Real-time and light-weighted unsupervised video object segmentation network

作者：

Highlights：

• A real time unsupervised video object segmentation network without manually labeled segmentation mask for the first frame.

• The Dynamic ASPP module dynamically selects features with different scales and greatly reduce the parameters.

• The RNN-Conv module can be stacked for higher level spatiotemporal patterns, preventing gradient from explosion and disappearance.

摘要

•A real time unsupervised video object segmentation network without manually labeled segmentation mask for the first frame.•The Dynamic ASPP module dynamically selects features with different scales and greatly reduce the parameters.•The RNN-Conv module can be stacked for higher level spatiotemporal patterns, preventing gradient from explosion and disappearance.

论文关键词：Unsupervised video object segmentation,Salient object detection

论文评审过程：Received 2 March 2020, Revised 10 February 2021, Accepted 17 June 2021, Available online 18 June 2021, Version of Record 10 July 2021.

论文官网地址：https://doi.org/10.1016/j.patcog.2021.108120