Salient Contour-Aware Based Twice Learning Strategy for Saliency Detection

Abstract

Fully convolutional neural networks (FCNs) have shown outstanding performance in many computer vision tasks including salient object detection. However, most deep learning-based saliency detection models are too complicated. They cause difficulties in training. Additionally, the performance of those overly complex deep learning models is limited, and the price performance ratio of those complex models is very low. To address the problems of existing deep-learning-based methods, we introduce a new research field called saliency contour detection and design a new dataset for saliency contour detection. Inspired by the human sketching process, we propose a novel contour-aware algorithm using FCNs with a twice learning strategy for saliency detection, which imitates and dissects the process of human cognition. Extensive experimental evaluations demonstrate the effectiveness of our proposed method against other outstanding methods.