Beyond Scalar Neuron

Beyond Scalar Neuron

论文:Beyond Scalar Neuron:Adopting V ector-Neuron Capsules for Long-Term Person Re-Identification

作者:悉尼科技大学

代码:code

摘要

背景:目前的人员再识别(reid)工作主要集中在一个人不太可能改变衣服的短期情景。然而,在长期重新定位的情况下,一个人有很大的机会改变衣服。一个复杂的再识别系统应该考虑到这样的变化。

主要贡献:

①本文提出引了一个名为Celeb-reID的大规模reid数据集。与之前的数据集不同,同一个人的Celeb-reID数据集中有不同的服装,总共有1052个id和34,186张图像,使Celeb-reID成为迄今为止最大的长期re-ID数据集。

②为了应对变装,我们建议使用矢量神经元(VN)胶囊代替传统的标量神经元(SN)来设计我们的网络。与SN相比,VN中的一个额外维度信息可以感知同一人的服装变化。我们引入了一个设计良好的ReIDCaps网络,并集成了胶囊来处理person re-ID任务。我们的网络采用了软嵌入注意(SEA)和特征稀疏表示(FSR)机制来提高性能。

模型评估:提出的长期reid数据集和两个常用的短期reid数据集进行了实验。综合分析给出了证明挑战在我们的数据集。实验结果表明,我们的ReIDCaps可以在长期情况下大大超过现有的最先进的方法

网络结构

[外链图片转存失败,源站可能有防盗链机制,建议将图片保存下来直接上传(img-1ezREWW1-1606134313680)(C:\Users\Mr.fei\AppData\Roaming\Typora\typora-user-images\image-20

### Scalar in Programming or Data Context In the context of programming and data, a scalar refers to a single value rather than a collection or array of values. Scalars are fundamental units of data that represent individual elements such as integers, floating-point numbers, characters, or booleans[^1]. In many programming languages, scalars serve as the building blocks for more complex data structures. For example, in C++, scalars can be used directly in operations involving overloaded operators. When performing arithmetic, logical, or comparison operations, scalars interact with each other based on the defined behavior of these operators. Overloaded operators allow custom behavior for standard data types, including scalars, enabling developers to define how operators should behave when applied to user-defined types. In the context of parallel computing, such as CUDA programming, scalars play a critical role in defining thread-level computations. Each thread in a parallel execution model may process a scalar value independently, contributing to the overall computation of a kernel function. The inherent thread model ensures that scalar operations are efficiently distributed across multiple threads, enhancing performance through parallelism[^3]. Additionally, in machine learning frameworks, scalars often represent individual weights or biases within neural networks. During backpropagation, scalar values are updated using gradient descent to minimize the cost function. These updates occur iteratively until convergence, ensuring that each scalar contributes optimally to the network's predictive capabilities[^2]. ```python # Example of scalar usage in Python scalar_value = 42 # A scalar integer floating_scalar = 3.14 # A scalar float # Demonstrating scalar operations result = scalar_value + floating_scalar # Adding two scalar values print(result) ```
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