massens / saliency-360salient-2017 Goto Github PK

Hi,
I am trying to use your model on 360-deg images. The model predicted 936 fixation points. Is there any post-processing on the predicted fixations so to choose relevant points ?

Here the shape of the predictions :

print(preds.shape)

(936, 5)

Hi,
I'm new to both python as well as deep learning environment. I successfully implemented the code by giving an image as an input and saving the predicted scan-path output as a .mat file. Can someone kindly help me with how to visualize the scan-path over the input image?

Thanks a lot :D

2017/09/28: IMPORTANT: MILA will stop developing Theano and the next release (renamed to 1.0) will be the last main release.
http://deeplearning.net/software/theano/

It's there any kind of Plan B to finish this project?
Can it be solved by Tensorflow?

Hello !
How did you do your training exactly please ?
Because the article says the model is trained on 36 images (of Salient360 dataset), which may seem small.
Have you done an other training for SaltiNet (or an equivalent) on more images ? (to predict saliency volumes, or saliency maps).
Thank you !