CALL FOR PAPERS : 

Workshop Description 

The need for analysis and understanding of content captured by cameras in video makes the event detection task a pivotal one in computer vision, with many applications such as intelligent visual surveillance of human activities, intelligent visual surveillance of animal activities, optical motion capture, and multimedia applications. While complexity of the challenges needed to be handled for these applications is increasing and the environments are more and more various due to novel acquisition devices, quite a plethora of methods have appeared since the 1990s. However, there are two key points that distinguish the actual situation:

(i)  The diversification of video datasets: Modern datasets offer significantly greater variety compared to those from the start of 2010’s that contained exclusively urban scenes. With the ongoing advances in data acquisition and continuous progress in sensors’ technologies, these datasets could be expanded by adding videos of maritime scenes, underwater scenes, natural ground scenes (i.e., forest), or complex-component urban settings. The evoked challenges are thus more and more complex. This naturally leads us to the second motivation:

(ii)  The use of deep learning-based models, in particular graph neural networks and few-shot learning models, becomes essential to be able to monitor such a dataset with deployable algorithms.

In this context, the needs for an up-to-date picture of the current landscape in the field as well as an adapted framework to compare their performance are crucial. To address these points, together with encouraging the participation of researchers from different communities, the workshop is complemented by a challenge on the Event Detection in Videos. Its results will be announced and discussed during the workshop. 

More details concerning the methodology adopted for the challenge such as dataset hierarchy, evaluation strategy, scenarios definition, could be found in the associated arXiv paper.

   Invited speakers :

–  Anthony Cioppa, Montefiore Institute of the University of Liège, Belgium.

–  Hideo Saito, Department of Information and Computer Science, Faculty of Science and Technology, Keio University, Tokyo.

For any question concerning the workshop, please contact : Anastasia Zakharova anastasia.zakharova (at) univ-lr.fr

Supported by: