Caffe编译（Win10+Vs2015+Cuda8.0）

参考博客

---

1. 编译
   
   • Windows下VS2015编译caffe
   • caffe window10 安装 Caffe在window10+VS2015安装
2. 测试minist数据集
   
   • CAFFE学习笔记（一）Caffe_Example之训练mnist

环境

---

1. Win10
2. Vs2015 or Vs2013
3. cmake-3.7.2-win64-x64.zip
4. cudnn-8.0-windows10-x64-v5.1
5. Cuda8.0
6. Anaconda2-4.3.0.1-Windows-x86_64 （Python2.7 or 3.5）
7. Git-2.12.0-64-bit
8. Caffe-Windows源码

编译步骤

---

1. 安装上述所有环境
2. 配置环境变量
   
   • cmake的bin目录加入path
   • cudnn的cuda下bin目录加入path
   • Anaconda安装目录加入path（python就在根目录下）
3. 解压Caffe-Windows，最好不要含有中文路径

4. 进入…\caffe-windows\scripts，用文本编辑器如Notepad++打开build_win.cmd文件（我的目录F:\vs2015\caffe\caffe-windows\scripts\build_win.cmd），下面进行修改编译配置：
   
         我的配置如下，如果你是Vs2013，把MSVC_VERSION=14改为12。WITH_NINJA=0，除非你想用ninja编译器来进行编译。Python版本设置为2或者3，取决于装的版本，这里编译只支持2.7或者3.5版本，其他不用怎么改变。

   else (
       :: Change the settings here to match your setup
       :: Change MSVC_VERSION to 12 to use VS 2013
       if NOT DEFINED MSVC_VERSION set MSVC_VERSION=14
       :: Change to 1 to use Ninja generator (builds much faster)
       if NOT DEFINED WITH_NINJA set WITH_NINJA=0
       :: Change to 1 to build caffe without CUDA support
       if NOT DEFINED CPU_ONLY set CPU_ONLY=0
       :: Change to Debug to build Debug. This is only relevant for the Ninja generator the Visual Studio generator will generate both Debug and Release configs
       if NOT DEFINED CMAKE_CONFIG set CMAKE_CONFIG=Release
       :: Set to 1 to use NCCL
       if NOT DEFINED USE_NCCL set USE_NCCL=0
       :: Change to 1 to build a caffe.dll
       if NOT DEFINED CMAKE_BUILD_SHARED_LIBS set CMAKE_BUILD_SHARED_LIBS=0
       :: Change to 3 if using python 3.5 (only 2.7 and 3.5 are supported)
       if NOT DEFINED PYTHON_VERSION set PYTHON_VERSION=2
       :: Change these options for your needs.
       if NOT DEFINED BUILD_PYTHON set BUILD_PYTHON=1
       if NOT DEFINED BUILD_PYTHON_LAYER set BUILD_PYTHON_LAYER=1
       if NOT DEFINED BUILD_MATLAB set BUILD_MATLAB=0
       :: If python is on your path leave this alone
       if NOT DEFINED PYTHON_EXE set PYTHON_EXE=python
       :: Run the tests
       if NOT DEFINED RUN_TESTS set RUN_TESTS=0
       :: Run lint
       if NOT DEFINED RUN_LINT set RUN_LINT=0
       :: Build the install target
       if NOT DEFINED RUN_INSTALL set RUN_INSTALL=0
   )
   

