Fine-grained categorization via CNN-based automatic extraction and integration of object-level and part-level features

作者：

Highlights：

• A novel CNN-based system for fine-grained categorization

• Only class labels are available during training with no other annotations.

• We seek to interpret the hidden layer feature maps of a well-trained CNN.

• Robust object & part detection, pose estimation, boosted classification accuracy

• Ingenious use of the features learned by CNN which can find wide applications

摘要

•A novel CNN-based system for fine-grained categorization•Only class labels are available during training with no other annotations.•We seek to interpret the hidden layer feature maps of a well-trained CNN.•Robust object & part detection, pose estimation, boosted classification accuracy•Ingenious use of the features learned by CNN which can find wide applications

论文关键词：Fine-grained categorization,Part-based-features,Automatic part detection,CNN-based

论文评审过程：Received 23 June 2016, Revised 3 March 2017, Accepted 16 June 2017, Available online 24 June 2017, Version of Record 13 July 2017.

论文官网地址：https://doi.org/10.1016/j.imavis.2017.06.003