caffe实现年龄及性别预测

一、相关代码及训练好的模型

eveningglow/age-and-gender-classification: Age and Gender Classification using Convolutional Neural Network  https://github.com/eveningglow/age-and-gender-classification

二、部署

1、打开Caffe.sln工程，编译方法见：https://www.cnblogs.com/smbx-ztbz/p/9380273.html

2、将相关源文件及模型拷贝至如下目录：

3、在examples中新建工程，且将对应源码添加进来

4、属性设置：

（1）进入“C/C++”，选中“常规”，“附加包含目录”输入如下：

D:Projectscaffe_gpucaffeuildinclude
D:Projectscaffe_gpucaffeuild
C:Users	ingpan.caffedependencieslibraries_v140_x64_py35_1.1.0librariesincludeoost-1_61
C:Users	ingpan.caffedependencieslibraries_v140_x64_py35_1.1.0librariesinclude
C:Program FilesNVIDIA GPU Computing ToolkitCUDAv8.0include
C:Users	ingpan.caffedependencieslibraries_v140_x64_py35_1.1.0librariesincludeopencv
D:Projectscaffe_gpucaffeinclude
C:Users	ingpan.caffedependencieslibraries_v140_x64_py35_1.1.0librariesInclude

其中tingpan改成自己电脑的用户名。

（2） “C/C++” –>“预处理器”—> “预处理器定义”， 输入如下：

WIN32
_WINDOWS
NDEBUG
CAFFE_VERSION=1.0.0
BOOST_ALL_NO_LIB
USE_LMDB
USE_LEVELDB
USE_CUDNN
USE_OPENCV
CMAKE_WINDOWS_BUILD
GLOG_NO_ABBREVIATED_SEVERITIES
GOOGLE_GLOG_DLL_DECL=__declspec(dllimport)
GOOGLE_GLOG_DLL_DECL_FOR_UNITTESTS=__declspec(dllimport)
H5_BUILT_AS_DYNAMIC_LIB=1
CMAKE_INTDIR="Release"

（3）“链接器” –>”输入” –>“附加依赖项”

kernel32.lib
user32.lib
gdi32.lib
winspool.lib
shell32.lib
ole32.lib
oleaut32.lib
uuid.lib
comdlg32.lib
advapi32.lib
D:Projectscaffe_gpucaffeuildinstalllibcaffe.lib
D:Projectscaffe_gpucaffeuildinstalllibcaffeproto.lib
C:Users	ingpan.caffedependencieslibraries_v140_x64_py35_1.1.0librariesliboost_system-vc140-mt-1_61.lib
C:Users	ingpan.caffedependencieslibraries_v140_x64_py35_1.1.0librariesliboost_thread-vc140-mt-1_61.lib
C:Users	ingpan.caffedependencieslibraries_v140_x64_py35_1.1.0librariesliboost_filesystem-vc140-mt-1_61.lib
C:Users	ingpan.caffedependencieslibraries_v140_x64_py35_1.1.0librarieslibglog.lib
C:Users	ingpan.caffedependencieslibraries_v140_x64_py35_1.1.0librariesLibgflags.lib
shlwapi.lib
C:Users	ingpan.caffedependencieslibraries_v140_x64_py35_1.1.0librariesliblibprotobuf.lib
C:Users	ingpan.caffedependencieslibraries_v140_x64_py35_1.1.0librarieslibcaffehdf5_hl.lib
C:Users	ingpan.caffedependencieslibraries_v140_x64_py35_1.1.0librarieslibcaffehdf5.lib
C:Users	ingpan.caffedependencieslibraries_v140_x64_py35_1.1.0librariescmake..libcaffezlib.lib
C:Users	ingpan.caffedependencieslibraries_v140_x64_py35_1.1.0librariesliblmdb.lib
ntdll.lib
C:Users	ingpan.caffedependencieslibraries_v140_x64_py35_1.1.0librarieslibleveldb.lib
C:Users	ingpan.caffedependencieslibraries_v140_x64_py35_1.1.0librariescmake..liboost_date_time-vc140-mt-1_61.lib
C:Users	ingpan.caffedependencieslibraries_v140_x64_py35_1.1.0librariescmake..liboost_filesystem-vc140-mt-1_61.lib
C:Users	ingpan.caffedependencieslibraries_v140_x64_py35_1.1.0librariescmake..liboost_system-vc140-mt-1_61.lib
C:Users	ingpan.caffedependencieslibraries_v140_x64_py35_1.1.0librarieslibsnappy_static.lib
C:Users	ingpan.caffedependencieslibraries_v140_x64_py35_1.1.0librarieslibcaffezlib.lib
C:Program FilesNVIDIA GPU Computing ToolkitCUDAv8.0libx64cudart.lib
C:Program FilesNVIDIA GPU Computing ToolkitCUDAv8.0libx64curand.lib
C:Program FilesNVIDIA GPU Computing ToolkitCUDAv8.0libx64cublas.lib
C:Program FilesNVIDIA GPU Computing ToolkitCUDAv8.0libx64cudnn.lib
C:Users	ingpan.caffedependencieslibraries_v140_x64_py35_1.1.0librariesx64vc14libopencv_highgui310.lib
C:Users	ingpan.caffedependencieslibraries_v140_x64_py35_1.1.0librariesx64vc14libopencv_imgcodecs310.lib
C:Users	ingpan.caffedependencieslibraries_v140_x64_py35_1.1.0librariesx64vc14libopencv_imgproc310.lib
C:Users	ingpan.caffedependencieslibraries_v140_x64_py35_1.1.0librariesx64vc14libopencv_core310.lib
C:Users	ingpan.caffedependencieslibraries_v140_x64_py35_1.1.0librariesliblibopenblas.dll.a
C:Users	ingpanAppDataLocalProgramsPythonPython35libspython35.lib
C:Users	ingpan.caffedependencieslibraries_v140_x64_py35_1.1.0librariesliboost_python-vc140-mt-1_61.lib

去掉勾选 “从父级或项目默认设置继承”。其中tingpan改成自己电脑的用户名。

（4）将D:Projectscaffe_gpucaffeuildinstallin添加到环境变量。

5、编译

如果出现一些错误，提示缺少dll库文件，则从C:Users ingpan.caffedependencieslibraries_v140_x64_py35_1.1.0librariesx64vc14in或C:Users ingpan.caffedependencieslibraries_v140_x64_py35_1.1.0librariesin中拷贝对应的dll文件到D:Projectscaffe_gpucaffeuildinstallin目录下。

6、测试

参数输入：

model/deploy_gender2.prototxt model/gender_net.caffemodel model/deploy_age2.prototxt model/age_net.caffemodel model/mean.binaryproto img/0008.jpg

输出结果如下：

7、说明

deploy_age2网络结构

deploy_gender2网络结构

性别估计和年龄估计使用的是相同的网络结构，不同之处在于年龄估计fc8层的输出个数为8，而年龄估计的输出个数为2。