【挤压-激励-网络】Squeeze-Excitation-Networks

亮堂梦

已于 2023-10-26 17:45:40 修改

阅读量812

点赞数

文章标签：深度学习

于 2023-10-26 17:43:53 首次发布

本文链接：https://blog.csdn.net/qq_31828933/article/details/133923778

版权

标题：Squeeze-and-Excitation Networks

1.介绍

2.SE模块

（1）Squeeze:全局信息嵌入

（2）Excitation:自适应重新校正

（3）Exemplars:SE-Inception and SE-ResNet模型

类型：研究型；

内容：SE模块

GitHub：https：//githorb.com/hujie-frank/senet

标题：Squeeze-and-Excitation Networks

摘要：

Convolutional neural networks are built upon the convolution operation, which extracts informative features by fusing spatial and channel-wise information together within local receptive fields. In order to boost the representational power of a network, several recent approaches have shown the benefit of enhancing spatial encoding. In this work, we focus on the channel relationship and propose a novel architectural unit, which we term the “Squeeze- and-Excitation” (SE) block, that adaptively recalibrates channel-wise feature responses by explicitly modelling in- terdependencies between channels. We demonstrate that by stacking these blocks together, we can construct SENet architectures that generalise extremely well across challenging datasets. Crucially, we find that SE blocks produce significant performance improvements for existing state-of- the-art deep architectures at minimal additional computational cost. SENets formed the foundation of our ILSVRC 2017 classification submission which won first place and significantly reduced the top-5 error to 2.251%, achieving a ∼25% relative improvement over the winning en- try of 2016. Code and models are available at https: //github.com/hujie-frank/SENet.

卷积神经网络是建立在卷积运算基础之上的，它将空间和通道维度信息在局部感受野内融合起来提取信息特征。为了提高网络的表征能力，最近的几种方法已经表明了增强空间编码的好处。在本次的研究中，我们聚焦于通道关系并提出了一个新颖的框架单元，我们称之为“激励-挤压”（SE）模块，依据通道之间显化模型的相互依赖关系适应性地重新矫正通道维度特征响应。我们证明了通过将这些模块堆叠到一起就可以构建SE网络架构，这个架构在通过挑战性的数据集泛化性极佳。关键是，我们发现SE模块在现存的最先进的深度架构中，花费很小的额外计算量就可以达到显著的提升效果。SENets是我们ILSVRC2017分类提交的基础，以排名第一的位置显著地将前五的错误减少至2.251%，相较于2016年获得的提升了25%。

1.介绍

SE构建块的基本结构如下图所示：

对于任意给定的变换,我们就可以构造一个SE模块来重新校准，特征U首先通过挤压Squeeze操作，跨越空间维度W*H聚合特征响应来产生通道描述符。之后是激励Excitation操作，通过依赖性的自门机制为每个通道学习特征采样激活，控制每个通道的激励。然后特征映射U重新加权生成SE块输出，此输出可以直接输入到后面的层中。