NeuralTalk2---自动产生图片的语句描述

一、安装torch

This code is written in Lua and requires  Torch

. If you're on Ubuntu, installing Torch in your home directory may look something like:

$ curl -s https://raw.githubusercontent.com/torch/ezinstall/master/install-deps | bash
$ git clone https://github.com/torch/distro.git ~/torch --recursive
$ cd ~/torch

$ ./install.sh  # and enter "yes" at the end to modify your bashrc 

$ source ~/.bashrc
如果上述方式不行,可以试如下方法：

# 1 .Install curl

sudo apt-get install curl 

# 2.Install the basic package dependencies that LuaJIT and Torch require.

 curl -s https://raw.githubusercontent.com/torch/ezinstall/master/install-deps | bash 

# 3.Clone Torch to ~/sfw/torch, this will installs LuaJIT, LuaRocks, and then uses LuaRocks (the lua package manager) to install core packages like torch, nn and paths, as well as a few other packages.

git clone https://github.com/torch/distro.git ~/sfw/torch --recursive 

# 4.Install Torch

cd ~/sfw/torch; ./install.sh 

# 5.Refresh your env variables

source ~/.bashrc

这个没什么说的，搞定以后，你会在你的额bashrc文件下看到类似这样的一行:

. /home/work-station/torch/install/bin/torch-activate

卸载torch

rm -rf ~/torch

二、安装需要的包

See the Torch installation documentation for more details. After Torch is installed we need to get a few more packages usingLuaRocks (which already came with the Torch install). In particular:

$ luarocks install nn
$ luarocks install nngraph 
$ luarocks install image 

三、安装cjson

http://www.kyne.com.au/~mark/software/download/lua-cjson-2.1.0.tar.gz 下载文件

$  tar -xzvf lua-cjson- 2.1 . 0 .tar.gz 

$  cd lua-cjson- 2.1 . 0  

$  luarocks make

四、安装cuda（这里是6.5），可以自行换成你想安装的版本

******************************************************************************************************

 CUDA6.5安装

******************************************************************************************************

CUDA6.5（ https://developer.nvidia.com/cuda-downloads ）

$ sudo chmod +x ./ cuda_6.5.14_linux_64.run

将CUDA6.5解压成三个文件($sudo sh cuda*.run --noexec --target .)（.前面有空格）

分别为：

CUDA安装包: cuda-linux64-rel-6.5.14-18749181.run

NVIDIA驱动: NVIDIA-Linux-x86_64-340.29.run

SAMPLE包: cuda-samples-linux-6.5.14-18745345.run

(这里就不安装NVIDIA驱动)

开始安装CUDA，

$ sudo ./cuda-linux64-rel-6.5.14-18749181.run

安装完成后需要在/etc/profile中

①添加环境变量,

PATH=/usr/local/cuda-6.5/bin:$PATH$ export PATH

LD_LIBRARY_PATH=/usr/local/cuda-6.5/lib64:$LD_LIBRARY_PATH$ export LD_LIBRARY_PATH

保存后, 执行下列命令, 使环境变量立即生效

$source /etc/profile

②添加lib路径，

在/etc/ld.so.conf.d/新建文件cuda.conf，并编辑

cd /etc/ld.so.conf.d

$ sudo touch cuda.conf 

$ sudo gedit cuda.conf

/usr/local/cuda-6.5/lib64

执行下列命令使之立刻生效

$ sudo ldconfig

继续，安装CUDA SAMPLE，安装下列依赖包

$sudo apt-get install freeglut3-dev build-essential libx11-dev libxmu-dev libxi-dev libgl1-mesa-glx libglu1-mesa libglu1-mesa-dev

然后用下述命令安装sample文件

$sudo ./cuda-samples-linux-6.5.14-18745345.run

完成后编译Sample文件,更若干分钟后

$ cd /usr/local/cuda-6.5/samples

$ sudo make

 编译完成后， 进入 samples/bin/x86_64/linux/release

$ sudo ./deviceQuery 

如果出现下列显卡信息，则驱动及cuda安装成功：

./deviceQuery Starting...

