A novel decoder based on Bayesian rules for task‐driven object segmentation

As a challenging problem in computer vision, salient object segmentation has attracted increasing attention recent years. Though lot of works based on encoder–decoder have been made, these methods can only recognize and segment one class objects, but cannot the other classes objects same image. To address this issue, paper proposes novel decoder Bayesian rules to perform task-driven segmentation, which control signal is added determine need be segmented. What's more, rule established decoder, set as prior, latent features learned encoder transferred corresponding layer observation, thus posterior probability each with respect specific-class calculated, belonging This proposed method evaluated for several benchmark datasets, including MS COCO, DUT-OMRON, ECSSD etc. Experimental results show that approach tends accurate, detailed, complete improves performance compared previous state-of-the-art.