Task-based parameter isolation for foreground segmentation without catastrophic forgetting using multi-scale region and edges fusion network

作者：

Highlights：

• A novel multi-scale region and edge fusion network (REFNet).

• REFNet detects the region and edges of objects to provide accurate detection.

• A novel task-based parameter isolation (TBPI) continual learning technique.

• TBPI improves the performance of any CNN-based model.

摘要

•A novel multi-scale region and edge fusion network (REFNet).•REFNet detects the region and edges of objects to provide accurate detection.•A novel task-based parameter isolation (TBPI) continual learning technique.•TBPI improves the performance of any CNN-based model.

论文关键词：Foreground segmentation,Moving objects,Deep learning,Continual learning,Parameter isolation

论文评审过程：Received 4 May 2021, Revised 25 June 2021, Accepted 29 June 2021, Available online 3 July 2021, Version of Record 9 July 2021.

论文官网地址：https://doi.org/10.1016/j.imavis.2021.104248