Saliency map and saliency_fix map files could not be found

[YanDingXin]

Hi，excuse me.The saliency map and saliency_fix map files are used in the predict_dreyeve_sequence. Py project.I hope you can reply after reading it. Thank you very much.

`_ I_ff[0, :, 0, :, :] = x OF_ff[0, :, 0, :, :] = of SEG_ff[0, :, 0, :, :] = seg

Y_sal[0, 0] = read_image(join(sequence_dir, 'saliency', '{:06d}.png'.format(sample)), channels_first=False,
                         color=False, resize_dim=(h, w))
Y_fix[0, 0] = read_image(join(sequence_dir, 'saliency_fix', '{:06d}.png'.format(sample)), channels_first=False,
                         color=False, resize_dim=(h, w))

return [I_ff, I_s, I_c, OF_ff, OF_s, OF_c, SEG_ff, SEG_s, SEG_c], [Y_sal, Y_fix]

_`

those code come from predict_dreyeve_sequence.py.

[DavideA]

Hi and thank you for your interest.

What is referred to as saliency is the first version of the ground truth maps, created as described in the first workshop paper.

In the journal paper, we relied on a slightly different procedure to build a fixation map for a frame. This is what is referred to as saliency_fix in the code. Basically, everything that has the _fix postfix identifies such "new" fixation maps.

We consider the latter one (fix) the proper ground truth annotation of our dataset. You are strongly encouraged to rely on it for the evaluation of your method. The saliency version should only be used to reproduce results in our conference paper.

Edit: the fixation videos you get once you download the dataset are relative to the fix ground truth. So, you should only unroll them into the saliency_fix folder and comment out the lines loading from the saliency folder. Indeed, the variable Y_sal in the lines you mentioned should not be used for training at all.

Hope this helps, D

[YanDingXin]

Hi and thank you for your interest. What is referred to as saliency is the first version of the ground truth maps, created as described in the first workshop paper. In the journal paper, we relied on a slightly different procedure to build a fixation map for a frame. This is what is referred to as saliency_fix in the code. Basically, everything that has the _fix postfix identifies such "new" fixation maps. We consider the latter one (fix) the proper ground truth annotation of our dataset. You are strongly encouraged to rely on it for the evaluation of your method. The saliency version should only be used to reproduce results in our conference paper. Edit: the fixation videos you get once you download the dataset are relative to the fix ground truth. So, you should only unroll them into the saliency_fix folder and comment out the lines loading from the saliency folder. Indeed, the variable Y_sal in the lines you mentioned should not be used for training at all. Hope this helps, D

Well, thank you very much for your quick reply, which has helped me a lot.I will continue to try to implement this experiment.Thanks again.

[varunjammula]

Hi, how do I compute the saliency_fix images? I could not find the code to get these frames. Can you please let me know how to fix this? Thanks!