A Lightened CNN for Deep Face Representation
来源:互联网 发布:决战武林宠物进阶数据 编辑:程序博客网 时间:2024/05/01 21:56
一个小巧的人脸识别CNN网络
https://github.com/AlfredXiangWu/face_verification_experiment
当前基于CNN网络的人脸识别的文献可以说是满天飞,虽然效果不错,但是计算量大是一个问题。导致其难以用于嵌入式设备或手机里。当前关于人脸识别的CNN网络问题如下:1)很深的CNN网络导致一个大的模型,提取特征的时间较长。2)基于 ReLU 激活函数学习到的特征往往是 high dimensional and sparse,所以采用 Joint Bayesian [19] or metric learning [12]得到一个 low-dimensional and compact representation。所以我们这里希望使用一个小的CNN网络,直接得到一个可以快速计算的 low-dimensional representation特征。
The contributions are summarized as follows:
1)一个新的激活函数用于CNN网络的卷积层, 它可以学习到 compact 特征
Max-Feature-Map (MFM) activation function
2)设计了两个小网络,One contains 4 convolution layers, 4 max-pooling layers and 2 fully connected layers and totally contains about 4M parameters, the other reduces the kernel size of convolution layers and employs Network in Network (NIN) [11] between convolution layers.
3)提出的网络,效果不错,时间短。 the CPU time of extracting face feature vector based on CNN is nearly 67ms
3 Architecture
3.1. Max-Feature-Map Activation Function
Max-Feature-Map 激活函数 受 maxout networks[4]启发,定义如下:
激活函数的梯度如下
3.2. The Lightened CNN Framework
结果:
- A Lightened CNN for Deep Face Representation
- 《A Lightened CNN for Deep Face Representation》论文解读 本文来自中科院,原文地址为: https://arxiv.org/abs/1511.02683
- 人脸识别方向论文笔记(1)-- A Light CNN for Deep Face Representation With Noisy Labels
- 【论文笔记】Learning Deep Face Representation
- Face++人脸识别:Learning Deep Face Representation
- A Discriminative Feature Learning Approach for Deep Face Recognition, ECCV16.
- ECCV A Discriminative Feature Learning Approach for Deep Face Recognition
- A Discriminative feature learning approach for deep face recognition
- A Discriminative Feature Learning Approach for Deep Face Recognition
- 阅读A Discriminative Feature Learning Approach for Deep Face Recognition
- A Discriminative Feature Learning Approach for Deep Face Recognition
- 人脸验证:Lightened CNN
- 【深度学习论文笔记】Learning Deep Face Representation
- Deep Learning Face Representation by Joint Identification-Verification
- Deep Learning Face Representation by Joint Identification-Verification
- Deep Learning Face Representation from Predicting 10,000 Classes
- DeeoID:Deep learning face representation from predicting 10,000 classes
- DeepID2:Deep Learning Face Representation by Joint Identification-Verification
- CentOS启动卡在开机进度条certmonger后
- Fabric自动化运维工具实现远程SSH部署及系统管理
- A10负载配置及主备手动切换命令
- Nginx实现TCP反向代理
- zabbix monitor mysql
- A Lightened CNN for Deep Face Representation
- Windows Server 2003 使用vSphere5.5连接ESXI5.5 “客户端无法向服务器发送完整的请求” 的解决办法
- Shanghai
- zabbix 编译安装
- Hadoop HA
- FastDFS+Nginx安装配置
- FastDFS配置和运行中出现的问题及解决(FAQ)
- MySQL之truncate表后恢复思路整理(前提是有备份且开启binlog)
- TPCC-MySQL基准测试