MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
来源:互联网 发布:史丹利的寓言mac版 编辑:程序博客网 时间:2024/05/29 10:58
TensorFlow实现:https://github.com/Zehaos/MobileNet
caffe实现:https://github.com/pby5/MobileNet_Caffe
MobileNets用于手机或嵌入式视觉应用,提出使用depthwise separable convolutions 构造轻量化的深度神经网络,并使用两个全局超参数Width Multiplier和Resolution Multiplier减少参数数量。
模型结构
1. Depthwise Separable Convolution
标准的卷积滤波和特征组合在一步内完成,Depthwise Separable卷积将标准的卷积分解为一个depthwise卷积和一个1*1的pointwise卷积。depthwise卷积对每个输入通道用一个滤波器计算,pointwise卷积使用1*1的卷积组合depthwise的结果。这个分解可以有效降低计算量和模型大小,分解过程如下图所示:
分解前后计算量对比为,分解后计算量降低了8~9倍:
2. 网络结构
MobileNet第一层是全卷积,其余卷积层使用depthwise separable卷积构建,除最后一层,所有层与Batch Norm和ReLU衔接,最后一层输入到softmax层。MobileNet的网络结构为:
下图显示了分解卷积与常规卷积的区别:
MobileNet花费了95%的计算时间在1*1的卷积上,主要参数也在1*1卷积中,其余的参数在全连接层中,如下表所示:
3. 模型瘦身-width乘子
输入通道数乘子
4. 降低分辨率-分辨率乘子
降低输入图像或特征图的分辨率,计算量下降
实验结果
MobileNet与GoogleNet,Vgg16对比,准确度差不多,但参数少很多
物体检测结果,使用Faster-RCNN,SSD框架与VGG,Inception V2对比:
- MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
- MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
- MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
- 手机CNN网络模型--MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
- 网络小型化MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
- Reading Note: MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
- [论文阅读]MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
- MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications个人理解
- [论文解读] MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
- MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications论文笔记
- 轻量化网络:MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
- How to design DL model(1):Efficient Convolutional Neural Networks for Mobile Vision Applications
- 论文记录_MobileNets Efficient Convolutional Neural Networks for Mobile Vision Application
- 论文记录_MobileNets Efficient Convolutional Neural Networks for Mobile Vision Application
- Pruning Convolutional Neural Networks for Resource Efficient Inference代码详解
- ShuffleNet: An Extremely Efficient Convolutional Neural Network for Mobile Devices
- 《Hardware Accelerated Convolutional Neural Networks for Synthetic Vision Systems》笔记
- CNN网络量化 - Quantized Convolutional Neural Networks for Mobile Devices
- 用数组名作函数参数
- Keil uVision5 下载程序 add flash programming algorithm选项缺少需要的算法解决办法
- Java面试题全集(中)
- CodeData配置
- linux下tcpdump的使用
- MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications
- (二十三)Jmeter启动提示WARNING: Could not open/create prefs root node---学习笔记
- 移植OpenBLAS到hi3519
- Android 调用系统功能实现图片选择器,你可能会遇到的问题汇总
- tmux快捷键记录
- 嵌入式学习方法
- 判断字符是否为空的一些方法理解
- SQL 基本查询语句
- Spring的 @Autowired注解的详细解释