 CUDA Device Query (Runtime API) version (CUDART static linking)

Detected 1 CUDA Capable device(s)

Device 0: "GeForce GTX 750 Ti"
 CUDA Driver Version / Runtime Version          6.5 / 6.5
 CUDA Capability Major/Minor version number:    5.0
 Total amount of global memory:                 2048 MBytes (2147155968 bytes)
 ( 5) Multiprocessors, (128) CUDA Cores/MP:     640 CUDA Cores
 GPU Clock rate:                                1084 MHz (1.08 GHz)
 Memory Clock rate:                             2700 Mhz
 Memory Bus Width:                              128-bit
 L2 Cache Size:                                 2097152 bytes
 Maximum Texture Dimension Size (x,y,z)         1D=(65536), 2D=(65536, 65536), 3D=(4096, 4096, 4096)
 Maximum Layered 1D Texture Size, (num) layers  1D=(16384), 2048 layers
 Maximum Layered 2D Texture Size, (num) layers  2D=(16384, 16384), 2048 layers
 Total amount of constant memory:               65536 bytes
 Total amount of shared memory per block:       49152 bytes
 Total number of registers available per block: 65536
 Warp size:                                     32
 Maximum number of threads per multiprocessor:  2048
 Maximum number of threads per block:           1024
 Max dimension size of a thread block (x,y,z): (1024, 1024, 64)
 Max dimension size of a grid size    (x,y,z): (2147483647, 65535, 65535)
 Maximum memory pitch:                          2147483647 bytes
 Texture alignment:                             512 bytes
 Concurrent copy and kernel execution:          Yes with 1 copy engine(s)
 Run time limit on kernels:                     Yes
 Integrated GPU sharing Host Memory:            No
 Support host page-locked memory mapping:       Yes
 Alignment requirement for Surfaces:            Yes
 Device has ECC support:                        Disabled
 Device supports Unified Addressing (UVA):      Yes
 Device PCI Bus ID / PCI location ID:           1 / 0
 Compute Mode:
    < Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >

deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 6.5, CUDA Runtime Version = 6.5, NumDevs = 1, Device0 = GeForceGTX 750 Ti
Result = PASS

五、安装cuda相关的包

Then get the cutorch and cunnpackages:

$ luarocks install cutorch
$ luarocks install cunn

六、安装cudnn

https://developer.nvidia.com/cudnn 下载cudnn的安装包

tar -xzvf cudnn-6.5-linux-R1.tgz
cd cudnn-6.5-linux-R1 

sudo cp lib* /usr/local/cuda/lib64/ 

sudo cp cudnn.h /usr/local/cuda/include/

执行后发现还是找不到库， 报错

error while loading shared libraries: libcudnn.so.6.5: cannot open shared objectfile: No such file or directory

而lib文件夹是在系统路径里的，用ls -al发现是文件权限的问题，因此用下述命令先删除软连接

cd /usr/local/cuda/lib64/ 

sudo rm -rf libcudnn.so libcudnn.so.6.5

然后修改文件权限，并创建新的软连接

sudo chmod u=rwx,g=rx,o=rx libcudnn.so.6.5.18

sudo ln -sf libcudnn.so.6.5.18 libcudnn.so.6.5

sudo ln -sf libcudnn.so.6.5 libcudnn.so

sudo ldconfig

以下是装过caffe后出现的问题：

我当时就是在这个位置出问题的，我的机子之前安装caffe的时候安装过cuda和cudnn，我把所有的都安装完，开始跑示例的时候，这个时候出错了：

These bindings are for version 4005 or above, while the loaded CuDNN is version: 4004 
Are you using an older version of CuDNN?

