首页|基于关联和识别的少样本目标检测

基于关联和识别的少样本目标检测

Few-Shot Object Detection Based on Association and Discrimination

扫码查看
原文链接
  • NETL
  • NSTL
  • 万方数据
当前基于深度学习的目标检测算法已较为成熟.然而,基于少量样本检测新类仍具有挑战性,因为少样本条件下的深度学习容易导致特征空间退化.现有工作采用整体微调范式在丰富样本的基类上进行预训练,在此基础上构建新类的特征空间.然而,新类基于多个基类隐式地构造特征空间,其结构较为分散,导致基类与新类之间可分性较差.采用对新类和与其相似的基类进行关联再识别的方法进行少样本目标检测.通过引入动态感兴趣区域头,提升模型对训练样本的利用率,基于二者间的语义相似度,显式地为新类构建特征空间.通过解耦基类和新类的分类分支、添加通道注意力模块及增加边界损失函数,提升二者间的可分性.在标准PASCAL VOC数据集上的实验结果表明,所提方法的nAP50均值较TFA、MPSR及DiGeo分别提升10.2、5.4、7.8.
Deep learning-based object detection algorithms have matured considerably.However,detecting novel classes based on a limited number of samples remains challenging as deep learning can easily lead to feature space degradation under few-shot conditions.Most of the existing methods employ a holistic fine-tuning paradigm to pretrain on base classes with abundant samples and subsequently construct feature spaces for the novel classes.However,the novel class implicitly constructs a feature space based on multiple base classes,and its structure is relatively dispersed,thereby leading to poor separability between the base class and the novel class.This study proposes the method of associating a novel class with a similar base class and then discriminating each class for few-shot object detection.By introducing dynamic region of interest headers,the model improves the utilization of training samples and explicitly constructs a feature space for new classes based on the semantic similarity between the two.Furthermore,by decoupling the classification branches of the base and new classes,integrating channel attention modules,and implementing boundary loss functions,we substantially improve the separability between the classes.Experimental results on the standard PASCAL VOC dataset reveal that our method surpasses the nAP50 mean scores of TFA,MPSR,and DiGeo by 10.2,5.4,and 7.8,respectively.

few-shot object detectionassociation and discriminationdynamic region of interest headchannel attentionmargin loss

贾剑利、韩慧妍、况立群、韩方正、郑心怡、张秀权

展开 >

中北大学计算机科学与技术学院,山西 太原 030051

机器视觉与虚拟现实山西省重点实验室,山西 太原 030051

山西省视觉信息处理及智能机器人工程研究中心,山西 太原 030051

少样本目标检测 关联和识别 动态感兴趣区域头 通道注意力 边界损失

山西省科技重大专项"揭榜挂帅"项目山西省科技成果转化引导专项&&中北大学机器视觉与虚拟现实重点实验室研究基金

202201150401021202104021301055202303021211153447-110103

2024

10.3788/LOP231658

激光与光电子学进展
中国科学院上海光学精密机械研究所

激光与光电子学进展

CSTPCD北大核心
影响因子：1.153
ISSN：1006-4125
年,卷(期)：2024.61(8)
  • 32