Context-aware network for RGB-D salient object detection
作者：
Highlights：
• We propose the CAN architecture to learn discriminative feature representations for saliency detection in RGB D images, by modeling multi modal and multi scale context dependencies within the context aware fusion and context dependent deconvolution. It is demonstrated that the proposed end to end CAN can a chieve favourable performance compared with state of the art methods.
• A context aware fusion unit based on the LSTM architecture (MCFLSTM) is developed to learn complementary contexts from two modalities. The positive effect of this fusion approach is demonstrated experimentally.
• A hierarchical LSTM structure called HSCLSTM is proposed to progressively refine saliency cues by modelling the context dependencies among different scales. Its effectiveness is also verified by experimental results.
摘要
•We propose the CAN architecture to learn discriminative feature representations for saliency detection in RGB D images, by modeling multi modal and multi scale context dependencies within the context aware fusion and context dependent deconvolution. It is demonstrated that the proposed end to end CAN can a chieve favourable performance compared with state of the art methods.•A context aware fusion unit based on the LSTM architecture (MCFLSTM) is developed to learn complementary contexts from two modalities. The positive effect of this fusion approach is demonstrated experimentally.•A hierarchical LSTM structure called HSCLSTM is proposed to progressively refine saliency cues by modelling the context dependencies among different scales. Its effectiveness is also verified by experimental results.
论文关键词：Stereoscopic saliency analysis,3D images,Multi-modal context fusion,Context-dependent deconvolution
论文评审过程：Received 19 December 2019, Revised 1 June 2020, Accepted 6 September 2020, Available online 15 September 2020, Version of Record 20 September 2020.
论文官网地址：https://doi.org/10.1016/j.patcog.2020.107630
