event Type Classification in Images}},
author={Müller-Budack, Eric and Springstein, Matthias and Hakimov, Sherzod and Mrutzek, Kevin and Ewerth, Ralph},
year={2021},
booktitle={The IEEE Winter Conference on Applications of Computer Vision},
}
title={{Ontology-driven Event Type Classification in Images}},
author={Müller-Budack, Eric and Springstein, Matthias and Hakimov, Sherzod and Mrutzek, Kevin and Ewerth, Ralph},
year={2021},
booktitle={The IEEE Winter Conference on Applications of Computer Vision},
}