Selective feature fusion network for salient object detection

Fully convolutional neural networks have achieved great success in salient object detection, which the effective use of multi-layer features plays a critical role. Based on this advantage, many saliency detectors emerged recent years, and most them designed series network structures to integrate multi-level generated by backbone network. However, information different layer play roles how effectively is still challenge. In article, selective feature fusion consists module (SFM) an attention-guide hierarchical emphasis (AEM) proposed. Most previous works mainly addition concatenation, as difference, SFM adaptively selects important from input fusion, avoids introducing too much redundant information. Besides, AEM combines spatial attention channel enhance simply iteration, further improve accuracy detection. Experiments five datasets show that proposed method achieve satisfactory results when comparing other state-of-the-art detection approaches.