USED: A Large Scale Social Event Detection Dataset

  Abstract

Event discovery from single pictures is a challenging problem that has raised significant interest in the last decade. We believe that deep learning can prove to be a breakthrough also in this research area, providing a successful framework to learn the inherent event-related characteristics of a media depicting the relevant moments of an event, notwithstanding the variety of visual appearances. In this context, it is well known that the main limitation in the application of CNNs is the unavailability of large datasets. To this aim, in this paper we provide a large scale properly annotated and balanced dataset of 525,000 images, covering every aspect of 14 different types of social events, selected among the most shared ones in social network. Such a large-scale collection of event-related images is intended to become a powerful support tool for the research community in multi- media analysis by providing a common benchmark for training, testing, validation and comparison of existing and novel algorithms. In this paper, we provide a detailed description of how the dataset is collected, organized and how it can be beneficial for the researchers in the multimedia analysis domain. By providing this dataset, we hope to gather research community in the multimedia and signal processing domains to advance this application.

  Data Description

The dataset is composed of 525,000 images, which are arranged into 14 different types of social events, selected among the most shared ones in social network. In order to make it balance, we collected an equal number of images (35,000) per event-class from Flickr using the respective API. In the image collection, we tried our best to cover every aspect of the considered social events by collecting images for same events with diverse contents in terms of viewpoints, colors, group pictures vs. single portrait and outdoor vs. indoor images, where the high variability of the represented in- formation can be effectively explored to ensure better performances in event classification. For example, in graduation, sports and wedding event-classes we collected single person pictures, group pictures and the pictures taken at the time of celebration. Similarly, in ski holiday and mountain trip classes our dataset covers both the pictures taken in green mountains as well as images of white and bare mountains. Another important characteristic of this dataset is the diversity in culture. For example, in wedding image collection we tried to cover diverse cultures by collecting wedding images from different communities and cultures (e.g., wedding images from Asian and European countries have been collected).

  Datasets Comparison

EiMM Dataset

CLass

Concert

Graduation

Mountain Trip

Meeting

Picnic

Sea-holiday

Ski-holiday

Wedding

#Images

1085

1815

2051

795

1627

2253

1817

1776

 

SED  Dataset

Class

Concert

Conference

Exhibition

Fashion

Protest

Sports

Theater

-

#Images

71556

2975

342

1556

1039

403

4342

-

 

 

Our Dataset

Class

Concert

Graduation

Mountain Trip

Meeting

Picnic

Sea-holiday

Ski-holiday

Wedding

#Images

35,000

35,000

35,000

35,000

35,000

35,000

35,000

35,000

  Download files

CSV files
Zip file containing labels in csv format
 
Datasets
Training Set
Test Set

  Contacts

For further details please contact:

Kashif Ahmad

Nicola Conci