Saliency describe the spatial locations in an image that attract human eye gaze. There are various approaches to generate saliency map, here we try to explore the use of one such method which uses convolutional neural networks, MLNet
The model combines features extracted at different levels of the CNN and the network tries to minimize the loss function through stochastic gradient descent. The model is trained and tested on the SALICON dataset. This approach differs from other CNN based approaches in that it does not employ fully convolutional networks and extracts feature maps from different levels of the network.
Marcella Cornia, Lorenzo Baraldi, Giuseppe Serra, and Rita Cucchiara. A Deep Multi- Level Network for Saliency Prediction. In International Conference on Pattern Recog- nition (ICPR), 2016. link - https://arxiv.org/pdf/1609.01064.pdf