A Dual Weighting Label Assignment Scheme for Object Detection

被引:80
作者
Li, Shuai [1 ]
He, Chenhang [1 ]
Li, Ruihuang [1 ]
Zhang, Lei [1 ]
机构
[1] Hong Kong Polytech Univ, Hong Kong, Peoples R China
来源
2022 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR) | 2022年
关键词
D O I
10.1109/CVPR52688.2022.00917
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Label assignment (LA), which aims to assign each training sample a positive (pos) and a negative (neg) loss weight, plays an important role in object detection. Existing LA methods mostly focus on the design of pos weighting function, while the neg weight is directly derived from the pos weight. Such a mechanism limits the learning capacity of detectors. In this paper, we explore a new weighting paradigm, termed dual weighting (DW), to specify pos and neg weights separately. We first identify the key influential factors of pos/neg weights by analyzing the evaluation metrics in object detection, and then design the pos and neg weighting functions based on them. Specifically, the pos weight of a sample is determined by the consistency degree between its classification and localization scores, while the neg weight is decomposed into two terms: the probability that it is a neg sample and its importance conditioned on being a neg sample. Such a weighting strategy offers greater flexibility to distinguish between important and less important samples, resulting in a more effective object detector. Equipped with the proposed DW method, a single FCOS-ResNet-50 detector can reach 41.5% mAP on COCO under 1x schedule, outperforming other existing LA methods. It consistently improves the baselines on COCO by a large margin under various backbones without bells and whistles. Code is available at https://github.com/strongwolf/DW.
引用
收藏
页码:9377 / 9386
页数:10
相关论文
共 50 条
[1]  
Bochkovskiy A, 2020, ARXIV, DOI 10.48550/ARXIV.2004.10934
[2]   Large-Scale Machine Learning with Stochastic Gradient Descent [J].
Bottou, Leon .
COMPSTAT'2010: 19TH INTERNATIONAL CONFERENCE ON COMPUTATIONAL STATISTICS, 2010, :177-186
[3]   Prime Sample Attention in Object Detection [J].
Cao, Yuhang ;
Chen, Kai ;
Loy, Chen Change ;
Lin, Dahua .
2020 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2020), 2020, :11580-11588
[4]   End-to-End Object Detection with Transformers [J].
Carion, Nicolas ;
Massa, Francisco ;
Synnaeve, Gabriel ;
Usunier, Nicolas ;
Kirillov, Alexander ;
Zagoruyko, Sergey .
COMPUTER VISION - ECCV 2020, PT I, 2020, 12346 :213-229
[5]   Blind receiver for OFDM systems via sequential Monte Carlo in factor graphs [J].
Chen Rong ;
Zhang Hai-bin ;
Xu You-yun ;
Liu Xin-zhao .
JOURNAL OF ZHEJIANG UNIVERSITY-SCIENCE A, 2007, 8 (01) :1-9
[6]   Dynamic DETR: End-to-End Object Detection with Dynamic Attention [J].
Dai, Xiyang ;
Chen, Yinpeng ;
Yang, Jianwei ;
Zhang, Pengchuan ;
Yuan, Lu ;
Zhang, Lei .
2021 IEEE/CVF INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV 2021), 2021, :2968-2977
[7]  
Deng J, 2009, PROC CVPR IEEE, P248, DOI 10.1109/CVPRW.2009.5206848
[8]  
Dosovitskiy A., 2021, P 9 INT C LEARN REPR
[9]   TOOD: Task-aligned One-stage Object Detection [J].
Feng, Chengjian ;
Zhong, Yujie ;
Gao, Yu ;
Scott, Matthew R. ;
Huang, Weilin .
2021 IEEE/CVF INTERNATIONAL CONFERENCE ON COMPUTER VISION (ICCV 2021), 2021, :3490-3499
[10]  
Feng Wansen, 2019, ARXIV190607155