NCMS: Towards accurate anchor free object detection through ℓ2 norm calibration and multi-feature selection

作者：

Highlights：

• A novel norm calibration operation to balance the norm of feature pyramid network.

• A multi-feature selection strategy to boost the anchor free object detector.

• We achieved single-model single-scale AP 44% on the COCO test dataset.

摘要

•A novel norm calibration operation to balance the norm of feature pyramid network.•A multi-feature selection strategy to boost the anchor free object detector.•We achieved single-model single-scale AP 44% on the COCO test dataset.

论文关键词：Object detection,Norm calibration,Feature selection

论文评审过程：Received 4 December 2019, Revised 22 May 2020, Accepted 22 July 2020, Available online 27 July 2020, Version of Record 8 August 2020.

论文官网地址：https://doi.org/10.1016/j.cviu.2020.103050