深度学习框架 Torch 7 问题笔记

Summary on deep learning framework --- Torch7 

2018-07-22 21:30:28

1. 尝试第一个 CNN 的 torch版本, 代码如下:

　　

 --    We now have 5 steps left to do in training our first torch neural network
 --    1. Load and normalize data
 --    2. Define Neural Network
 --    3. Define Loss function
 --    4. Train network on training data
 --    5. Test network on test data.

 --    1. Load and normalize data
 require 'paths'
 require 'image';
 if (not paths.filep("cifar10torchsmall.zip")) then
     os.execute('wget -c https://s3.amazonaws.com/torch7/data/cifar10torchsmall.zip')
     os.execute('unzip cifar10torchsmall.zip')
 end
 trainset = torch.load('cifar10-train.t7')
 testset = torch.load('cifar10-test.t7')
 classes = {'airplane', 'automobile', 'bird', 'cat',
            'deer', 'dog', 'frog', 'horse', 'ship', 'truck'}

 print(trainset)
 print(#trainset.data)

 itorch.image(trainset.data[]) -- display the 100-th image in dataset
 print(classes[trainset.label[]])

 -- ignore setmetatable for now, it is a feature beyond the scope of this tutorial.
 -- It sets the index operator
 setmetatable(trainset,
     {__index = function(t, i)
                     return {t.data[i], t.label[i]}
                 end}
 );
 trainset.data = trainset.data:double()  -- convert the data from a ByteTensor to a DoubleTensor.

 function trainset:size()
     return self.data:size()
 end

 print(trainset:size())
 print(trainset[])
 itorch.image(trainset[][])

 redChannel = trainset.data[{ {}, {}, {}, {} }] -- this pick {all images, 1st channel, all vertical pixels, all horizontal pixels}
 print(#redChannel)

 -- TODO:fill
 mean = {}
 stdv = {}
 for i = , do
     mean[i] = trainset.data[{ {}, {i}, {}, {} }]:mean()  -- mean estimation
     print('Channel ' .. i .. ' , Mean: ' .. mean[i])
     trainset.data[{ {}, {i}, {}, {} }]:add(-mean[i]) -- mean subtraction 

     stdv[i] = trainset.data[ { {}, {i}, {}, {} }]:std()  -- std estimation
     print('Channel ' .. i .. ' , Standard Deviation: ' .. stdv[i])
     trainset.data[{ {}, {i}, {}, {} }]:div(stdv[i])  -- std scaling
 end 

 --    2. Define Neural Network
 net = nn.Sequential()
 net:add(nn.SpatialConvolution(, , , )) -- 3 input image channels, 6 output channels, 5x5 convolution kernel
 net:add(nn.ReLU())                       -- non-linearity
 net:add(nn.SpatialMaxPooling(,,,))     -- A max-pooling operation that looks at 2x2 windows and finds the max.
 net:add(nn.SpatialConvolution(, , , ))
 net:add(nn.ReLU())                       -- non-linearity
 net:add(nn.SpatialMaxPooling(,,,))
 net:add(nn.View(**))                    -- reshapes from a 3D tensor of 16x5x5 into 1D tensor of 16*5*5
 net:add(nn.Linear(**, ))             -- fully connected layer (matrix multiplication between input and weights)
 net:add(nn.ReLU())                       -- non-linearity
 net:add(nn.Linear(, ))
 net:add(nn.ReLU())                       -- non-linearity
 net:add(nn.Linear(, ))                   -- 10 is the number of outputs of the network (in this case, 10 digits)
 net:add(nn.LogSoftMax())                     -- converts the output to a log-probability. Useful for classification problems

 -- 3. Let us difine the Loss function
 criterion = nn.ClassNLLCriterion()

 -- 4. Train the neural network
 trainer = nn.StochasticGradient(net, criterion)
 trainer.learningRate = 0.001
 trainer.maxIteration =  -- just do 5 epochs of training.
 trainer:train(trainset)

