Attention-based multi-branch multi-level salient object detection network for 360°panoramic images

Le résumé fourni par la source

To address the problems of high complexity, poor generalization, low detection accuracy and loss of detail information in current salient objects detection methods for 360°panoramic image, an attention-based multi-branch and multi-level network is proposed in this paper, which uses a multi-branch structure to complement the information among different scales and compensate for the loss of detail information. The backbone network extracts five levels of features, and performs attention computation on deep-level features from both spatial and channel perspectives to complete the semantic information processing; Multi scale feature extraction is performed on shallow-level features, followed by feature mining operations on high and low confidence regions to further refine salient features and complete detailed information processing. The detail map supervision loss function monitors the salient objects details of 360°panoramic images with a larger detail range to improve the detection effect. In this paper, the existing data sets are integrated and noised to improve the challenge of data sets. This paper has carried out sufficient experiments on the 360 SOD data set, the subjective and objective experimental results show that the performance of this method is better than the current 14 advanced methods, and it can further improve the detection accuracy of salient objects in 360°panoramic images.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.