PointNet++复现(各种避坑)
因为科研需要,最近要研究PoinetNet++网络,在安装过程中遇到各种问题,终于还是艰难上了车,把如何安装的过程写下来,希望帮助各位避坑。以上仅是博主自己遇到的错误,会存在一定局限性,如有错误,欢迎指正。
github原文地址:https://github.com/charlesq34/pointnet2
一、安装Tensorflow
强烈建议安装Tensorflow版本低于2.0,否则需要大量修改源代码!!!本文采用AnaConda2安装Tensorflow,Tensorflow官网上pip安装,源码安装,docker都试过,不仅相对比较麻烦,还会有各种问题。
- 下载Anaconda
Anaconda安装教程:https://blog.csdn.net/lwplwf/article/details/79162470
建议下载python2.7版本 - 安装cuda
博主是在tensorflow安装教程内学习安装的cuda,并没有遇到问题,建议安装版本cuda10.1
原链接:https://tensorflow.google.cn/install/gpu
# Add NVIDIA package repositories
wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/cuda-repo-ubuntu1804_10.1.243-1_amd64.deb
sudo dpkg -i cuda-repo-ubuntu1804_10.1.243-1_amd64.deb
sudo apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/7fa2af80.pub
sudo apt-get update
wget http://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1804/x86_64/nvidia-machine-learning-repo-ubuntu1804_1.0.0-1_amd64.deb
sudo apt install ./nvidia-machine-learning-repo-ubuntu1804_1.0.0-1_amd64.deb
sudo apt-get update
# Install NVIDIA driver
sudo apt-get install --no-install-recommends nvidia-driver-418
# Reboot. Check that GPUs are visible using the command: nvidia-smi
# Install development and runtime libraries (~4GB)
sudo apt-get install --no-install-recommends cuda-10-1
sudo apt-get install --no-install-recommends libcudnn7=7.6.4.38-1+cuda10.1
sudo apt-get install --no-install-recommends libcudnn7-dev=7.6.4.38-1+cuda10.1
# Install TensorRT. Requires that libcudnn7 is installed above.
sudo apt-get install -y --no-install-recommends libnvinfer6=6.0.1-1+cuda10.1
sudo apt-get install -y --no-install-recommends libnvinfer-dev=6.0.1-1+cuda10.1
sudo apt-get install -y --no-install-recommends libnvinfer-plugin6=6.0.1-1+cuda10.1
- 安装Tensorflow
创建虚拟环境:
conda create -n tf
进入虚拟环境:
source activate tf
在虚拟环境中安装Tensorflow
如果只安装CPU版本:
conda install tensorflow
安装GPU版本:
conda install tensorflow-gpu
建议安装gpu版本,且指定版本<2.0的,博主安装的1.14:conda install tensorflow-gpu==1.14
现在只需等待即可
- 在虚拟环境中测试Tensorflow是否安装成功
source activate tf
python
impotr tensorflow as tf
如果没有报错就安装成功了,想看tensorflow版本就再加一句:
tf.__version__
二、PoinetNet++复现
- 下载pointnet++
https://github.com/charlesq34/pointnet2下载到本地并解压 - 可能遇到的问题
如果python为3.7,需要把所有xrange改为range
print后需加括号括起来 - 编译tf_ops文件夹
文件夹内功3个子文件夹,需分别修改.sh文件进行编译的到.cu.o和.so.so文件
修改.sh文件参阅:https://blog.csdn.net/qq_40196164/article/details/84638410?depth_1-utm_source=distribute.pc_relevant.none-task-blog-BlogCommendFromBaidu-3&utm_source=distribute.pc_relevant.none-task-blog-BlogCommendFromBaidu-3
补充上文博主尚未提及的地方:
在sh文件末尾会有
-ltensorflow_framework -O2 -D_GLIBCXX_USE_CXX11_ABI=0
这两处也要根据电脑情况做修改:
/anaconda2/envs/tf/lib/python3.7/site-packages/tensorflow内如果有libtensflow文件,就应该把-ltensorflow_framework改成l:文件名,如l:libtensorflow_framework.so.1
如果电脑gcc版本高于4.0,把 -D_GLIBCXX_USE_CXX11_ABI=0注释掉即可。
附上博主自己调试时的sh文件。
/usr/local/cuda-10.1/bin/nvcc tf_sampling_g.cu -o tf_sampling_g.cu.o -c -O2 -DGOOGLE_CUDA=1 -x cu -Xcompiler -fPIC
g++ -std=c++11 tf_sampling.cpp tf_sampling_g.cu.o -o tf_sampling_so.so -shared -fPIC -I /home/gfb/anaconda2/envs/tf/lib/python3.7/site-packages/tensorflow/include -I /usr/local/cuda-10.1/include -I /home/gfb/anaconda2/envs/tf/lib/python3.7/site-packages/tensorflow/include/external/nsync/public -lcudart -L /usr/local/cuda-10.1/lib64/ -L /home/gfb/anaconda2/envs/tf/lib/python3.7/site-packages/tensorflow -l:libtensorflow_framework.so.1 -O2 #-D_GLIBCXX_USE_CXX11_ABI=0
需要注意的是,每个人配置不同,需找到自己电脑上何时的位置。
- 编译train.py
在pointnet2-master文件夹内进入终端,输入source activate tf进去虚拟环境,进而输入python train.py开始PoineNet++之旅。
参考博客:
https://blog.csdn.net/qq_40196164/article/details/84638410?depth_1-utm_source=distribute.pc_relevant.none-task-blog-BlogCommendFromBaidu-3&utm_source=distribute.pc_relevant.none-task-blog-BlogCommendFromBaidu-3
https://blog.csdn.net/m0_37864814/article/details/82112029