Win10 Tensorflow 配置Mask_RCNN

1、安装Anaconda3

下载地址  Anaconda 官网下载地址：https://www.continuum.io/downloads 

下载以后，点击exe程序，开始安装，详细的安装过程（图片参考：https://blog.csdn.net/u012318074/article/details/77075209）

点击 Next

点击 I Agree

选择两个钟任意一个，如果是新手，个人建议选择第二个

选择安装目录，最好不要是中文路径

这两个都要选择，这样会把Python加入环境变量

等待安装完成

2、安装tensorflow

这里安装的是tensorflow  CPU版本

安装过程：

打开开始菜单，选择

在命令行输入：

pip install tensorflow==1.6  (1.6为tensorflow版本，具体版本可自定义)

安装完成以后在命令行输入  python进入python命令行

然后输入：   import tensorflow as tf       回车，如果不报错，说明tensorflow安装成功

注：如果想安装tensorflow  GPU版本（GPU版本确实比CPU快，有条件的话推荐安装GPU版本），请先确认自己电脑上是否有GPU（打开驱动精灵或者鲁大师或者360驱动进行硬件检测，看显卡一栏，即可找到GPU），如果有，先安装cuda，然后下载对应版本的cudnn，再运行上面的命令可安装tensorflow-gpu版本

3、配置Mask_RCNN

①下载Mask_RCNN：项目源代码地址：https://github.com/matterport/Mask_RCNN

②在COCO数据集上预训练权重（mask_rcnn_coco.h5），

下载地址：https://github.com/matterport/Mask_RCNN/releases

下载完成以后放置在Mask_RCNN目录下即可

③安装pycocotools

下载COCO:

下载地址：https://github.com/philferriere/cocoapi

下载完成以后在前面打开的Anaconda命令窗口中用cd命令切换到cocoapi下的PythonAPI目录

然后在命令行输入  python setup.py build_ext --inplace          (注：在此之前确保你的电脑上已经安装了Visual Studio2015，如果Visual Studio没有C++模块，需要打开Visual Studio2015，并安装C++模块，具体安装过程可搜索相关的参考文档)

