Multi-scale structural kernel representation for object detection

作者：

Highlights：

• The first attempt to integrate high-order statistics into deep CNNs for effective object detection.

• The proposed high-order statistical module preserves the spatial information while taking account into their special geometry structures.

• Performing favorably in comparison to the state-of-the-art methods and showing good generalization ability to other dense prediction tasks.

摘要

•The first attempt to integrate high-order statistics into deep CNNs for effective object detection.•The proposed high-order statistical module preserves the spatial information while taking account into their special geometry structures.•Performing favorably in comparison to the state-of-the-art methods and showing good generalization ability to other dense prediction tasks.

论文关键词：Object detection,High-order statistics,Polynomial kernel,Matrix power normalization

论文评审过程：Received 19 February 2020, Revised 5 August 2020, Accepted 11 August 2020, Available online 27 August 2020, Version of Record 7 September 2020.

论文官网地址：https://doi.org/10.1016/j.patcog.2020.107593