CINet: Cascaded Interaction With Eroded Deep Supervision Strategy for Saliency Detection

Salient object detection (SOD) has garnered significant interest because of its pivotal role in numerous computer vision and graphics applications. Deep convolutional neural networks have been widely applied in salient object detection and have achieved remarkable results in this field. To enhance the network representation ability, a very important means is to increase the depth of the neural network to learn as many hierarchical features as possible. However, the information related to the input image features will be lost with increasing in network depth and existing models suffer from information distortion caused by interpolation during upsampling and downsampling. In response to this drawback, this article focuses on the feature level and label level to address this significant challenge. On the one hand, a novel cascaded interaction network with a guidance module named global-local aligned attention (GAA) is designed to reduce the negative impact of interpolation on the feature side. On the other hand, a deep supervision strategy based on edge erosion is proposed to reduce the negative guidance of label interpolation on lateral output. Extensive experiments on five popular datasets demonstrate the superiority of our method.