search

imatge-upc / salgan

SalGAN: Visual Saliency Prediction with Generative Adversarial Networks
https://imatge-upc.github.io/salgan
MIT License
368 stars 106 forks source link

what is the data format for running the Salgan #16

Closed endeavorui closed 6 years ago

endeavorui commented 7 years ago

sorry to interrupt again. I am still struggle to parse the json file from SALICON dataset into the format required to run salgan. Would you please specify the parsing methods? or please clarify what is saliency map and fixations? My best guess is saliency map equals to annotations and the fixations are coordinates of the dots. If yes, what is the data structure of fixations?

Since the salgan used the 'mat' file, please help to confirm it.

endeavorui commented 7 years ago

In addition, the fixations for one image comes from multiple workers. Should I choose either one or combine them all?

junting commented 7 years ago

Hi @endeavorui,

Fixations maps are build based on fixation annotations of all workers. Saliency map is just a continuous map and is obtained by blurring the fixation map with a gaussian filter.

We were using the '.mat' file, the fixation maps were stored in this format.

Best, Junting

endeavorui commented 7 years ago

Is it possible to share with some samples of saliency maps and fixations?

junting commented 7 years ago

Hi @endeavorui ,

You can find some examples here.http://saliency.mit.edu/exampleImgs/banner.001.png