Towards A Unified Framework For Finding What Is Interesting In Videos
Finished Project
Abstract The goal of this project is to develop and apply effective computer vision techniques that can automatically detect "what is interesting" in such videos. Here, what is meant by interesting may refer to different notions such as where people look in videos, salient objects, interesting motion patterns or moments in videos. All these topics listed above form the subject matter of the proposed project. It is important to note that each notion of interestingness involves different computational problems that need to be solved. This project, unlike the previous work, will investigate all these interrelated concepts within a unified framework and this will allow us to detect different levels of interestingness in a more accurate way.Although the aforementioned problems are closely related to each other, most of the existing literature treats them separately – as mentioned above. In this project, we bridge this gap by developing various techniques and methodologies for solving each task, which take advantage of simultaneously using additional information from other sources of interestingness.

Sponsors: The Scientific and Technological Research Council of Turkey (TUBITAK) Career Development Program (Award# 113E497)

Related Publications Spatio-Temporal Saliency Networks for Dynamic Saliency Prediction
IEEE Transactions on Multimedia
Cagdas Bak, Aysun Kocak, Erkut Erdem, Aykut Erdem
A Comparative Study for Feature Integration Strategies in Dynamic Saliency Estimation
Signal Processing: Image Communication, 51, pp. 13-25, February 2017
Yasin Kavak, Erkut Erdem, Aykut Erdem
Two-Stream Convolutional Networks for Dynamic Saliency Prediction
arXiv preprint arXiv:1607.04730
Cagdas Bak, Aykut Erdem, Erkut Erdem
Kişisel Görüntü Kümelerinin İçsel Özellikler Kullanılarak Özetlenmesi (Summarizing Personal Image Collections with Intrinsic Properties)
24. IEEE Sinyal İşleme ve İletişim Uygulamaları Kurultayı (SIU 2016), Zonguldak, Mayis 2016
Goksu Erdogan, Bora Celikkale, Aykut Erdem, Erkut Erdem
İnsan Kalabalıklarının Baskın Kümeler Tabanlı Analizi (Dominant Sets Based Analysis of Human Crowds)
24. IEEE Sinyal İşleme ve İletişim Uygulamaları Kurultayı (SIU 2016), Zonguldak, Mayis 2016
Burçak Asal, Aykut Erdem, Erkut Erdem
Baskın Kümeler ile Hareket Yörüngelerinin Öbeklenmesi (Clustering Motion Trajectories via Dominant Sets)
24. IEEE Sinyal İşleme ve İletişim Uygulamaları Kurultayı (SIU 2016), Zonguldak, Mayis 2016
Çağdaş Bak, Aykut Erdem, Erkut Erdem