MaskCOV: A random mask covariance network for ultra-fine-grained visual categorization
作者：
Highlights：
• This paper proposes a novel random mask covariance network (MaskCOV) to study the challenging problem of ultra-fine-grained visual categorization (ultra-FGVC).
• A new self-supervised learning with a powerful in-image data augmentation scheme is proposed for enabling the network to localise those discriminative yet compact regions.
• The proposed MaskCOV focuses training within the local discriminative regions, enforcing the regional and correlation learning for subtle detail classification in the ultra-FGVC.
• Encouraging experimental results demonstrate the capability of MaskCOV in advancing research from the fine-grained to the ultra-fine-grained visual categorization.
摘要
•This paper proposes a novel random mask covariance network (MaskCOV) to study the challenging problem of ultra-fine-grained visual categorization (ultra-FGVC).•A new self-supervised learning with a powerful in-image data augmentation scheme is proposed for enabling the network to localise those discriminative yet compact regions.•The proposed MaskCOV focuses training within the local discriminative regions, enforcing the regional and correlation learning for subtle detail classification in the ultra-FGVC.•Encouraging experimental results demonstrate the capability of MaskCOV in advancing research from the fine-grained to the ultra-fine-grained visual categorization.
论文关键词：Ultra-fine-grained visual categorization,Fine-grained visual categorization,Covariance matrix,Self-supervised learning
论文评审过程：Received 31 July 2020, Revised 13 May 2021, Accepted 16 May 2021, Available online 2 June 2021, Version of Record 17 June 2021.
论文官网地址：https://doi.org/10.1016/j.patcog.2021.108067
