search

marcellacornia / sam

Predicting Human Eye Fixations via an LSTM-based Saliency Attentive Model. IEEE Transactions on Image Processing (2018)
https://ieeexplore.ieee.org/document/8400593
MIT License
208 stars 76 forks source link

use your sam-vgg outputs but nss-score is not so well #6

Closed ArchWing closed 6 years ago

ArchWing commented 6 years ago

I use your provided outputs of SALICON and I evaluate it with the MIT nss matlab source file but I get nss score=1.871215 but in paper you get 3.226 is there something I do wrong, I use SALICON2017 data,thx

marcellacornia commented 6 years ago

Hi @ArchWing, thanks for downloading our code.

In our paper, we trained and tested our network on the original SALICON dataset. The 2017 version was released about four months ago. The new ground-truth saliency maps are very different from which we used in our experiments.

Also, the results on the SALICON dataset were obtained by submitting our predicted saliency maps on the CodaLab page. For this reason, results obtained with the MIT Matlab code may be slightly different.

ArchWing commented 6 years ago

thank you so much!