5. 修改完build_win.cmd保存好后，打开cmd命令提示符，进入该文件目录，执行

   F:\vs2015\caffe\caffe-windows\scripts>build_win.cmd
   

   等待编译，先下载依赖包，然后进行caffe编译，大约20-30分钟编译成功。最后结果大致：

     CUSTOMBUILD : -- warning : gp_resolved_file_type non-absolute file 'api-ms-win-crt-string-l1-1-0.dll' returning type
   'system' -- possibly incorrect [F:\vs2015\caffe\caffe-windows\scripts\build\tools\upgrade_solver_proto_text_install_pre
   requisites.vcxproj]
     CUSTOMBUILD : warning : cannot resolve item 'api-ms-win-crt-heap-l1-1-0.dll' [F:\vs2015\caffe\caffe-windows\scripts\b
   uild\tools\upgrade_solver_proto_text_install_prerequisites.vcxproj]
     CUSTOMBUILD : -- warning : gp_resolved_file_type non-absolute file 'api-ms-win-crt-heap-l1-1-0.dll' returning type 's
   ystem' -- possibly incorrect [F:\vs2015\caffe\caffe-windows\scripts\build\tools\upgrade_solver_proto_text_install_prere
   quisites.vcxproj]
     CUSTOMBUILD : warning : cannot resolve item 'api-ms-win-crt-stdio-l1-1-0.dll' [F:\vs2015\caffe\caffe-windows\scripts\
   build\tools\upgrade_solver_proto_text_install_prerequisites.vcxproj]
     CUSTOMBUILD : -- warning : gp_resolved_file_type non-absolute file 'api-ms-win-crt-stdio-l1-1-0.dll' returning type '
   system' -- possibly incorrect [F:\vs2015\caffe\caffe-windows\scripts\build\tools\upgrade_solver_proto_text_install_prer
   equisites.vcxproj]
     CUSTOMBUILD : warning : cannot resolve item 'api-ms-win-crt-convert-l1-1-0.dll' [F:\vs2015\caffe\caffe-windows\script
   s\build\tools\upgrade_solver_proto_text_install_prerequisites.vcxproj]
     CUSTOMBUILD : -- warning : gp_resolved_file_type non-absolute file 'api-ms-win-crt-convert-l1-1-0.dll' returning type
    'system' -- possibly incorrect [F:\vs2015\caffe\caffe-windows\scripts\build\tools\upgrade_solver_proto_text_install_pr
   erequisites.vcxproj]
   
       13407 个警告
       0 个错误
   
   已用时间 00:19:12.74
   

6. 编译完成后，会在…/caffe-windows/scripts目录下多出build文件夹，里面的Caffe.sln用Vs2015打开就可以了，如果想再编译，选择Release方式。在…/caffe-windows/scripts/build/tools/Release下有编译生成的Caffe.exe,使用它便可以测试数据集。

测试mnist数据集

---

1. 下载mnist leveldb格式的数据集，包含mnist-train-leveldb以及mnist-test-leveldb。
2. 在…\caffe-windows\examples\mnist目录下，有lenet_train.prototxt、lenet_solver.prototxt和train_lenet.sh文件。先明确作用：
   
   • 在lenet_train.prototxt中：
     
     • 给出用于训练的MNIST数据mnist-train-leveldb以及mnist-test-leveldb的路径；
     • 在此定义这个网络都有哪些层，给出每一层的定义。
   • 在lenet_solver.prototxt中：
     
     • 在此指明我们要训练的网络是lenet_train.prototxt，给出lenet_train.prototxt的路径；
     • 在此指定是用CPU还是GPU。
   • 文件train_lenet.sh，在里面给出：
     
     • caffe编译后生成的可执行文件exe的名称（带不带“.exe”都可以），后面跟“train”代表选择“训练”；
     • 刚才写好的lenet_solver.prototxt的路径。
3. 在…\caffe-windows\scripts\build\tools\Release下新建一个文件夹，我的叫mnist_test。将mnist-test-leveldb、 mnist-train-leveldb 和…\caffe-windows\examples\mnist目录下的lenet_train_test.prototxt、lenet_solver.prototxt这四个文件全部拷贝到mnist_test文件夹中去。

4. 设定prototxt文件，给出相应的路径：

   • lenet_train_test.prototxt，这个文件即之前说的lenet_train.prototxt，为了便于区分在后面加了一个_test。路径改变如下(修改部分截取，未修改部分没有贴上来)：

     name: "LeNet"
     layer {
       name: "mnist"
       type: "Data"
       top: "data"
       top: "label"
       include {
         phase: TRAIN
       }
       transform_param {
         scale: 0.00390625
       }
       data_param {
         source: "mnist_test/mnist-train-leveldb"  ---------》修改部分
         batch_size: 64
         backend: LEVELDB  ---------》修改部分
       }
     }
     layer {
       name: "mnist"
       type: "Data"
       top: "data"
       top: "label"
       include {
         phase: TEST
       }
       transform_param {
         scale: 0.00390625
       }
       data_param {
         source: "mnist_test/mnist-test-leveldb"   ---------》修改部分
         batch_size: 100
         backend: LEVELDB  ---------》修改部分
       }
     }
     ......
     .....
     

