Configuration for 84*84 input images about sam

[kkhetarpal]

I was able to overcome this by using shape_r_gt =11 = shape_c_gt. I get the salmap. However, it doesn't match what I get when I use the original image of 360*480. Any suggestions what I am missing ?

What would be the correct steps to ensure a correct salmap for a 84*84 image input ?
Thanks a lot in advance for your help.

[marcellacornia]

Hi @kkhetarpal,
thanks for downloading our code.

The parameters of our configuration file have been designed for images from the SALICON dataset that have all the same size of 480x640. However, our code is able to predict saliency maps for images with different sizes. In our code, we pre-process the input images to bring them to 240x320 by padding them and keeping the original aspect ratio.

So, if you want to use our network without re-training it, I suggest you use the original parameters.