Localization-Aware Active Learning for Object Detection (ACCV)

2024-04-24 12:26:22

原文链接

这是一篇CCF-C文章，简要描述一下思想：

目标检测任务的主动学习中的数据选择大多数都基于classification来，作者同时考虑了classification和localization。

主要提出了两个针对localization的方法：

1. Localization Tightness

如上图所示，计算了region proposal的输出框和最终预测框的IoU来表示Tightness：
T(B0j)=IoU(B0j,R0j)T(B_0^j)=IoU(B_0^j,R_0^j)T(B0j)=IoU(B0j,R0j)
然后再结合classification，综合得出metric:
J(B0j)=∣T(B0j)+Pmax(B0j)−1∣J(B_0^j)=|T(B_0^j)+P_{max}(B_0^j)-1|J(B0j)=∣T(B0j)+Pmax(B0j)−1∣
这里我不敢苟同作者的metric，但是至少其思路挺好的

但是上述方法只能用于two-stage网络，region proposal也就是SS or RPN的output

2.Localization Stability

如图所示，通过增强图片的高斯噪声，评价噪声图片得到的框与原始图片得到的框的差距作为Stability：
SB(B0j)=∑n=1NIoU⁡(B0j,Cn(B0j))NS_B\left(B_0^j\right)=\frac{\sum_{n=1}^N \operatorname{IoU}\left(B_0^j, C_n\left(B_0^j\right)\right)}{N} SB(B0j)=N∑n=1NIoU(B0j,Cn(B0j))
然后再结合classification，综合得出metric:
SI(Ii)=∑j=1MPmax⁡(B0j)SB(B0j)∑j=1MPmax⁡(B0j)S_I\left(I_i\right)=\frac{\sum_{j=1}^M P_{\max }\left(B_0^j\right) S_B\left(B_0^j\right)}{\sum_{j=1}^M P_{\max }\left(B_0^j\right)} SI(Ii)=∑j=1MPmax(B0j)∑j=1MPmax(B0j)SB(B0j)
也就是根据预测的score加权评价。

这个metric同时适用于two-stage和one-stage网络

Experiment

这个虽然在(a)图上看得不明显，但是(b)图以Random方法为基线比较幅度这个小trick还挺好的！！这样更可以让审稿人看出方法得优越性！

TheEnd.The\ End.The End.

Localization-Aware Active Learning for Object Detection (ACCV)相关推荐

[论文解读]Deep active learning for object detection
Deep active learning for object detection 文章目录 Deep active learning for object detection 简介摘要初步以前 ...
[论文阅读] Multiple Instance Active Learning for Object Detection
论文地址:https://openaccess.thecvf.com/content/CVPR2021/html/Yuan_Multiple_Instance_Active_Learning_for_ ...
DenseBox: Unifying Landmark Localization with End to End Object Detection
DenseBox: Unifying Landmark Localization with End to End Object Detection arxiv: http://arxiv.org/ab ...
Libra R-CNN: Towards Balanced Learning for Object Detection
论文基本信息标题: Libra R-CNN: Towards Balanced Learning for Object Detection 作者: JiangmiaoPang, Kai Chen, ...
Delving into Localization Errors for Monocular 3D Object Detection 论文学习
论文地址:Delving into Localization Errors for Monocular 3D Object Detection Github地址:Delving into Locali ...
【论文笔记】：Libra R-CNN: Towards Balanced Learning for Object Detection
&Title: Libra R-CNN: Towards Balanced Learning for Object Detection &Summary 检测不平衡问题包括:样本层面( ...
《Recent Advances in Deep Learning for Object Detection 》笔记
最近看了一篇目标检测的综述,之前对目标检测的认识不是很多,所以简单地记录一下笔记,由于是很早之前写的,对目标检测的很多概念都还不是很清楚,简单记录一下.这篇论文主要讲了目前的目标检测算法的一些设置.检 ...
论文阅读笔记五十三：Libra R-CNN: Towards Balanced Learning for Object Detection(CVPR2019)
论文原址:https://arxiv.org/pdf/1904.02701.pdf github:https://github.com/OceanPang/Libra_R-CNN 摘要相比模型的结构 ...
Week 8 Deep learning for object detection
目标检测(Object Detection)_图像算法AI的博客-CSDN博客_目标检测

最新文章

热门文章