Agnostic attribute segmentation of dynamic scenes with limited spatio-temporal resolution
作者：
Highlights：
• We introduce an effective agnostic attribute video object segmentation method.
• For robust segmentation of dynamic scenes with limited spatio-temporal resolution.
• We extract more discriminative agnostic proposal, location and appearance features.
• Jointly impose spatio-temporal consistency constraints on the features using a CRF.
• Feasibility and superiority are demonstrated in segmenting highly dynamic scenes.
摘要
•We introduce an effective agnostic attribute video object segmentation method.•For robust segmentation of dynamic scenes with limited spatio-temporal resolution.•We extract more discriminative agnostic proposal, location and appearance features.•Jointly impose spatio-temporal consistency constraints on the features using a CRF.•Feasibility and superiority are demonstrated in segmenting highly dynamic scenes.
论文关键词：Video object segmentation,Conditional random field,Class-agnostic,Semantic space,Spatio-temporal resolution
论文评审过程：Received 2 December 2016, Revised 6 October 2017, Accepted 27 February 2019, Available online 27 February 2019, Version of Record 5 March 2019.
论文官网地址：https://doi.org/10.1016/j.patcog.2019.02.026
