WebSaliency is a large-scale, high-quality gaze dataset of 450 freely viewed webpage images. Each  image  was viewed by 13 or 14 viewers (41 in total). The dataset is three times larger than the existing FiWI dataset that has 149 images. FiWI grouped pages manually into 3 broad categories (textual, pictorial and mixed) but is too small to produce other meaningful clusters. The size of each image is 1280 x 720. WebSaliency is large enough that it can be automatically clustered into 6 meaningful categories that can also be used for applications such as improving visual search. To ensure a uniform distribution of stimuli from across our six clusters, we selected 75 pages per cluster from among our 55k PageSegNet training examples, thus collecting 450 images in total. A total of 41 participants (19 females, 22 males; age range 17-23; with normal or corrected-to-normal vision) participated in our data collection. After calibration, participants were instructed to freely view each webpage image for 5 seconds. The order of presentation of the web pages was randomized. Eye position was measured using an EyeLink 1000 eye-tracker.

We hope you enjoy using WebSaliency!



The WebSaliency dataset contains :


If you use WebSaliency, please cite as:

title={Predicting visual attention in graphic design documents},
author={Chakraborty, Souradeep and Wei, Zijun and Kelton, Conor and Ahn, Seoyoung and Balasubramanian, Aruna and Zelinsky, Gregory J and Samaras, Dimitris},
journal={IEEE Transactions on Multimedia},



  1. You agree not to reproduce, duplicate, copy, sell, trade, resell or exploit for any commercial purposes, any portion of the images and any portion of derived data.
  2. You agree not to further copy, publish or distribute any portion of the WebSaliency dataset. Except, for internal use at a single site within the same organization it is allowed to make copies of the database.
  3. The EyeCogLab and/or CVLab @Stony Brook University reserves the right to terminate your access to the database at any time.