Saliency Prediction Based On Lightweight Attention Mechanism
Yu Wang, Ming Lei Zhu · Journal of Physics Conference Series · 2020
Abstract Saliency prediction refers to an algorithm that extracts salient regions from natural scenes. In the field of deep learning, receptive field limits the accuracy of pixel classification and thus affects the accuracy of saliency prediction. This study proposes a new saliency prediction method that uses the improved feature pyramid attention (FPA) to gain multi-scale context information to solve the above-mentioned problems. FPA’s number of parameters and cost of calculation are also decreased. Experimental results show that this method can obtain more accurate results than the existing saliency prediction methods without increasing the calculation amount.