Salient Object Detection in the Deep Learning Era： An In-depth Survey

https://arxiv.org/pdf/1904.09146.pdf

• Summary of research on evaluating the salience of images using Deep Learning
• 

The rest of the paper is organized as follows. Section 2 explains the proposed taxonomies, each accompanied with one or two most representative models. Section 3 examines the most notable SOD datasets, whereas Section 4 describes several widely used SOD metrics. Section 5 benchmarks several deep SOD models and provides in-depth analyses. Section 6 provides further discussions and presents open issues and future research directions of the field. Finally, Section 7 concludes the paper.

• Classification based on the architecture of Neural Network
  • There are pixel-based methods and CNN-based methods.
• Classification based on the level of supervision
  • Annotation-based methods, such as human gaze.
  • Weak methods (highlighting the areas of objects in the middle stages of datasets like ImageNet)#weakly supervised learning
    • Visual Complexity Analysis Using Deep Intermediate-Layer Features seems relevant.
• Additionally, there is a classification of Single Task Learning vs Multi Task Learning (MTL).