使用Fast RCNN跑通Demo

参考：http://www.cnblogs.com/louyihang-loves-baiyan/p/4885659.html

1. 下载源码
   

   git clone --recursive https://github.com/rbgirshick/fast-rcnn.git

2. 下载预训练模型
   cd fast-rcnn
   执行以下命令
   

   ./data/scripts/fetch_fast_rcnn_models.sh

   这个fetch_fast_rcnn_models.sh文件就是帮你下载模型的。
   下载下来的模型会放在
   fast-rcnn/data/fast_rcnn_models文件夹下面
   (fast-rcnn为克隆下来的根目录)
3. 配置相应的文件
   cd fast-rcnn/caffe-fast-rcnn
   执行命令
   cp Makefile.config.example Makefile.config
   然后
   sudo gedit Makefile.config进行以下修改然后保存退出
   取消注释：USE_CUDNN = 1
                        WITH_PYTHON_LAYER = 1
                        USE_PKG_CONFIG = 1
   修改路径
   INCLUDE_DIRS 应该添加上 /usr/include/hdf5/serial
   LIBRARY_DIRS 添加上 /usr/lib/x86_x64-linux-gnu/hdf5/serial
4. 编译Cython module
   cd fast-rcnn/lib
5. 编译caffe和pycaffe
   cd fast-rcnn/caffe-fast-rcnn
   make
   make的时候可能会出现如下错误：
   n@kevin:~/fast-rcnn/caffe-fast-rcnn$ make
   PROTOC src/caffe/proto/caffe.proto
   CXX .build_release/src/caffe/proto/caffe.pb.cc
   CXX src/caffe/common.cpp
   In file included from ./include/caffe/util/device_alternate.hpp:40:0,
                    from ./include/caffe/common.hpp:19,
                    from src/caffe/common.cpp:5:
   ./include/caffe/util/cudnn.hpp: In function ‘void caffe::cudnn::createPoolingDesc(cudnnPoolingStruct**, caffe::PoolingParameter_PoolMethod, cudnnPoolingMode_t*, int, int, int, int, int, int)’:
   ./include/caffe/util/cudnn.hpp:124:41: error: too few arguments to function ‘cudnnStatus_t cudnnSetPooling2dDescriptor(cudnnPoolingDescriptor_t, cudnnPoolingMode_t, cudnnNanPropagation_t, int, int, int, int, int, int)’
            pad_h, pad_w, stride_h, stride_w));
                                            ^
   ./include/caffe/util/cudnn.hpp:12:28: note: in definition of macro ‘CUDNN_CHECK’
        cudnnStatus_t status = condition; \
                               ^
   In file included from ./include/caffe/util/cudnn.hpp:5:0,
                    from ./include/caffe/util/device_alternate.hpp:40,
                    from ./include/caffe/common.hpp:19,
                    from src/caffe/common.cpp:5:
   /usr/local/cuda/include/cudnn.h:803:27: note: declared here
    cudnnStatus_t CUDNNWINAPI cudnnSetPooling2dDescriptor(
                              ^
   Makefile:501: recipe for target '.build_release/src/caffe/common.o' failed
   make: *** [.build_release/src/caffe/common.o] Error 1
   
   
   解决办法：参考：http://blog.csdn.net/errors_in_life/article/details/70916583
   进入自己根目录下原来的Caffe下面拷贝相应的文件进入caffe-fast-rcnn
   

   1.将./include/caffe/util/cudnn.hpp 换成最新版的caffe里的cudnn的实现，即相应的cudnn.hpp.

   2. 将./include/caffe/layers文件夹整个拷贝到对应的caffe-fast-rcnn里的include/caffe目录下，缺少该文件夹
   

   3.将./src/caffe/layer里的，所有以cudnn开头的文件，例如cudnn_lrn_layer.cu，cudnn_pooling_layer.cpp，cudnn_sigmoid_layer.cu。

      都替换成最新版的caffe里的相应的同名文件。
   然后重新编译，解决
   make pycaffe
6. 运行Demo文件
   cd fast-rcnn/tools
   python demo.py --net caffenet
   
   
   OK!
   
   可能出现的错误：
   LD -o .build_release/lib/libcaffe.so
   /usr/bin/ld: 找不到 -lhdf5_hl
   collect2: error: ld returned 1 exit status
   Makefile:493: recipe for target '.build_release/lib/libcaffe.so' failed
   
   
   解决办法：
   修改caffe-fast-rcnn下的Makefile.config
   

   //重要的一项将# Whatever else you find you need goes here.下面的INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/includeLIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /usr/lib修改为：INCLUDE_DIRS :=  $(PYTHON_INCLUDE) /usr/local/include /usr/include/hdf5/serialLIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /usr/lib /usr/lib/x86_64-linux-gnu /usr/lib/x86_64-linux-gnu/hdf5/serial//这是因为ubuntu16.04的文件包含位置发生了变化，尤其是需要用到的hdf5的位置，所以需要更改这一路径
   cd /usr/lib/x86_64-linux-gnu\\然后根据情况执行下面两句：sudo ln -s libhdf5_serial.so.10.1.0 libhdf5.sosudo ln -s libhdf5_serial_hl.so.10.0.2 libhdf5_hl.so