配置相关note

1. error：_C.cpython-36m-x86_64-linux-gnu.so: undefined symbol:_ZN6caffe26detail37_typeMetaDataInstance_preallocated_32E

一开始是报错lib.cudaart.so.10.0找不到的错，下载了cudatoolkit=10.0再报的这个错

解决：把之前的build文件和之前的******.so文件删掉，重新编译就好了，记得版本问题

 

2.numpy.core.multiarray failed to import

https://blog.csdn.net/m0_37733057/article/details/88426147

 

3. 换源

https://blog.csdn.net/sscc_learning/article/details/85339529

 

4. ubuntu18.04安装anaconda3

https://blog.csdn.net/qq_15192373/article/details/81091098

 

5. 安装python-pcl（ubuntu18.04）

https://blog.csdn.net/weixin_38705903/article/details/88641948

https://blog.csdn.net/fanjiule/article/details/80913894

步骤：提前安装好torch numpy matplotlib什么的

git clone https://github.com/PointCloudLibrary/pcl.git
sudo apt-get install  libboost-all-dev libeigen3-dev libflann-dev python libusb-1.0-0-dev libudev-dev freeglut3-dev doxygen graphviz libpng12-dev libgtest-dev libxmu-dev libxi-dev libpcap-dev libqhull-dev libvtk5-qt4-dev python-vtk libvtk-java
sudo apt-get install --fix-missing  libboost-all-dev libeigen3-dev libflann-dev python libusb-1.0-0-dev libudev-dev freeglut3-dev doxygen graphviz libpng12-dev libgtest-dev libxmu-dev libxi-dev libpcap-dev libqhull-dev libvtk5-qt4-dev python-vtk libvtk-java
cd pcl
mkdir build
cd build
cmake  -D CMAKE_BUILD_TYPE=None  -D BUILD_GPU=ON  -D BUILD_apps=ON  -D BUILD_examples=ON ..

 (图片来自https://blog.csdn.net/fanjiule/article/details/80913894)

git clone https://github.com/strawlab/python-pcl.git
pip install --upgrade pip
pip install cython==0.25.2

cd python-pcl

sudo apt-get install software-properties-common python-software-properties
sudo add-apt-repository ppa:v-launchpad-jochen-sprickerhof-de/pcl
sudo apt-get update
sudo apt-get install libpcl-dev

sudo python setup.py build_ext -i
python setup.py install

安装完成pip list中会有python-pcl 0.3

 

6.ubuntu anaconda升级torch和torchvision

# 进入环境卸载之前的torch torchvision
pip uninstall torch torchvision

# 系统是cuda9.0版本的(nvcc -V可查)，换源成清华源
pip install torch==1.5.0 torchvision==0.6.0 -i https://pypi.tuna.tsinghua.edu.cn/simple

# 这里注意官方给的比较繁琐的pip的安装方法，其实没有必要
# 比如pip install torch==1.5.0+cu92 torchvision==0.6.0+cu92 -f https://download.pytorch.org/whl/torch_stable.html
# 只需使用其中的torch torchvision版本对应信息避免torch torchvision冲突就好了

7.plt.imsave('path',image, cmap='inferno')

其中参数cmap的含义：

意思就是把标量映射到了RGB值