安装完成标志：

-------------------------------------------------

(E:Anaconda3install1) E:TensorflowcocoapiPythonAPI>python setup.py build_ext --inplace
running build_ext
building 'pycocotools._mask' extension
E:Visual Studio 2015installVCBINx86_amd64cl.exe /c /nologo /Ox /W3 /GL /DNDEBUG /MD -IE:Anaconda3install1libsite-packages umpycoreinclude -I../common -IE:Anaconda3install1include -IE:Anaconda3install1include "-IE:Visual Studio 2015installVCINCLUDE" "-IE:Visual Studio 2015installVCATLMFCINCLUDE" "-IC:Program Files (x86)Windows Kits10include10.0.10240.0ucrt" "-IC:Program Files (x86)Windows KitsNETFXSDK4.6.1includeum" "-IC:Program Files (x86)Windows Kits8.1include\shared" "-IC:Program Files (x86)Windows Kits8.1include\um" "-IC:Program Files (x86)Windows Kits8.1include\winrt" /Tcpycocotools/_mask.c /Fobuild emp.win-amd64-3.6Releasepycocotools/_mask.obj
_mask.c
e:anaconda3install1libsite-packages umpycoreinclude umpy py_1_7_deprecated_api.h(12) : Warning Msg: Using deprecated NumPy API, disable it by #defining NPY_NO_DEPRECATED_API NPY_1_7_API_VERSION
pycocotools/_mask.c(3070): warning C4244: “=”: 从“Py_ssize_t”转换到“siz”，可能丢失数据
pycocotools/_mask.c(3505): warning C4244: “函数”: 从“npy_intp”转换到“siz”，可能丢失数据
pycocotools/_mask.c(6935): warning C4244: “=”: 从“npy_intp”转换到“siz”，可能丢失数据
pycocotools/_mask.c(7742): warning C4244: “=”: 从“Py_ssize_t”转换到“siz”，可能丢失数据
pycocotools/_mask.c(15343): warning C4244: “初始化”: 从“double”转换到“float”，可能丢失数据
pycocotools/_mask.c(15349): warning C4244: “初始化”: 从“double”转换到“float”，可能丢失数据
E:Visual Studio 2015installVCBINx86_amd64cl.exe /c /nologo /Ox /W3 /GL /DNDEBUG /MD -IE:Anaconda3install1libsite-packages umpycoreinclude -I../common -IE:Anaconda3install1include -IE:Anaconda3install1include "-IE:Visual Studio 2015installVCINCLUDE" "-IE:Visual Studio 2015installVCATLMFCINCLUDE" "-IC:Program Files (x86)Windows Kits10include10.0.10240.0ucrt" "-IC:Program Files (x86)Windows KitsNETFXSDK4.6.1includeum" "-IC:Program Files (x86)Windows Kits8.1include\shared" "-IC:Program Files (x86)Windows Kits8.1include\um" "-IC:Program Files (x86)Windows Kits8.1include\winrt" /Tc../common/maskApi.c /Fobuild emp.win-amd64-3.6Release../common/maskApi.obj
maskApi.c
E:Visual Studio 2015installVCBINx86_amd64link.exe /nologo /INCREMENTAL:NO /LTCG /DLL /MANIFEST:EMBED,ID=2 /MANIFESTUAC:NO /LIBPATH:E:Anaconda3install1libs /LIBPATH:E:Anaconda3install1PCbuildamd64 "/LIBPATH:E:Visual Studio 2015installVCLIBamd64" "/LIBPATH:E:Visual Studio 2015installVCATLMFCLIBamd64" "/LIBPATH:C:Program Files (x86)Windows Kits10lib10.0.10240.0ucrtx64" "/LIBPATH:C:Program Files (x86)Windows KitsNETFXSDK4.6.1libumx64" "/LIBPATH:C:Program Files (x86)Windows Kits8.1libwinv6.3umx64" /EXPORT:PyInit__mask build emp.win-amd64-3.6Releasepycocotools/_mask.obj build emp.win-amd64-3.6Release../common/maskApi.obj /OUT:E:TensorflowcocoapiPythonAPIpycocotools\_mask.cp36-win_amd64.pyd /IMPLIB:build emp.win-amd64-3.6Releasepycocotools\_mask.cp36-win_amd64.lib
_mask.obj : warning LNK4197: 多次指定导出“PyInit__mask”；使用第一个规范
正在创建库 build emp.win-amd64-3.6Releasepycocotools\_mask.cp36-win_amd64.lib 和对象 build emp.win-amd64-3.6Releasepycocotools\_mask.cp36-win_amd64.exp
正在生成代码
已完成代码的生成

