Transformed Saliency Dataset

==========================

Transformed Saliency Dataset

==========================

 

*DOWNLOAD LINK* 

-----------

https://drive.google.com/drive/folders/1qXVU6deYqdM2ZTyJQTxJyVWId9bTUceJ... （Google drive）

 

*DESCRIPTION*

-----------

We present a dataset of human visual attention on 2D images during scene free viewing. This dataset includes 1900 images, which are corrputed by various image transformations. This dataset is manually annotated with human eye-movement data recorded by Tobii X120 eye-tracker. This dataset provides a new benchmark to measure the robustness of saliency prediction models on various transformed scenes.

 

*SIZE* 

-----------

Total file size: 3.73GB

 

*PACKING LIST* 

-----------

├── Transformed_Saliency_Dataset

│   └── fixation

│   └── fixation_img

│   └── image

│   └── map

│   └── Scaled_Image

 

1. fixation: *.mat (Discrete human eye-movement fixation points in ".mat" format, Resolution: 1080*1920*1)

2. fixation_img: *.jpg (Discrete human eye-movement fixation points in ".jpg" format, Resolution: 1080*1920*1)

3. image: *.BMP (Source RGB stimuli, Resolution: 1080*1920*3) 

4. map: *.BMP (Density human gaze maps of source RGB stimuli, Resolution: 1080*1920*1) 

5. Scaled_Image: *.BMP (Downsampled source RGB stimuli, Resolution: 270*480*3) 

 

*Detailed Transformation Types* 

-----------

├── Image Acquisition Stage:

│   └── 2 levels of Motion Blur

│   └── 2 levels of Gaussian Noise

├── Image Transmission Stage:

│   └── 2 levels of JPEG Compression

├── Image Displaying Stage:

│   └── 2 levels of Contrast Change

│   └── 2 levels of Rotation Degree

│   └── 3 levels of Shearing Transformations

├── Other:

│   └── Inversion Transformation

│   └── Mirroring Transformation

│   └── Line Drawing

│   └── 2 levels of Cropping

 

*EYE-TRACKING SETTINGS*

-----------  

1. Total number of images: 1900

2. Scenes: indoor, outdoor, cartoon, synthetic patterns, and fractals.

3. Resolution: 1080*1920 

4. One degree of visual angle: 1 degree of horizontal visual angle corresponding to 56.91 pixels, and  1 degree of vertical visual angle corresponding to 56.55 pixels

5. Observers: 10 subjects per image, age ranging from 18 to 35 years

6. Task: free-viewing

7. Duration: 4 sec per image

8. Extra: Eye-movement data is recorded by Tobii X120 eye-tracker, 60Hz. 

 

*CITATION* 

-----------

@article{ZhaohuiTIP19,

author = {Z. Che and A. Borji and G. Zhai and X. Min and G. Guo and P. L. Callet},

title = {How is Gaze Influenced by Image Transformations? Dataset and Model},

journal = {IEEE Transactions on Image Processing},

year = {2019}

}

 

*CONTACT INFORMATION* 

-----------

Maintained by Zhaohui Che, Shanghai Jiao Tong University, Shanghai, China; 

E-mails:    chezhaohui@sjtu.edu.cn;    chezhaohuihy@gmail.com;

Last Update: 04.Oct.2019