论文阅读:轻量级reid:Learning Generalisable Omni-Scale Representations for Person Re-Identification

论文地址:
https://arxiv.org/pdf/1910.06827.pdf
Github:
https://github.com/KaiyangZhou/deep-person-reid

摘要

有效的行人再识别模型应该学习特征表示,这些特征表示既可以用于区别外观相似的人员,又可以在无需任务调整下用于跨数据集部署。

在本文中,我们提出了新的CNN架构来应对这两个挑战。首先,我们提出了一个被称为全尺度网络(OSNet)的CNN来学习特征,它不仅可以捕捉不同的空间尺度,而且可以封装多个尺度的协同组合,即全尺度特征。基本构建块由多个卷积流组成,每个卷积流检测一定范围内的特征。对于全尺度特征学习,提出了一种统一的聚合门,将多尺度特征与信道权值动态融合。OSNet是轻量级的,因为它的构建块包含分解卷积。

其次,为了改进通用特征学习,我们在OSNet中引入实例规范化层来处理跨数据集的差异。为了确定这些层在体系结构中的最佳位置,我们提出了一种有效的可微体系结构搜索算法。

大量的实验表明,在传统的相同数据集设置下,尽管OSNet比现有的re-ID模型要小得多,但它仍能实现最先进的性能。在更具挑战性和实用性的跨数据集设置中,OSNet击败了最新的无监督域自适应方法,同时并不需要任何目标数据来进行模型自适应。

2 related work

(1)Multi-Scale and Multi-Stream CNNs
基于ResNeXt,(多流相同尺度),OSNet(多流不同尺度);

建立网络在Inception [ 46 ],[47]上,其中inception多个流是最初专为降低计算成本而设计,精心设计的卷积层和池化层的混合物。在相比之下,我们的构建基块使用比例控制因子使要捕获的空间比例多样化。

在轻量化设计方面,我们的OSNet是与MobileNet相似,但凭经验修改的排序更好地为全尺度特征的学习。
(2)Domain Generalisation
unsupervised domain adaptation (UDA) methods [7], [8], [9], [10]:
DG methods [54], [55]
adapted for DG [56]
Our DG re-ID solution is based on adding instance normalisation (IN) layers [13].use neural architecture search to optimally explore the capability of IN in DG
多尺度动态融合模块思想如下:
在这里插入图片描述

3 OMNI -S CALE N ETWORK FOR P ERSON R E -ID

可分离卷积:lite模式后再加个conv1x1如图4更高效。
在这里插入图片描述
(1)主要是动态聚合多尺度特征
动态尺度融合是通过一种新型的聚合门(AG),这是一种可学习的神经网络,且 shared。
G is implemented as a mini-network composed of a non-parametric global average pooling layer [69] and a multi-layer perceptron (MLP) with one ReLU-activated hidden layer, followed by the sigmoid
activation. To reduce parameter overhead, we follow [70],[71] to reduce the MLP’s hidden dimension with a reductionratio, which is set to 16.
the AG is shared across all the feature streams in the same omni-scale residual block
在这里插入图片描述
网络结构如下:
AIN为自动搜索插入IN(instance normalisation)层的构架。(Different from batch normalisation (BN), which normalises each sample using statistics computed over a mini-batch,IN performs normalisation on each sample using its own mean and standard deviation)在这里插入图片描述
与之前resnext和inception网络的区别:

严格遵循OSNet中的多流设计指数决定的比例递增原理;
具体来说,不同的流具有不同的重新感知场,但使用相同的Lite 3×3布局,这样的设计可以更有效地覆盖各种scale;inception是原始的通过共享设计最终具有较低的计算成本具有多个流的计算,
鉴于OSNet旨在有选择地融合多个不同接收场大小的特征流为了学习全方位的功能

  • 0
    点赞
  • 3
    收藏
    觉得还不错? 一键收藏
  • 0
    评论
Deep person re-identification is the task of recognizing a person across different camera views in a surveillance system. It is a challenging problem due to variations in lighting, pose, and occlusion. To address this problem, researchers have proposed various deep learning models that can learn discriminative features for person re-identification. However, achieving state-of-the-art performance often requires carefully designed training strategies and model architectures. One approach to improving the performance of deep person re-identification is to use a "bag of tricks" consisting of various techniques that have been shown to be effective in other computer vision tasks. These techniques include data augmentation, label smoothing, mixup, warm-up learning rates, and more. By combining these techniques, researchers have been able to achieve significant improvements in re-identification accuracy. In addition to using a bag of tricks, it is also important to establish a strong baseline for deep person re-identification. A strong baseline provides a foundation for future research and enables fair comparisons between different methods. A typical baseline for re-identification consists of a deep convolutional neural network (CNN) trained on a large-scale dataset such as Market-1501 or DukeMTMC-reID. The baseline should also include appropriate data preprocessing, such as resizing and normalization, and evaluation metrics, such as mean average precision (mAP) and cumulative matching characteristic (CMC) curves. Overall, combining a bag of tricks with a strong baseline can lead to significant improvements in deep person re-identification performance. This can have important practical applications in surveillance systems, where accurate person recognition is essential for ensuring public safety.

“相关推荐”对你有帮助么?

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值