Search

Although the salient object detection (SOD) methods based on fully convolutional networks have made extraordinary achievements, it is still a challenge to accurately detect salient objects with complicated structure from cluttered real-world scenes due to their rarely considering the effectiveness and correlation of the captured different scale context and how to efficient interaction of complementary information. Motivate by this, in this paper, a novel Dense Multi-scale Inference Network (DMINet) is proposed for the accurate SOD task, which mainly consists of a dual-stream multi-receptive field module and a residual multi-mode interaction strategy. The former uses the well-designed different receptive field convolution operations and dense guidance connections to efficiently capture and utilize multi-scale contextual features for better salient objects inferring, while the latter adopts diverse interaction manners to adequately interact complementary information from multi-level features, generating powerful feature representations for predicting high-quality saliency maps. Quantitative and qualitative comparison results on five SOD datasets convincingly demonstrate that our DMINet performs favorably compared with 17 state-of-the-art SOD methods under different evaluation metrics. [ABSTRACT FROM AUTHOR]

Feature Pyramid Network (FPN) has been widely applied in the task of salient object detection (SOD), which has achieved great performance. However, most existing FPN-based SOD methods still have some limitations, such as insufficient guidance due to gradual dilution of semantic information, excessive computation leading to slow inference speed, and low efficiency of training models. In this paper, we design a novel Gradual Shrinkage and Cyclic Interaction Network (GSCINet) for efficient and accurate SOD, consisting of a Multi-Scale Contextual Attention Module (MSCAM) and an Adjacent Feature Shrinkage and Interaction Module (AFSIM). Specifically, the MSCAM aims at efficiently capturing multi-scale and multi-receptive-field contextual attention information through a series of well-designed convolutions and attention weight matrices of different scales to enhance the performance of initial input features. Subsequently, in AFSIM, we propose a gradual shrinkage structure and introduce a circular interaction mechanism to optimize the compressed features with less calculation cost, thereby enabling fast and accurate inference of salient objects. Extensive experimental results demonstrate the high efficiency and superiority of GSCINet against 17 state-of-the-art (SOTA) saliency detection methods under multiple evaluation metrics. [ABSTRACT FROM AUTHOR]