   • lenet_solver.prototxt 设置相对路径：

     # The train/test net protocol buffer definition
     net: "mnist_test/lenet_train_test.prototxt"  ---------》修改部分
     # test_iter specifies how many forward passes the test should carry out.
     # In the case of MNIST, we have test batch size 100 and 100 test iterations,
     # covering the full 10,000 testing images.
     test_iter: 100
     # Carry out testing every 500 training iterations.
     test_interval: 500
     # The base learning rate, momentum and the weight decay of the network.
     base_lr: 0.01
     momentum: 0.9
     weight_decay: 0.0005
     # The learning rate policy
     lr_policy: "inv"
     gamma: 0.0001
     power: 0.75
     # Display every 100 iterations
     display: 100
     # The maximum number of iterations
     max_iter: 10000
     # snapshot intermediate results
     snapshot: 5000
     snapshot_prefix: "mnist_test/lenet"   ---------》修改部分
     # solver mode: CPU or GPU
     solver_mode: GPU    ---------》使用GPU
     

5. 打开cmd，进入caffe.exe的目录：

       C:\Users\Mr.sorrow>f:
   
       F:\>cd F:\vs2015\caffe\caffe-windows\scripts\build\tools\Release
   
       F:\vs2015\caffe\caffe-windows\scripts\build\tools\Release>
   

   运行：

       caffe.exe train --solver=mnist_test/lenet_solver.prototxt
   

   结果：

       .......
       .......
       I0309 13:46:14.446481  3708 solver.cpp:219] Iteration 9600 (165.724 iter/s, 0.603412s/100 iters), loss = 0.0029485
       I0309 13:46:14.446481  3708 solver.cpp:238]     Train net output #0: loss = 0.0029485 (* 1 = 0.0029485 loss)
       I0309 13:46:14.447450  3708 sgd_solver.cpp:105] Iteration 9600, lr = 0.00603682
       I0309 13:46:15.059074  3708 solver.cpp:219] Iteration 9700 (164.181 iter/s, 0.609083s/100 iters), loss = 0.00178553
       I0309 13:46:15.059074  3708 solver.cpp:238]     Train net output #0: loss = 0.00178552 (* 1 = 0.00178552 loss)
       I0309 13:46:15.060080  3708 sgd_solver.cpp:105] Iteration 9700, lr = 0.00601382
       I0309 13:46:15.664719  3708 solver.cpp:219] Iteration 9800 (165.979 iter/s, 0.602486s/100 iters), loss = 0.0119374
       I0309 13:46:15.665698  3708 solver.cpp:238]     Train net output #0: loss = 0.0119374 (* 1 = 0.0119374 loss)
       I0309 13:46:15.665698  3708 sgd_solver.cpp:105] Iteration 9800, lr = 0.00599102
       I0309 13:46:16.286525  3708 solver.cpp:219] Iteration 9900 (161.895 iter/s, 0.617683s/100 iters), loss = 0.00379853
       I0309 13:46:16.286990  3708 solver.cpp:238]     Train net output #0: loss = 0.00379854 (* 1 = 0.00379854 loss)
       I0309 13:46:16.287490  3708 sgd_solver.cpp:105] Iteration 9900, lr = 0.00596843
       I0309 13:46:16.890439  3708 solver.cpp:448] Snapshotting to binary proto file mnist_test/mnist/lenet_iter_10000.caffemodel
       I0309 13:46:16.903506  3708 sgd_solver.cpp:273] Snapshotting solver state to binary proto file mnist_test/mnist/lenet_iter_10000.solverstate
       I0309 13:46:16.910498  3708 solver.cpp:311] Iteration 10000, loss = 0.00267322
       I0309 13:46:16.910498  3708 solver.cpp:331] Iteration 10000, Testing net (#0)
       I0309 13:46:17.131078  7880 data_layer.cpp:73] Restarting data prefetching from start.
       I0309 13:46:17.140102  3708 solver.cpp:398]     Test net output #0: accuracy = 0.99
       I0309 13:46:17.140102  3708 solver.cpp:398]     Test net output #1: loss = 0.0297552 (* 1 = 0.0297552 loss)
       I0309 13:46:17.143108  3708 solver.cpp:316] Optimization Done.
       I0309 13:46:17.143108  3708 caffe.cpp:260] Optimization Done.
   

   在….\caffe-windows\scripts\build\tools\Release\mnist_test目录下，即生成四个文件：

   • lenet_iter_5000.caffemodel
   • lenet_iter_5000.solverstate
   • lenet_iter_10000.caffemodel
   • lenet_iter_10000.solverstate