 -- 5. Test the network, print accuracy
 print(classes[testset.label[]])
 itorch.image(testset.data[])

 testset.data = testset.data:double()   -- convert from Byte tensor to Double tensor
 for i=, do -- over each image channel
     testset.data[{ {}, {i}, {}, {}  }]:add(-mean[i]) -- mean subtraction
     testset.data[{ {}, {i}, {}, {}  }]:div(stdv[i]) -- std scaling
 end

 -- for fun, print the mean and standard-deviation of example-100
 horse = testset.data[]
 print(horse:mean(), horse:std())

 print(classes[testset.label[]])
 itorch.image(testset.data[])
 predicted = net:forward(testset.data[])

 -- the output of the network is Log-Probabilities. To convert them to probabilities, you have to take e^x
 print(predicted:exp())

 for i=,predicted:size() do
     print(classes[i], predicted[i])
 end

 -- test the accuracy
 correct =
 for i=, do
     local groundtruth = testset.label[i]
     local prediction = net:forward(testset.data[i])
     local confidences, indices = torch.sort(prediction, true)  -- true means sort in descending order
     if groundtruth == indices[] then
         correct = correct +
     end
 end

 print(correct, *correct/ .. ' % ')

 class_performance = {, , , , , , , , , }
 for i=, do
     local groundtruth = testset.label[i]
     local prediction = net:forward(testset.data[i])
     local confidences, indices = torch.sort(prediction, true)  -- true means sort in descending order
     if groundtruth == indices[] then
         class_performance[groundtruth] = class_performance[groundtruth] +
     end
 end

 for i=,#classes do
     print(classes[i], *class_performance[i]/ .. ' %')
 end

 require 'cunn';
 net = net:cuda()
 criterion = criterion:cuda()
 trainset.data = trainset.data:cuda()
 trainset.label = trainset.label:cuda()

 trainer = nn.StochasticGradient(net, criterion)
 trainer.learningRate = 0.001
 trainer.maxIteration =  -- just do 5 epochs of training.

 trainer:train(trainset)

　　

　　  那么,运行起来 却出现如下的问题:

　　(1).

/home/wangxiao/torch/install/bin/luajit: ./train_network.lua:26: attempt to index global 'itorch' (a nil value)
stack traceback:
./train_network.lua:26: in main chunk
[C]: in function 'dofile'
...xiao/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:145: in main chunk
[C]: at 0x00406670
wangxiao@AHU:~/Documents/Lua test examples$

　　  主要是 itorch 的问题, 另外就是 要引用 require 'nn' 来解决 无法辨别 nn 的问题.

　　我是把 带有 itorch 的那些行都暂时注释了.

2.  'libcudnn (R5) not found in library path.

wangxiao@AHU:~/Downloads/wide-residual-networks-master$ th ./train_Single_Multilabel_Image_Classification.lua
nil
/home/wangxiao/torch/install/bin/luajit: /home/wangxiao/torch/install/share/lua/5.1/trepl/init.lua:384: /home/wangxiao/torch/install/share/lua/5.1/trepl/init.lua:384: /home/wangxiao/torch/install/share/lua/5.1/cudnn/ffi.lua:1600: '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)

stack traceback:
[C]: in function 'error'
/home/wangxiao/torch/install/share/lua/5.1/trepl/init.lua:384: in function 'require'
./train_Single_Multilabel_Image_Classification.lua:8: in main chunk
[C]: in function 'dofile'
...xiao/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:145: in main chunk
[C]: at 0x00406670
wangxiao@AHU:~/Downloads/wide-residual-networks-master$

================================================================>>

答案是:

　　重新下载了 cudnn-7.5-linux-x64-v5.0-ga.tgz

　　并且重新配置了,但是依然提醒这个问题,那么,问题何在呢?查看了博客:http://blog.csdn.net/hungryof/article/details/51557666 中的内容:

---

坑4 可能出现’libcudnn not found in library path’的情况

截取其中一段错误信息：

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)

• 1
• 2

LD_LIBRARY_PATH是该环境变量，主要用于指定查找共享库（动态链接库）时除了默认路径之外的其他路径。由于刚才已经将 
“libcudnn*”复制到了/usr/local/cuda-7.5/lib64/下面，因此需要

1. sudo gedit /etc/ld.so.conf.d/cudnn.conf 就是新建一个conf文件。名字随便
2. 加入刚才的路径/usr/local/cuda-7.5/lib64/
3. 反正我还添加了/usr/local/cuda-7.5/include/,这个估计不要也行。
4. 保存后，再sudo ldconfig来更新缓存。（可能会出现libcudnn.so.5不是符号连接的问题，不过无所谓了！！）

此时运行

th neural_style.lua -gpu 0 -backend cudnn

• 1

成功了！！！！

============================================================>>>>

评价:　　按照这种做法试了,确实成功了! 赞一个 !!!

