行人检索--Beyond triplet loss: a deep quadruplet network for person re-identification
2017-04-19 14:27
916 查看
Beyond triplet loss: a deep quadruplet network for person re-identification
CVPR2017
https://arxiv.org/abs/1704.01719
本文使用深度学习进行行人检索,侧重点主要在损失函数的改进,提出了 quadruplet loss 用于减小类内方差 和 增加类间方差
![](https://oscdn.geek-share.com/Uploads/Images/Content/202011/07/5e80eb1b446d8b38b5ecf02b07fd35f9)
上图显示,在我们新的 quadruplet loss 作用下,对于训练数据的每个类别,我们减小了同类别方差,增加异类方差。
本文的网络结构:
![](https://oscdn.geek-share.com/Uploads/Images/Content/202011/07/d3f74a8030ecf470e8c81ad7f1bb5c12)
上图红色部分是重点,加入了第二个类别的负样本。
这里的 positive pair negative pair negative pair2 三个损失函数计算值中 positive pair 是最小的。这么做减小了同类别方差 。
![](https://oscdn.geek-share.com/Uploads/Images/Content/202011/07/c28200bf1d2a9be674c17d221b83c41e)
为了增加异类方差,这里我们设计了一个自动最大阈值采样策略,
Margin-based online hard negative mining
Thus the margin threshold is self-adaptive based on the two distributions of the trained model.
通过这个采样策略,使得我们的损失函数中边界阈值自适应得到
实验结果对比:
CVPR2017
https://arxiv.org/abs/1704.01719
本文使用深度学习进行行人检索,侧重点主要在损失函数的改进,提出了 quadruplet loss 用于减小类内方差 和 增加类间方差
上图显示,在我们新的 quadruplet loss 作用下,对于训练数据的每个类别,我们减小了同类别方差,增加异类方差。
本文的网络结构:
上图红色部分是重点,加入了第二个类别的负样本。
这里的 positive pair negative pair negative pair2 三个损失函数计算值中 positive pair 是最小的。这么做减小了同类别方差 。
为了增加异类方差,这里我们设计了一个自动最大阈值采样策略,
Margin-based online hard negative mining
Thus the margin threshold is self-adaptive based on the two distributions of the trained model.
通过这个采样策略,使得我们的损失函数中边界阈值自适应得到
实验结果对比:
相关文章推荐
- 行人检索"Deep Ranking for Person Re-identification via Joint Representation Learning"
- 行人检索 - Embedding Deep Metric for Person Re-identification
- 行人检索“An Improved Deep Learning Architecture for Person Re-Identification”
- 【论文笔记】Margin Sample Mining Loss: A Deep Learning Based Method for Person Re-identification
- Paper Reading: In Defense of the Triplet Loss for Person Re-Identification
- 【Person Re-ID】In Defense of the Triplet Loss for Person Re-Identification
- A Multi-task Deep Network for Person Re-identification
- 【行人识别】Deep Transfer Learning for Person Re-identification
- 【行人识别】Deep Transfer Learning for Person Re-identification
- 【Person Re-ID】Margin Sample Mining Loss: A Deep Learning Based Method for Person Re-identification
- 1703.In Defense of the Triplet Loss for Person Re-Identification 论文阅读笔记
- 【Person Re-ID】Deep Representation Learning with Part Loss for Person Re-Identification
- 【论文笔记】In Defense of the Triplet Loss for Person Re-Identification
- 【行人识别】Deep Transfer Learning for Person Re-identification
- 行人检索 Learning a Discriminative Null Space for Person Re-identification
- 【2016_ECCV】Embedding Deep Metric for Person Re-identification: A Study Against Large Variations
- 【行人重识别】A Discriminatively Learned CNN Embedding for Person Re-identification
- [论文翻译]Pedestrian Alignment Network for Large-scale Person Re-Identification
- Deep Ranking for Person Re-identification via Joint Representation Learning
- 论文阅读:An Enhanced Deep Feature Representation for Person Re-identification