Temporal consistent portrait video segmentation

作者：

Highlights：

• A novel correspondence prediction paradigm for high-quality feature reconstruction.

• A PVS framework to exploit inter-frame correspondence through meta-learning.

• A PVS dataset serving as a benchmark to promote future research.

摘要

•A novel correspondence prediction paradigm for high-quality feature reconstruction.•A PVS framework to exploit inter-frame correspondence through meta-learning.•A PVS dataset serving as a benchmark to promote future research.

论文关键词：Portrait video segmentation,Meta-learning,Feature reconstruction,Feature aggregation block

论文评审过程：Received 18 September 2020, Revised 27 February 2021, Accepted 5 May 2021, Available online 30 June 2021, Version of Record 14 July 2021.

论文官网地址：https://doi.org/10.1016/j.patcog.2021.108143