Improved U-Net-Like Network for Visual Saliency Detection Based on Pyramid Feature Attention

Figure 1

Architecture overview of our proposed pyramid feature attention-based U-Net-like (PFAU-Net) model. CFE and CA stand for context-aware feature extraction and channel attention, respectively. Compared with the U-Net-like backbone used in [22], two main improvements are made, including (1) introducing context-aware pyramid feature module into deep layers of the encoder to pass multiscale and multireceptive field high-level features to the decoder through shortcut and (2) adding a feature pyramid fusion path for the decoder to extract multilevel semantic information related to saliency targets.