AFLNet: Adversarial focal loss network for RGB-D salient object detection

Highlights：

• We infer adversarial focal loss function for our proposed model to address the issue of foreground-background class imbalance in SOD, which can improve the detection accuracy.

• We substitute adversarial learning for pixel-wise loss to refine the saliency map, whose high-order characteristics can capture contextual information of the salient object to achieve the clear boundaries and consistent object.

• We utilize the inception model to fuse sufficiently the high-level features of RGB and depth information in deep layers. Scale adaptation, less parameters and parallel computation abilities of the inception not only reduce computational cost but also afford better fusion effect.