https://github.com/happynear/caffe-windows
1.首先请选择cuda9.0 ，cudNN7.05，虽然按照下图有矛盾之处（cuda8.0）
2.只需要libcaffe.lib即可
3.拷贝classification.cpp文件到新工程下，将其分解为.h以及新的.cpp
4.依赖文件：
4.1 拷贝caffe-windows/include 下的caffe文件到新工程下
4.2 拷贝\caffe-windows\windows\thirdparty 到新工程下
4.3拷贝 c盘cuda9.0文件夹到新工程下，重命名为cuda
4.4新建caffe_lib文件夹，将caffe-windows编译出来的lib放到该文件下
5.工程配置
5.1 包含目录（主要thirdparty 跟 caffe-windows的caffe目录）
./include;./thirdparty;./thirdparty/Protobuf/include;./thirdparty/Boost;./thirdparty/OpenBLAS/include;./cuda/include;./thirdparty/GLog/include;./thirdparty/GFlags/Include;./caffe_lib;./thirdparty\OpenCV;./thirdparty\OpenCV\include;./thirdparty\OpenCV\include\opencv;./thirdpartyOpenCV\include\opencv2;./thirdparty\OpenBLAS\include
5.2 库目录（主要thirdparty ，cuda 跟 caffe_lib）
./thirdparty/Boost/lib64-msvc-14.0;./thirdparty/Protobuf/lib;./cuda\lib\x64;./thirdparty\GFlags\Lib;./thirdparty\GLog\lib;./thirdparty\LEVELDB\lib;./thirdparty\NCCL\lib;./thirdparty/OpenCV\x64\vc14\lib;./thirdparty\HDF5\lib;./thirdparty\LMDB\lib;./thirdparty\OpenBLAS\lib;./caffe_lib
5.3 附加依赖项目（thirdparty ，cuda ，以及 libcaffe）
glog.lib;libopenblas.dll.a;zlib.lib;szip.lib;libcaffe.lib;gflags_nothreadsd.lib;libboost_chrono-vc140-mt-1_65_1.lib;libboost_chrono-vc140-mt-gd-1_65_1.lib;libboost_date_time-vc140-mt-1_65_1.lib;libboost_date_time-vc140-mt-gd-1_65_1.lib;libboost_filesystem-vc140-mt-1_65_1.lib;libboost_filesystem-vc140-mt-gd-1_65_1.lib;libboost_python-vc140-mt-1_65_1.lib;libboost_python-vc140-mt-gd-1_65_1.lib;libboost_python3-vc140-mt-1_65_1.lib;libboost_python3-vc140-mt-gd-1_65_1.lib;libboost_system-vc140-mt-1_65_1.lib;libboost_system-vc140-mt-gd-1_65_1.lib;libboost_thread-vc140-mt-1_65_1.lib;libboost_thread-vc140-mt-gd-1_65_1.lib;libprotobuf.lib;libprotobuf-lite.lib;libprotoc.lib;cublas.lib;cudart.lib;cuda.lib;cufft.lib;curand.lib;cudnn.lib;gflags.lib;gflagsd.lib;gflags_nothreads.lib;hdf5.lib;hdf5_hl.lib;Shlwapi.lib;LevelDb.lib;lmdb.lib;opencv_world310.lib;
6 其他头文件："layerhead.h"
namespace caffe
{
extern INSTANTIATE_CLASS(InputLayer);
//REGISTER_LAYER_CLASS(Input);
extern INSTANTIATE_CLASS(InnerProductLayer);
//REGISTER_LAYER_CLASS(InnerProduct);
extern INSTANTIATE_CLASS(DropoutLayer);
//REGISTER_LAYER_CLASS(Dropout);
extern INSTANTIATE_CLASS(ConvolutionLayer);
REGISTER_LAYER_CLASS(Convolution);
extern INSTANTIATE_CLASS(BatchNormLayer);
//REGISTER_LAYER_CLASS(BatchNorm);
extern INSTANTIATE_CLASS(ReLULayer);
REGISTER_LAYER_CLASS(ReLU);
extern INSTANTIATE_CLASS(PoolingLayer);
REGISTER_LAYER_CLASS(Pooling);
extern INSTANTIATE_CLASS(LRNLayer);
REGISTER_LAYER_CLASS(LRN);
extern INSTANTIATE_CLASS(SoftmaxLayer);
REGISTER_LAYER_CLASS(Softmax);
extern INSTANTIATE_CLASS(EltwiseLayer);
//REGISTER_LAYER_CLASS(Eltwise);
extern INSTANTIATE_CLASS(PowerLayer);
//REGISTER_LAYER_CLASS(Power);
extern INSTANTIATE_CLASS(SplitLayer);
//REGISTER_LAYER_CLASS(Split);
extern INSTANTIATE_CLASS(ScaleLayer);
//REGISTER_LAYER_CLASS(Scale);
extern INSTANTIATE_CLASS(BiasLayer);
//REGISTER_LAYER_CLASS(Bias);
extern INSTANTIATE_CLASS(ReshapeLayer);
//extern INSTANTIATE_CLASS(ROIPoolingLayer);
}
7.测试代码
//std::cout << "" << std::endl;
//深度学习初始化
cv::Mat img(32, 32, CV_8UC3);
string resnet_model_file = "models/ResNet56/res56_cifar_train_test.prototxt";
string resnet_trained_file = "models/ResNet56/cifar10_res56_iter_120000.caffemodel";
string resnet_mean_file = "models/ResNet56/mean.binaryproto";
string resnet_label_file = "models/ResNet56/labels.txt";
Classifier classifier(resnet_model_file, resnet_trained_file, resnet_mean_file, resnet_label_file);
std::cout << ""<<std::endl;
img = cv::imread("0.jpg");
std::vector<Prediction> predictions = classifier.Classify(img);
IplImage Ip_img = IplImage(img);
string str1 = predictions[0].first;
string str2;
std::cout << str1.c_str() << predictions[0].second << std::endl;
8.结果