---

　　3. 利用 gm 加载图像时,提示错误,但是装上那个包仍然提示错误:

　　  

　　

Load library:

gm = require 'graphicsmagick'

First, we provide two high-level functions to load/save directly into/form tensors:

img = gm.load('/path/to/image.png' [, type])    -- type = 'float' (default) | 'double' | 'byte'
gm.save('/path/to/image.jpg' [,quality])        -- quality = 0 to 100 (for jpegs only)

The following provide a more controlled flow for loading/saving jpegs.

Create an image, from a file:

image = gm.Image('/path/to/image.png')
-- or
image = gm.Image()
image:load('/path/to/image.png')

　　但是悲剧的仍然有错, 只好换了用  image.load() 的方式加载图像:
　　

--To load as byte tensor for rgb imagefile
local img = image.load(imagefile,3,'byte')

---

 
　　4. Torch 保存 txt 文件：
　　-- save opt 
　　file = torch.DiskFile(paths.concat(opt.checkpoints_dir, opt.name, 'opt.txt'), 'w') 
　　file:writeObject(opt) 
　　file:close() 
　　
　　5. Torch 创建新的文件夹
　　opts.modelPath = opt.modelDir .. opt.modelName 
　　if not paths.dirp(opt.modelPath) then 
　　　　paths.mkdir(opts.modelPath) 
　　end 

　　6. Torch Lua 保存 图像到文件夹 
　　借助 image package，首先安装： luarocks install image 
　　然后 require 'image' 
　　就可以使用了： local img = image.save('./saved_pos_neg_image/candidate_' .. tostring(i) .. tostring(j) .. '.png', pos_patch, 1, 32, 32) 

　　7. module 'bit' not found:No LuaRocks module found for bit

wangxiao@AHU:/media/wangxiao/724eaeef-e688-4b09-9cc9-dfaca44079b2/fast-neural-style-master$ th ./train.lua
/home/wangxiao/torch/install/bin/lua: /home/wangxiao/torch/install/share/lua/5.2/trepl/init.lua:389: /home/wangxiao/torch/install/share/lua/5.2/trepl/init.lua:389: /home/wangxiao/torch/install/share/lua/5.2/trepl/init.lua:389: module 'bit' not found:No LuaRocks module found for bit
no field package.preload['bit']
no file '/home/wangxiao/.luarocks/share/lua/5.2/bit.lua'
no file '/home/wangxiao/.luarocks/share/lua/5.2/bit/init.lua'
no file '/home/wangxiao/torch/install/share/lua/5.2/bit.lua'
no file '/home/wangxiao/torch/install/share/lua/5.2/bit/init.lua'
no file '/home/wangxiao/.luarocks/share/lua/5.1/bit.lua'
no file '/home/wangxiao/.luarocks/share/lua/5.1/bit/init.lua'
no file '/home/wangxiao/torch/install/share/lua/5.1/bit.lua'
no file '/home/wangxiao/torch/install/share/lua/5.1/bit/init.lua'
no file './bit.lua'
no file '/home/wangxiao/torch/install/share/luajit-2.1.0-beta1/bit.lua'
no file '/usr/local/share/lua/5.1/bit.lua'
no file '/usr/local/share/lua/5.1/bit/init.lua'
no file '/home/wangxiao/.luarocks/lib/lua/5.2/bit.so'
no file '/home/wangxiao/torch/install/lib/lua/5.2/bit.so'
no file '/home/wangxiao/torch/install/lib/bit.so'
no file '/home/wangxiao/.luarocks/lib/lua/5.1/bit.so'
no file '/home/wangxiao/torch/install/lib/lua/5.1/bit.so'
no file './bit.so'
no file '/usr/local/lib/lua/5.1/bit.so'
no file '/usr/local/lib/lua/5.1/loadall.so'
stack traceback:
[C]: in function 'error'
/home/wangxiao/torch/install/share/lua/5.2/trepl/init.lua:389: in function 'require'
./train.lua:5: in main chunk
[C]: in function 'dofile'
...xiao/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:145: in main chunk
[C]: in ?
wangxiao@AHU:/media/wangxiao/724eaeef-e688-4b09-9cc9-dfaca44079b2/fast-neural-style-master$

