学习日记2020-11-24论文学习

BASNet: Boundary-Aware Salient Object Detection

1. A novel boundary-aware salient object detection net-work: BASNet
2. A novel hybrid loss that fuses BCE, SSIM and IoU – three levels: pixel-level, patch-level and map-level,
3. 
   

Label Decoupling Framework for Salient Object Detection

1.to decompose a saliency label into body map and detail map to supervise the model, respectively.
2.FIN is designed with two branches to adapt to body map and detail map respectively.

Adaptive Graph Convolutional Network with Attention Graph Clustering for Co-saliency Detection

1. AGCN provide an adaptive graph convolutional network design to simultaneously capture the intra- and inter-image correspondence
2. AGCM an attention graph clustering module to differentiate the common objects from salient fore-grounds
3. an end-to-end computational framework with encoder-decoder CNN structure to jointly optimize the graph clustering task and the co-saliency detection objective
   
   
   
   
   

Weakly-Supervised Salient Object Detection via Scribble Annotations

1.a new weakly-supervised salient object detection method
2. introduce a new scribble based saliency dataset S-DUTS
3.we propose a gated structure-aware loss
4.design a scribble boosting scheme to expand our scribble annotations,
5. a new evaluation metric to measure

Interactive Two-Stream Decoder for Accurate and Fast Saliency Detection

1.the correlation between the saliency maps and corresponding contour maps.(FCF)
2. a lightweight Interactive Two-Stream Decoder (ITSD) – cues- the saliency and contour maps
3. an Adaptive ConTour (ACT) loss