Partial Person ReID

最新推荐文章于 2024-08-08 07:53:25 发布

Sugaiiii777

最新推荐文章于 2024-08-08 07:53:25 发布

阅读量444

点赞数

分类专栏：机器学习 ReID 文章标签：深度学习神经网络机器学习

本文链接：https://blog.csdn.net/sinat_37439203/article/details/107379056

版权

本文提出了一种基于Spatial Transformer Networks (STNReID) 的部分人体重识别方法，特别是在深度学习领域的应用。由于只能使用部分图像进行匹配，部分人体重识别是一项具有挑战性的任务。Luo等人利用STNReID框架，该框架可以训练现有的整体人体图像数据集。STNReID包含两个模块：用于从整体图像采样到部分图像的几何变换网络（STN）模块，以及用于提取整体、部分和变换图像特征的重识别模块。通过观察STN模块和重识别模块之间的对抗，最终得到了针对部分人体重识别的强大STNReID模型。

摘要由CSDN通过智能技术生成

This paper proposed a method bases on Spatial Transformer Networks(STNReID) about Partial Person Re-identification(Re-ID), especially on deep learning. About Partial Person Re-ID is a challenging task in the present because we only use partial images, and then to match the images. Thus, there are few studies in this field, especially on deep learning, which have focused on matching partial person images of holistic person images. Luo et al. employed a Re-ID framework based on STNReID, which can be training as existing holistic person images datasets. STNReID has two modules, which include a spatial transformer network (STN) module and a ReID module. The STN module used for samples an affined image from the holistic image to the partial image. The ReID module extracted the features that holistic, partial, and affined images. Luo observed confrontation between the STN module and the ReID module, finally, acquire strong STNReID aim to partial ReID.

Related works——to describe the model and introduce relevant knowledge

Deep person ReID

DCNNs	DCNNs have superior performance, based on the loss functions of training DCNNs, which have two methods, robust representation learning, and deep metric learning.