在终端中执行：luarocks install luabitop 
就可以了。 

---

8.　　HDF5Group:read() - no such child 'media' for [HDF5Group 33554432 /]

/home/wangxiao/torch/install/bin/lua: /home/wangxiao/torch/install/share/lua/5.2/hdf5/group.lua:312: HDF5Group:read() - no such child 'media' for [HDF5Group 33554432 /]
stack traceback:
[C]: in function 'error'
/home/wangxiao/torch/install/share/lua/5.2/hdf5/group.lua:312: in function </home/wangxiao/torch/install/share/lua/5.2/hdf5/group.lua:302>
(...tail calls...)
./fast_neural_style/DataLoader.lua:44: in function '__init'
/home/wangxiao/torch/install/share/lua/5.2/torch/init.lua:91: in function </home/wangxiao/torch/install/share/lua/5.2/torch/init.lua:87>
[C]: in function 'DataLoader'
./train.lua:138: in function 'main'
./train.lua:327: in main chunk
[C]: in function 'dofile'
...xiao/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:145: in main chunk
[C]: in ?

最近在训练 类型迁移的代码，发现这个蛋疼的问题。哎。。纠结好几天了。。这个 hdf5 到底怎么回事 ？  求解释 ！！！

------------------------------------------------------------------------------------------------

　　后来发现, 是我自己的数据集路径设置的有问题, 如: 应该是 CoCo/train/image/

　　但是,我只是给定了 CoCo/train/ ...

---

　　9. 怎么设置 torch代码在哪块 GPU 上运行 ? 或者 怎么设置在两块卡上同时运行 ?

　　

　  　

　　如图所示: export CUDA_VISIBLE_DEVICES=0 即可指定代码在 GPU-0 上运行. 

　　

---

　　10.  When load the pre-trained VGG model, got the following errors:

　　　　MODULE data UNDEFINED
　　　　warning: module 'data [type 5]' not found
　　　　nn supports no groups!
　　　　warning: module 'conv2 [type 4]' not found
　　　　nn supports no groups!
　　　　warning: module 'conv4 [type 4]' not found
　　　　nn supports no groups!
　　　　warning: module 'conv5 [type 4]' not found

　　

 using cudnn
 Successfully loaded ./feature_transfer/AlexNet_files/bvlc_alexnet.caffemodel
 MODULE data UNDEFINED
 warning: module 'data [type 5]' not found
 nn supports no groups!
 warning: module 'conv2 [type 4]' not found
 nn supports no groups!
 warning: module 'conv4 [type 4]' not found
 nn supports no groups!
 warning: module 'conv5 [type 4]' not found