------------------------------------------------

然后在命令行输入：python setup.py build_ext install

-------------------------------------------------

(E:Anaconda3install1) E:TensorflowcocoapiPythonAPI>python setup.py build_ext install
running build_ext
building 'pycocotools._mask' extension
E:Visual Studio 2015installVCBINx86_amd64cl.exe /c /nologo /Ox /W3 /GL /DNDEBUG /MD -IC:UsersluoAppDataRoamingPythonPython36site-packages umpycoreinclude -I../common -IE:Anaconda3install1include -IE:Anaconda3install1include "-IE:Visual Studio 2015installVCINCLUDE" "-IE:Visual Studio 2015installVCATLMFCINCLUDE" "-IC:Program Files (x86)Windows Kits10include10.0.10240.0ucrt" "-IC:Program Files (x86)Windows KitsNETFXSDK4.6.1includeum" "-IC:Program Files (x86)Windows Kits8.1include\shared" "-IC:Program Files (x86)Windows Kits8.1include\um" "-IC:Program Files (x86)Windows Kits8.1include\winrt" /Tcpycocotools/_mask.c /Fobuild emp.win-amd64-3.6Releasepycocotools/_mask.obj
_mask.c
c:usersluoappdata oamingpythonpython36site-packages umpycoreinclude umpy py_1_7_deprecated_api.h(12) : Warning Msg: Using deprecated NumPy API, disable it by #defining NPY_NO_DEPRECATED_API NPY_1_7_API_VERSION
pycocotools/_mask.c(3070): warning C4244: “=”: 从“Py_ssize_t”转换到“siz”，可能丢失数据
pycocotools/_mask.c(3505): warning C4244: “函数”: 从“npy_intp”转换到“siz”，可能丢失数据
pycocotools/_mask.c(6935): warning C4244: “=”: 从“npy_intp”转换到“siz”，可能丢失数据
pycocotools/_mask.c(7742): warning C4244: “=”: 从“Py_ssize_t”转换到“siz”，可能丢失数据
pycocotools/_mask.c(15343): warning C4244: “初始化”: 从“double”转换到“float”，可能丢失数据
pycocotools/_mask.c(15349): warning C4244: “初始化”: 从“double”转换到“float”，可能丢失数据
E:Visual Studio 2015installVCBINx86_amd64cl.exe /c /nologo /Ox /W3 /GL /DNDEBUG /MD -IC:UsersluoAppDataRoamingPythonPython36site-packages umpycoreinclude -I../common -IE:Anaconda3install1include -IE:Anaconda3install1include "-IE:Visual Studio 2015installVCINCLUDE" "-IE:Visual Studio 2015installVCATLMFCINCLUDE" "-IC:Program Files (x86)Windows Kits10include10.0.10240.0ucrt" "-IC:Program Files (x86)Windows KitsNETFXSDK4.6.1includeum" "-IC:Program Files (x86)Windows Kits8.1include\shared" "-IC:Program Files (x86)Windows Kits8.1include\um" "-IC:Program Files (x86)Windows Kits8.1include\winrt" /Tc../common/maskApi.c /Fobuild emp.win-amd64-3.6Release../common/maskApi.obj
maskApi.c
E:Visual Studio 2015installVCBINx86_amd64link.exe /nologo /INCREMENTAL:NO /LTCG /DLL /MANIFEST:EMBED,ID=2 /MANIFESTUAC:NO /LIBPATH:E:Anaconda3install1libs /LIBPATH:E:Anaconda3install1PCbuildamd64 "/LIBPATH:E:Visual Studio 2015installVCLIBamd64" "/LIBPATH:E:Visual Studio 2015installVCATLMFCLIBamd64" "/LIBPATH:C:Program Files (x86)Windows Kits10lib10.0.10240.0ucrtx64" "/LIBPATH:C:Program Files (x86)Windows KitsNETFXSDK4.6.1libumx64" "/LIBPATH:C:Program Files (x86)Windows Kits8.1libwinv6.3umx64" /EXPORT:PyInit__mask build emp.win-amd64-3.6Releasepycocotools/_mask.obj build emp.win-amd64-3.6Release../common/maskApi.obj /OUT:buildlib.win-amd64-3.6pycocotools\_mask.cp36-win_amd64.pyd /IMPLIB:build emp.win-amd64-3.6Releasepycocotools\_mask.cp36-win_amd64.lib
_mask.obj : warning LNK4197: 多次指定导出“PyInit__mask”；使用第一个规范
正在创建库 build emp.win-amd64-3.6Releasepycocotools\_mask.cp36-win_amd64.lib 和对象 build emp.win-amd64-3.6Releasepycocotools\_mask.cp36-win_amd64.exp
正在生成代码
已完成代码的生成
running install
running build
running build_py
running install_lib
copying buildlib.win-amd64-3.6pycocotools\_mask.cp36-win_amd64.pyd -> E:Anaconda3install1Libsite-packagespycocotools
running install_egg_info
Writing E:Anaconda3install1Libsite-packagespycocotools-2.0-py3.6.egg-info

-------------------------------------------------

安装编译完成

④在命令行切换到Mask_RCNN目录下的   samples目录

然后你命令行输入：jupyter notebook   

也可以直接在开始菜单打开Jupyter Notebook，

⑤然后会发现该衬托关系打开了默认浏览器，并且打开新的页面，内容如下：

⑥然后点击  demo.ipynb，进入新页面

这是发现改程序有默认的一次运行结果，但这不影响我们后面的操作，

⑦在顶部菜单栏中选择   Cell —>  Run All

⑧注意查看下面有没有报错，如果没有报错，等待运行结果，一般报错位置在此处，可能会提示你更新numpty，此时只需要 pip install numpty，numpty会自动更新到新版本

⑨运行成功以后会出现类似第六步的效果图

⑩如果想运行自己的图片，只要打开  Mask_RCNN目录下的images目录，看到里面就有刚才运行的效果原图，可以把这些图片全部删掉，把自己的图片拷贝进去，不用修改图片名称，因为Mask_RCNN程序会自动遍历该目录下的图片，然后随机选取一张图片运行。