DSAL-GAN: DENOISING BASED SALIENCY PREDICTION WITH GENERATIVE ADVERSARIAL NETWORKS
1.Joint optimization of denoising and saliency prediction in a coupled end-to-end trainable GAN framework.
2. Use of cycle consistency loss to refine saliency predic-tion.
SE2Net: Siamese Edge-Enhancement Network for Salient Object Detection
1.A novel multi-stage siamese network is de-signed to jointly estimate the edges and regions from the low-level and high-level features
2.A novel edge-guided inference algorithm
SAC-Net: Spatial Attenuation Context for Salient Object Detection
1.the spatial attenuation context (SAC) module to recurrently propagate the image features over the whole feature maps with variable attenuation factors and learn to adaptively integrate the features through an attention mechanism in the module
Contour Loss: Boundary-Aware Learning for Salient Object Segmentation
1., Contour Loss The aimof Contour Loss is to guide the network to perceive the object boundaries to learn the boundary-wise distinctions between salient objects and background.
2.hierarchical global attention module (HGAM). HGAM is proposed to hierarchically attend to global contextual information for alleviating background distractions.
OGNet: Salient Object Detection with Output-guided Attention Module
- a new output-guided attention module built with outputs in various positions of an neural network
2.an intractable area loss function
Edge-guided Non-local Fully Convolutional Network for Salient Object Detection
- propose a novel edge-guided non-local fully convolutional network–embed the detailed edge information in a hierarchical manner and thus generate high-quality boundary-aware saliency maps
- an edge guidance block --incorporate the edge prior knowledge in the non-local feature learning framework
Contrast Prior and Fluid Pyramid Integration for RGBD Salient Object Detection
1.a contrast loss to utilize the contrast prior
2.a fluid pyramid integration strategy to make better use of multi-scale cross-modal features