 wangxiao@AHU:~/Downloads/multi-modal-visual-tracking$ qlua ./train_match_function_alexNet_version_2017_02_28.lua
 using cudnn
 Successfully loaded ./feature_transfer/AlexNet_files/bvlc_alexnet.caffemodel
 MODULE data UNDEFINED
 warning: module 'data [type 5]' not found
 nn supports no groups!
 warning: module 'conv2 [type 4]' not found
 nn supports no groups!
 warning: module 'conv4 [type 4]' not found
 nn supports no groups!
 warning: module 'conv5 [type 4]' not found
 conv1:
 conv3:
 fc6:
 fc7:
 fc8:
 nn.Sequential {
 [input -> () -> () -> () -> output]
 (): nn.SplitTable
 (): nn.ParallelTable {
 input
 |`-> (): nn.Sequential {
 | [input -> () -> () -> () -> () -> () -> () -> () -> () -> () -> () -> () -> () -> () -> () -> () -> () -> () -> () -> output]
 | (): nn.SpatialConvolution( -> , 11x11, ,)
 | (): nn.ReLU
 | (): nn.SpatialCrossMapLRN
 | (): nn.SpatialMaxPooling(3x3, ,)
 | (): nn.ReLU
 | (): nn.SpatialCrossMapLRN
 | (): nn.SpatialMaxPooling(3x3, ,)
 | (): nn.SpatialConvolution( -> , 3x3, ,, ,)
 | (): nn.ReLU
 | (): nn.ReLU
 | (): nn.ReLU
 | (): nn.SpatialMaxPooling(3x3, ,)
 | (): nn.View(-)
 | (): nn.Linear( -> )
 | (): nn.ReLU
 | (): nn.Dropout(0.500000)
 | (): nn.Linear( -> )
 | (): nn.ReLU
 | }
 `-> (): nn.Sequential {
 [input -> () -> () -> () -> () -> () -> () -> () -> () -> () -> () -> () -> () -> () -> () -> () -> () -> () -> () -> output]
 (): nn.SpatialConvolution( -> , 11x11, ,)
 (): nn.ReLU
 (): nn.SpatialCrossMapLRN
 (): nn.SpatialMaxPooling(3x3, ,)
 (): nn.ReLU
 (): nn.SpatialCrossMapLRN
 (): nn.SpatialMaxPooling(3x3, ,)
 (): nn.SpatialConvolution( -> , 3x3, ,, ,)
 (): nn.ReLU
 (): nn.ReLU
 (): nn.ReLU
 (): nn.SpatialMaxPooling(3x3, ,)
 (): nn.View(-)
 (): nn.Linear( -> )
 (): nn.ReLU
 (): nn.Dropout(0.500000)
 (): nn.Linear( -> )
 (): nn.ReLU
 }
 ... -> output
 }
 (): nn.PairwiseDistance
 }
 =================================================================================================================
 ================= AlextNet based Siamese Search for Visual Tracking ========================
 =================================================================================================================
 ==>> The Benchmark Contain:  videos ...
 deal with video / video name: BlurFace ... please waiting ...
 the num of gt bbox:
 the num of video frames:
 ========>>>> Begin to track  video name: nil-th frame, please waiting ...
 ========>>>> Begin to track  video name: nil-th frame, please waiting ... ............] ETA: 0ms | Step: 0ms
 ========>>>> Begin to track  video name: nil-th frame, please waiting ... ............] ETA: 39s424ms | Step: 80ms
 ========>>>> Begin to track  video name: nil-th frame, please waiting ... ............] ETA: 33s746ms | Step: 69ms
 ========>>>> Begin to track  video name: nil-th frame, please waiting ... ............] ETA: 31s817ms | Step: 65ms
 ========>>>> Begin to track  video name: nil-th frame, please waiting ... ............] ETA: 32s575ms | Step: 66ms
 ========>>>> Begin to track  video name: nil-th frame, please waiting ... ............] ETA: 34s376ms | Step: 70ms
 ========>>>> Begin to track  video name: nil-th frame, please waiting ... ............] ETA: 40s240ms | Step: 82ms
 ========>>>> Begin to track  video name: nil-th frame, please waiting ... ...........] ETA: 44s211ms | Step: 91ms
 ========>>>> Begin to track  video name: nil-th frame, please waiting ... ...........] ETA: 45s993ms | Step: 95ms
 ========>>>> Begin to track  video name: nil-th frame, please waiting ... ...........] ETA: 47s754ms | Step: 99ms
 ========>>>> Begin to track  video name: nil-th frame, please waiting ... ...........] ETA: 50s392ms | Step: 104ms
 ========>>>> Begin to track  video name: nil-th frame, please waiting ... ...........] ETA: 53s138ms | Step: 110ms
 ========>>>> Begin to track  video name: nil-th frame, please waiting ... ...........] ETA: 55s793ms | Step: 116ms
 ========>>>> Begin to track  video name: nil-th frame, please waiting ... ...........] ETA: 59s253ms | Step: 123ms
 ========>>>> Begin to track  video name: nil-th frame, please waiting ... ...........] ETA: 1m2s | Step: 130ms
 ========>>>> Begin to track  video name: nil-th frame, please waiting ... ...........] ETA: 1m5s | Step: 137ms
 ========>>>> Begin to track  video name: nil-th frame, please waiting ... ...........] ETA: 1m8s | Step: 143ms
 ========>>>> Begin to track  video name: nil-th frame, please waiting ... ...........] ETA: 1m11s | Step: 149ms
 //////////////////////////////////////////////////////////////////////////..............] ETA: 1m14s | Step: 157ms
 ==>> pos_proposal_list:
 ==>> neg_proposal_list:
 qlua: /home/wangxiao/torch/install/share/lua/5.1/nn/Container.lua::
 In  module of nn.Sequential:
 In  module of nn.ParallelTable:
 In  module of nn.Sequential:
 /home/wangxiao/torch/install/share/lua/5.1/nn/THNN.lua:: Need input of dimension  and input.size[] ==  but got input to be of shape: [ x  x ] at /tmp/luarocks_cunn-scm--/cunn/lib/THCUNN/generic/SpatialConvolutionMM.cu:
 stack traceback:
 [C]: in function 'v'
 /home/wangxiao/torch/install/share/lua/5.1/nn/THNN.lua:: in function 'SpatialConvolutionMM_updateOutput'
 ...ao/torch/install/share/lua/5.1/nn/SpatialConvolution.lua:: in function <...ao/torch/install/share/lua/5.1/nn/SpatialConvolution.lua:>
 [C]: in function 'xpcall'
 /home/wangxiao/torch/install/share/lua/5.1/nn/Container.lua:: in function 'rethrowErrors'
 ...e/wangxiao/torch/install/share/lua/5.1/nn/Sequential.lua:: in function <...e/wangxiao/torch/install/share/lua/5.1/nn/Sequential.lua:>
 [C]: in function 'xpcall'
 /home/wangxiao/torch/install/share/lua/5.1/nn/Container.lua:: in function 'rethrowErrors'
 ...angxiao/torch/install/share/lua/5.1/nn/ParallelTable.lua:: in function <...angxiao/torch/install/share/lua/5.1/nn/ParallelTable.lua:>
 [C]: in function 'xpcall'
 /home/wangxiao/torch/install/share/lua/5.1/nn/Container.lua:: in function 'rethrowErrors'
 ...e/wangxiao/torch/install/share/lua/5.1/nn/Sequential.lua:: in function 'forward'
 ./train_match_function_alexNet_version_2017_02_28.lua:: in function 'opfunc'
 /home/wangxiao/torch/install/share/lua/5.1/optim/adam.lua:: in function 'optim'
 ./train_match_function_alexNet_version_2017_02_28.lua:: in main chunk

