Document worth reading: “Salient Object Detection in the Deep Learning Era: An In-Depth Survey”

As an needed disadvantage in laptop computer imaginative and prescient, salient object detection (SOD) from pictures has been attracting an rising amount of study effort over the years. Recent advances in SOD, not surprisingly, are dominantly led by deep learning-based choices (named deep SOD) and mirrored by tons of of revealed papers. To facilitate the in-depth understanding of deep SODs, in this paper we provide an entire survey masking quite a few factors ranging from algorithm taxonomy to unsolved open factors. In particular, we first consider deep SOD algorithms from completely totally different views along with group construction, stage of supervision, learning paradigm and object/event stage detection. Following that, we summarize present SOD evaluation datasets and metrics. Then, we rigorously compile a radical benchmark outcomes of SOD methods based on earlier work, and provide detailed analysis of the comparability outcomes. Moreover, we look at the effectivity of SOD algorithms beneath completely totally different attributes, which have been barely explored beforehand, by creating a novel SOD dataset with rich attribute annotations. We extra analyze, for the first time in the topic, the robustness and transferability of deep SOD fashions w.r.t. adversarial assaults. We moreover look into the have an effect on of enter perturbations, and the generalization and hardness of present SOD datasets. Finally, we deal with quite a few open factors and challenges of SOD, and degree out potential evaluation directions in future. All the saliency prediction maps, our constructed dataset with annotations, and codes for evaluation are made publicly obtainable at https://github.com/wenguanwang/SODsurvey. Salient Object Detection in the Deep Learning Era: An In-Depth Survey