A saliency dectetion method on 360-degree images. This implementation calls two deep learning methods written in Python.
Python 2.7
Theano 0.9
Keras 1.2.0 (Note: Keras changed its API in the latest version so the version must be 1.2.0)
Other miscellaneous requirements
VGG16 Weights for SalGan. Store it somewhere and specify the path in constant.py
ResNet Weights for SAM. Put it in sam/weights
Note: The codes of the 2 deep models have been modified so the codes from the original sources won't work here.
In Matlab, type:
imgLoc = 'p31.jpg';
s = predictSaliency(imgLoc);
or
imgLoc = 'p31.jpg';
s = predictSaliency(imgLoc,true); //display the saliency image
or
imgLoc = 'p31.jpg';
s = predictSaliency(imgLoc,true,true); //display and save the saliency image
This implementation is based on two deep learning methods: Saliency Attentive Model and Visual Saliency Prediction with Generative Adversarial Networks. Please refer to the original papers for more details.