Static Saliency vs. Dynamic Saliency

Introduction

In this work, we give a comprehensive comparative study for the first time of dynamic saliency (video shots) and static saliency (key frames of the corresponding video shots), and two key observations are obtained: 1) video saliency is often different from, yet quite related with, image saliency, and 2) camera motions, such as tilting, panning or zooming, affect dynamic saliency significantly. Motivated by these observations, we propose a novel camera motion and image saliency aware model for dynamic saliency prediction.

Datasets

CAMO dataset   (Download)

Hollywood dataset (Download)

CAMO videos (Download)

Dataset info

Each dataset provides the following items:

- Key frame (stimuli)

- Previous key frame (which is used to compute optical flow if applicable)

- Static fixation

- Dynamic fixation

Additional Resources

The code to extract optical flow can be found at:

http://people.csail.mit.edu/celiu/OpticalFlow/

The code used to extract features for static saliency can be found at: 

http://people.csail.mit.edu/tjudd/WherePeopleLook/index.html

Citation

Please cite the following paper if you use our datasets in your research

Tam V. Nguyen, Mengdi Xu, Guangyu Gao, Mohan S. Kankanhalli, Qi Tian, Shuicheng Yan: Static saliency vs. dynamic saliency: a comparative study. ACM Multimedia 2013: 987-996

Contact or password request:

Please drop an email to: tamnguyen@nus.edu.sg