一周论文泛读总结20190415-20190421

1 Learning to Compare: Relation Network for Few-Shot Learning

参考

Learning to Compare: Relation Network for Few-Shot Learning 论文理解
一种相似度学习方法


将待检测图片特征与support set 图片特征连接计算相似度

2 Meta-Tracker: Fast and Robust Online Adaptation

参考

【小白笔记】目标跟踪Meta-Tracker: Fast and Robust Online Adaptation for Visual Object Trackers论文笔记

3 Person Transfer GAN to Bridge Domain Gap for Person Re-Identification

参考

【论文笔记】Person Transfer GAN to Bridge Domain Gap for Person Re-Identification
分析模型在其它数据表现较差的现象，利用GAN 网络对ReID数据集的图片迁移 ，转换数据集风格。还提出一个大型的ReID数据集MSMT17。

CycleGAN的损失

为了使变换后前景不变

4 Deep Regression Tracking with Shrinkage Loss

参考

【小白笔记】目标跟踪Deep Regression Tracking with Shrinkage Loss

ECCV 2018 DSLT:《Deep Regression Tracking with Shrinkage Loss》论文笔记
1.修改Loss降低平衡正负样本，

$$f(x)=\frac{1}{1+\exp (a(c-l))}$$

$$L_s=f(l)\ast l^2$$

2.残差连接，融合多个卷积层和特征图