Detection guided deconvolutional network for hierarchical feature learning

作者：

Highlights：

• Apply object detection information as high-level guidance to train a hierarchical image representation model.

• Propose a detection-guided hierarchical learning algorithm to train a deep deconvolutional network.

• Adopt non-negative sparse regularizer to help learn more reasonable features.

• Performances on three challenging benchmark datasets outperform baselines and some related works.

摘要

Highlights•Apply object detection information as high-level guidance to train a hierarchical image representation model.•Propose a detection-guided hierarchical learning algorithm to train a deep deconvolutional network.•Adopt non-negative sparse regularizer to help learn more reasonable features.•Performances on three challenging benchmark datasets outperform baselines and some related works.

论文关键词：Image representation,Deep leaning,Object recognition

论文评审过程：Received 2 September 2014, Revised 2 January 2015, Accepted 2 February 2015, Available online 11 February 2015.

论文官网地址：https://doi.org/10.1016/j.patcog.2015.02.002