The MVS database can be downloaded from DropBox or BiaduPan, code:mvsd.

The directory structure of the MVS dataset is as follows,

└── MVS  
    ├── Video-Number  
        ├── fixation_maps
        ├── fixations
        ├── frames
        ├── saliency_maps
└── video2frame.py

To get each frame of a video, you need to run:

python video2frame.py 

Visualization of predicted saliency maps of our and other compared approaches over two mobile video clips in MVS dataset.

If you have any questions, please contact wenshijie@buaa.edu.cn