意思很明显，就是cudnn版本低了 
好吧，开始版本升级中，中间参照了这个帖子，但是依然没解决我的问题，最后终于让我搞定了，特此记录

Step 1：删除/usr/local/lib/下的所有以libcudnn开头的so文件和链接文件，比如libcudnn.so.7.0.58 
Step 2：下载最新版本的cudnn，下载地址最后给出，首先要注册，过几天Nvidia才会给你下载权限 
Step 3：假设你已经下载了最新版本的cudnn，假设是* cudnn-7.0-linux-x64-v4.0-prod.tgz*，不说了解压，进入cuda目录，执行:

$ sudo cp include/* /usr/local/cuda/include/

$sudo cp lib64/* /usr/local/cuda/lib64/ 
搞定

这是cudnn与torch相关的内容（具体是什么我也不清楚）

https://github.com/soumith/cudnn.torch

下载cudnn.torch 

cd cudnn.torch

luarocks make cudnn-scm-1.rockspec

六、安装loadcaffe

$ sudo apt-get install libprotobuf-dev protobuf-compiler 

$ luarocks install loadcaffe

七、安装torch-hdf5和h5py

安装 torch-hdf5

sudo apt-get install libhdf5-serial-dev hdf5-tools
git clone git@github.com:deepmind/torch-hdf5.git 
#如果上面这句代码下载有问题，可以换成这句代码git clone https://github.com/deepmind/torch-hdf5
cd torch-hdf5
luarocks make hdf5-0-0.rockspec LIBHDF5_LIBDIR="/usr/lib/x86_64-linux-gnu/"

安装h5py

已经pip的情况下：

$ sudo pip install h5py

（可跳过）未能安装pip的情况下：

    
    

     
     $ sudo apt-get install python-pip python-dev build-essential 
    
    
    
    

     
     $ sudo pip install --upgrade pip 
    
    
    
    

     
     $ sudo pip install --upgrade virtualenv 
    
    
    
    

     
     
$ sudo pip install h5py
（可跳过）如果pip安装不成功可以试试下面安装pip的方法：
    
    
wget https://bootstrap.pypa.io/get-pip.py  --no-check-certificate
sudo python get-pip.py
八、下载已有模型，对数据进行预测
模型下载地址（具体下载地址：http://cs.stanford.edu/people/karpathy/neuraltalk2/checkpoint_v1.zip），下载解压就可以
准备好你的图片，之后执行：
 th eval.lua -model /path/to/model -image_folder /path/to/image/directory -num_images 10 
有可能出现如下问题：
user@user-XPS-8500:~/neural-style$ th neural_style.lua -gpu 0 -backend cudnn
nil 
/home/user/torch/install/bin/luajit: /home/user/torch/install/share/lua/5.1/trepl/init.lua:384: /home/ben/torch/install/share/lua/5.1/trepl/init.lua:384: /home/user/torch/install/share/lua/5.1/cudnn/ffi.lua:1279: 'libcudnn (R4) not found in library path.
Please install CuDNN from https://developer.nvidia.com/cuDNN
Then make sure files named as libcudnn.so.4 or libcudnn.4.dylib are placed in your library load path (for example /usr/local/lib , or manually add a path to LD_LIBRARY_PATH)

stack traceback:
    [C]: in function 'error'
    /home/user/torch/install/share/lua/5.1/trepl/init.lua:384: in function 'require'
    neural_style.lua:64: in function 'main'
    neural_style.lua:500: in main chunk
    [C]: in function 'dofile'
    .../user/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:145: in main chunk
    [C]: at 0x00406670
user@user-XPS-8500:~/neural-style$ 
我的解决方案是，执行如下代码：
luarocks install cudnn


写上你的model文件的路径和图片文件夹的路径，如果你想预测所有的图片将： 
-num_images 10 改为-num_images -1
启动httpServer
预测之后会在你的vis文件夹下生成vis.json文件在imgs中会有你预测的图片:
启动server：
$ cd vis
$ python -m SimpleHTTPServer
浏览器访问：
localhost:8000，就可以看见了
我的结果
我随机从百度上下了6个图片，预测结果是这样的：