 WARNING: If you see a stack trace below, it doesn't point to the place where this error occurred. Please use only the one above.
 stack traceback:
 [C]: at 0x7f86014df9c0
 [C]: in function 'error'
 /home/wangxiao/torch/install/share/lua/5.1/nn/Container.lua:: in function 'rethrowErrors'
 ...e/wangxiao/torch/install/share/lua/5.1/nn/Sequential.lua:: in function 'forward'
 ./train_match_function_alexNet_version_2017_02_28.lua:: in function 'opfunc'
 /home/wangxiao/torch/install/share/lua/5.1/optim/adam.lua:: in function 'optim'
 ./train_match_function_alexNet_version_2017_02_28.lua:: in main chunk
 wangxiao@AHU:~/Downloads/multi-modal-visual-tracking$

　　Just like the screen shot above, change the 'nn' into 'cudnn' will be ok and passed.

　　11. both (null) and torch.FloatTensor have no less-than operator

　　　　qlua: ./test_MM_tracker_VGG_.lua:254: both (null) and torch.FloatTensor have no less-than operator
　　　　stack traceback:
　　　　[C]: at 0x7f628816e9c0
　　　　[C]: in function '__lt'
　　　　./test_MM_tracker_VGG_.lua:254: in main chunk

　　

　　Because it is floatTensor () style and you can change it like this if you want this value printed in a for loop:　predictValue -->> predictValue[i] .

