BVMSOD: Bionic Vision Mechanism based Salient Object Detection

Salient object detection is proved to be important in vision tasks, and recent advances in this task is substantial, mostly benefiting from the explosive development of convolutional neural networks (CNNs). However, most of the existing methods are exploited to boost the metrics, while whether the proposed models are consistent with the human visual mechanism is rarely considered. In terms of this issue, we proposed the bionic vision mechanism based salient object detection framework, named BVMSOD. Our framework contains two branches, which have cross connection with each other to simulate the binocular vision pathway. In each single branch, there are hierarchical short connections between the features of different level layers, which is consistent with the mechanism that the human vision contains the primary vision and the advanced vision. Extensive experiments demonstrate that our method is effective and robust in various scenes. This reveals that there are numerous essential mechanisms in the bionic and brain-inspired intelligence, which could be combined with deep learning method in various vision tasks.