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

Difference between OutputMaps and FixationMaps #25

Open hiteshnitetc opened 6 years ago

hiteshnitetc commented 6 years ago

I want to train this SalGAN network on my own dataset. I have the original images and the ground-truth images (no fixation maps). What I have presumed is that in Train section of 01-data_preprocessing.py script, pathOutputImageswill have path to the original images and pathOutputMaps will have path to ground-truth images. Is my assumption correct?

I am also confused in what I should give as input to pathToFixationMaps in the same script. Is it necessary to give all three parameters (pathOutputImages, pathOutputMaps and pathToFixationMaps) for the training purpose. Can I train this network without fixation maps?

junting commented 6 years ago

Hi @hiteshnitetc

Sure, of course, you can. SalGAN is trained with images and ground truth saliency maps. The part about fixation map is something extra that we added later for other proposes. I will remove since they were not used for the training/testing.

Best,