Ubuntu 16.04+cuda 9.0+cudnn 7配置Torch运行DenseCap开源代码遇到问题

1、按照Torch官网给出安装教程

执行./install.sh时遇到

error: more than one operator "==" matches these operands:
            function "operator==(const __half &, const __half &)"
            function "operator==(half, half)"
            operand types are: half == half

解决方法：命令行输入

export TORCH_NVCC_FLAGS="-D__CUDA_NO_HALF_OPERATORS__"

2、安装DenseCap相关库stnbhwd

DenseCap提供安装命令如下：

luarocks install https://raw.githubusercontent.com/qassemoquab/stnbhwd/master/stnbhwd-scm-1.rockspec

出现编译错误：

[ 25%] Building C object CMakeFiles/stn.dir/init.c.o
[ 50%] Linking C shared module libstn.so
[ 50%] Built target stn
[ 75%] Building NVCC (Device) object CMakeFiles/custn.dir/custn_generated_init.cu.o
nvcc fatal   : Value 'sm_20' is not defined for option 'gpu-architecture'
CMake Error at custn_generated_init.cu.o.cmake:207 (message):
  Error generating
  /tmp/luarocks_stnbhwd-scm-1-4197/stnbhwd/build/CMakeFiles/custn.dir//./custn_generated_init.cu.o


CMakeFiles/custn.dir/build.make:63: recipe for target 'CMakeFiles/custn.dir/custn_generated_init.cu.o' failed
make[2]: *** [CMakeFiles/custn.dir/custn_generated_init.cu.o] Error 1
CMakeFiles/Makefile2:104: recipe for target 'CMakeFiles/custn.dir/all' failed
make[1]: *** [CMakeFiles/custn.dir/all] Error 2
Makefile:127: recipe for target 'all' failed
make: *** [all] Error 2

Error: Build error: Failed building.

解决方法：

1.下载源码到本地：git clone https://github.com/qassemoquab/stnbhwd
2.修改CMakeLists.txt：
From:
IF (CUDA_FOUND)
   LIST(APPEND CUDA_NVCC_FLAGS "-arch=sm_20")
to:
IF (CUDA_FOUND)
   LIST(APPEND CUDA_NVCC_FLAGS "-arch=sm_30")
3.在源码目录里执行本地安装：luarocks make

3、运行DenseCap

cudnn报错，以为是路径的问题：

/home/ubuntu/torch/install/bin/luajit: /home/ubuntu/torch/install/share/lua/5.1/trepl/init.lua:389: /home/ubuntu/torch/install/share/lua/5.1/trepl/init.lua:389: /home/ubuntu/torch/install/share/lua/5.1/cudnn/ffi.lua:1603: 'libcudnn (R5) not found in library path.
Please install CuDNN from https://developer.nvidia.com/cuDNN
Then make sure files named as libcudnn.so.5 or libcudnn.5.dylib are placed in
your library load path (for example /usr/local/lib , or manually add a path to LD_LIBRARY_PATH)

Alternatively, set the path to libcudnn.so.5 or libcudnn.5.dylib
to the environment variable CUDNN_PATH and rerun torch.
For example: export CUDNN_PATH="/usr/local/cuda/lib64/libcudnn.so.5"

stack traceback:
        [C]: in function 'error'
        /home/ubuntu/torch/install/share/lua/5.1/trepl/init.lua:389: in function 'require'
        neural_style.lua:350: in function 'setup_gpu'
        neural_style.lua:53: in function 'main'
        neural_style.lua:601: in main chunk
        [C]: in function 'dofile'
        ...untu/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:150: in main chunk
        [C]: at 0x00405d50

命令行执行语句：

export CUDNN_PATH="/usr/local/cuda-9.0/lib64/libcudnn.so.7"

出现新提示：

Found Environment variable CUDNN_PATH = /usr/local/cuda-9.0/lib64/libcudnn.so.7/home/ubuntu/torch/install/bin/luajit: /home/ubuntu/torch/install/share/lua/5.1/trepl/init.lua:389: /home/ubuntu/torch/install/share/lua/5.1/trepl/init.lua:389: /home/ubuntu/torch/install/share/lua/5.1/cudnn/ffi.lua:1618: These bindings are for CUDNN 5.x (5005 <= cudnn.version > 6000) , while the loaded CuDNN is version: 7003
Are you using an older or newer version of CuDNN?
stack traceback:
        [C]: in function 'error'
        /home/ubuntu/torch/install/share/lua/5.1/trepl/init.lua:389: in function 'require'
        neural_style.lua:350: in function 'setup_gpu'
        neural_style.lua:53: in function 'main'
        neural_style.lua:601: in main chunk
        [C]: in function 'dofile'
        ...untu/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:150: in main chunk
        [C]: at 0x00405d50

正确解决方法（应该是直接利用luarocks安装cudnn时默认安装了5版本）：

git clone https://github.com/soumith/cudnn.torch.git -b R7 && cd cudnn.torch && luarocks make cudnn-scm-1.rockspec

可能遇到执行luarocks提示没有操作权限的问题。

解决方法：
获取root权限，然后调用/home/usrname/torch/install/bin/luarocks即可，例如：
/home/usrname/torch/install/bin/luarocks make

代码运行示例：