　　

　　12.

========>>>> Begin to track the 6-th and the video name is ILSVRC2015_train_00109004 , please waiting ...
THCudaCheck FAIL file=/tmp/luarocks_cutorch-scm-1-707/cutorch/lib/THC/generic/THCStorage.cu line=66 error=2 : out of memory
qlua: cuda runtime error (2) : out of memory at /tmp/luarocks_cutorch-scm-1-707/cutorch/lib/THC/generic/THCStorage.cu:66
stack traceback:
[C]: at 0x7fa20a8f99c0
[C]: at 0x7fa1dddfbee0
[C]: in function 'Tensor'
./train_match_function_VGG_version_2017_03_02.lua:377: in main chunk
wangxiao@AHU:~/Downloads/multi-modal-visual-tracking$

Yes, it is just out of memory of GPU. Just turn the batchsize to a small value, it may work. It worked for me. Ha ha ...

13. luarocks install class does not have any effect, it still shown me the error: No Module named "class" in Torch.

　　==>> in terminal, install this package in sudo.

　　==>> then, it will be OK.

14. How to install opencv 3.1 on Ubuntu 14.04 ??? 

　　As we can found from: http://blog.csdn.net/a125930123/article/details/52091140

　　1. first, you should install torch successfully ;

　　2. then, just follow what the blog said here:

安装opencv3.1

1、安装必要的包

sudo apt-get install build-essential
sudo apt-get install cmake git libgtk2.0-dev pkg-config libavcodec-dev libavformat-dev libswscale-dev

2、下载opencv3.1

http://opencv.org/downloads.html
解压：unzip  opencv-3.1.0

3、安装
cd ~/opencv-3.1.0
mkdir build
cd build
cmake -D CMAKE_BUILD_TYPE=Release -D CMAKE_INSTALL_PREFIX=/usr/local ..  

sudo make -j24 
sudo make install -j24  
sudo /bin/bash -c 'echo "/usr/local/lib" > /etc/ld.so.conf.d/opencv.conf'
sudo ldconfig

安装完成

4、问题

在安装过程中可能会出现无法下载 ippicv_linux_20151201.tgz的问题。

解决方案：

手动下载ippicv_linux_20151201.tgz：https://raw.githubusercontent.com/Itseez/opencv_3rdparty/81a676001ca8075ada498583e4166079e5744668/ippicv/ippicv_linux_20151201.tgz

将下载好的文件  放入 opencv-3.1.0/3rdparty/ippicv/downloads/linux-808b791a6eac9ed78d32a7666804320e 中，如果已经存在 ，则替换掉，这样就可以安装完成了。

5、最后执行命令

luarocks install cv

OpenCV bindings for Torch安装成功。

But, maybe you may found some errors, such as:

cudalegacy/src/graphcuts.cpp:120:54: error: ‘NppiGraphcutState’ has not been declared    (solution draw from: http://blog.csdn.net/allyli0022/article/details/62859290)

At this moment, you need to change some files:

found graphcuts.cpp in opencv3.1, and do the following changes:

解决方案：需要修改一处源码：
在graphcuts.cpp中将

#if !defined (HAVE_CUDA) || defined (CUDA_DISABLER) 

改为

#if !defined (HAVE_CUDA) || defined (CUDA_DISABLER) || (CUDART_VERSION >= 8000) 

then, try again, it will be ok...this code just want to make opencv3.1 work under cuda 8.0, you know...skip that judge sentence...

---

15.  安装torch-hdf5 

sudo apt-get install libhdf5-serial-dev hdf5-tools
git clone https://github.com/deepmind/torch-hdf5
cd torch-hdf5
sudo luarocks make hdf5--.rockspec LIBHDF5_LIBDIR=”/usr/lib/x86_64-Linux-gnu/”

17. iTorch安装

git clone https://github.com/zeromq/zeromq4-1.git
mkdir build-zeromq
cd build-zeromq
cmake ..
make && make install
安装完之后，luarocks install itorch
之后可以通过luarocks list查看是否安装成功