Ubuntu16.04 + troch + cuda + deep-painterly-harmonization测试记录

deep-painterly-harmonization测试记录

---

安装CUDA & cudnn

参见另一篇Ubuntu 16.04 GTX950M + cuda9.0 + cuDNN7.0 + TensorFlow 1.5 / 1.8 安装记录

---

安装Pytorch

最后发现要求的不是pytorch而是torch

官网：https://pytorch.org/

中文文档：http://pytorch-cn.readthedocs.io/zh/latest/

# pip3 Python3.5 CUDA9.0
pip3 install http://download.pytorch.org/whl/cu90/torch-0.4.0-cp35-cp35m-linux_x86_64.whl 
pip3 install torchvision


#pip2 Python2.7 CUDA9.0
pip install http://download.pytorch.org/whl/cu90/torch-0.4.0-cp27-cp27mu-linux_x86_64.whl 
pip install torchvision 

# if the above command does not work, then you have python 2.7 UCS2, use this command 如果上面的命令不起作用，那么你有python 2.7 UCS2，使用这个命令
pip install http://download.pytorch.org/whl/cu90/torch-0.4.0-cp27-cp27m-linux_x86_64.whl

查看当前pytorch的版本

import torch
print(torch.__version__)

---

安装torch

git clone https://github.com/torch/distro.git ~/torch --recursive
cd ~/torch
bash install-deps
./install.sh

不幸的是，编译出错：

lib/THC/CMakeFiles/THC.dir/build.make:4243: recipe for target 'lib/THC/CMakeFiles/THC.dir/THC_generated_THCTensorMathPairwise.cu.o' failed
make[2]: *** [lib/THC/CMakeFiles/THC.dir/THC_generated_THCTensorMathPairwise.cu.o] Error 1
/pkgbuild/torch/torch/extra/cutorch/lib/THC/generic/THCTensorMath.cu(393): error: more than one operator "==" matches these operands:
            function "operator==(const __half &, const __half &)"
            function "operator==(half, half)"
            operand types are: half == half

/pkgbuild/torch/torch/extra/cutorch/lib/THC/generic/THCTensorMath.cu(414): error: more than one operator "==" matches these operands:
            function "operator==(const __half &, const __half &)"
            function "operator==(half, half)"
            operand types are: half == half

原因是cuda和torch的头文件都提供了相同的重载运算符，编译器不知道用哪一个。输入下面shell命令禁止使用cuda的头文件编译torch即可：
export TORCH_NVCC_FLAGS="-D__CUDA_NO_HALF_OPERATORS__"
重新运行./install.sh重新开始编译就OK了。

source ~/.zshrc之后输入命令th，如果能够出现下边的界面，安装成功

---

安装loadcaffe

# 依赖th
sudo apt-get install libprotobuf-dev protobuf-compiler

# Then install the package itself:
luarocks install loadcaffe

# In Ubuntu 16.04 you need to use gcc-5: 
CC=gcc-5 CXX=g++-5 luarocks install loadcaffe

#　测试
$ th
th> require 'loadcaffe'
{
  load : function: 0x4182c4c8
  C : userdata: 0x40636248
}
                                                                      [0.0719s] 
th> 

export TORCH_NVCC_FLAGS="-D__CUDA_NO_HALF_OPERATORS__"

直接在命令行用export添加路径，只能在当前shell脚本下可用，切换到另一个终端就会失效。要使设置永久有效，需要在如下文本里面添加命令，比如：

/etc/profile #开机执行,对所有shell用户执行环境变量设置,重启生效;常用
~/.bashrc #开机和登录执行bash shell采用;source ~/.bashrc生效,常用
/etc/bash.bashrc #~/.bashrc更广的版本
~/.profile #不能被bash读取

调用顺序:/etc/profile/-----> ~/.bashrc ------> /etc/bash.bashrc

# 参考 https://blog.csdn.net/Ychan_cc/article/details/66970190

• **错误信息:**Couldn’t load .caffemodel

  Couldn't load models/VGG_ILSVRC_19_layers.caffemodel
  /home/sun/torch/install/bin/luajit: neural_gram.lua:98: attempt to index a nil value
  stack traceback:
  neural_gram.lua:98: in function 'main'
  neural_gram.lua:450: in main chunk
  [C]: in function 'dofile'
  .../sun/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:150: in main chunk
  [C]: at 0x00405d50
  
  
  # 解决 检查.caffemodel文件是否损坏。 重新下载
  

• Out of memory

  THCudaCheck FAIL file=/home/sun/torch/extra/cutorch/lib/THC/generic/THCStorage.cu line=66 error=2 : out of memory
  /home/sun/torch/install/bin/luajit: neural_gram.lua:134: cuda runtime error (2) : out of memory at /home/sun/torch/extra/cutorch/lib/THC/generic/THCStorage.cu:66
  stack traceback:
  [C]: in function 'patchmatch_r'
  neural_gram.lua:134: in function 'main'
  neural_gram.lua:450: in main chunk
  [C]: in function 'dofile'
  .../sun/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:150: in main chunk
  [C]: at 0x00405d50
  
  # 解决 这是因为GPU内存不够用。用小一点的图片作为输入或者用小点的网络。
  
  
  vim gen_all.py
  
  # 找到 -image_size 700 一共有两行这个 i7-6700HQ Geforce 950M 4G显存
  
  -image_size 500   # 可以运行 但是很慢很慢 显存使用情况 3920MiB /  4046MiB 
  -image_size 300   # 这个还好一点

---

安装Octave

经查阅，通过sudo apt-get install octave 默认安装的为Octave-4.0.0, 4.0.0版本的好像有Bug，于是选择下面这个方式安装：

sudo apt-add-repository ppa:octave/stable
sudo apt-get update
sudo apt-get install octave

# 移除已安装的所有版本的Octave
# sudo apt-get remove octave
命令行 输入 octave 即可启动，然后固定在启动栏

运行deep-painterly-harmonization

# Download VGG-19:
sh models/download_models.sh

# Compile cuda_utils.cu (Adjust PREFIX and NVCC_PREFIX in makefile for your machine):
make clean && make

# To generate all results (in data/) using the provided scripts, simply run this in Python 
python gen_all.py

# and then run  in Matlab or Octave. The final output will be in results/.
run('filt_cnn_artifact.m')

部分错误参考链接 Github issue