Use pytorch to reproduce Deeplabv3+ (first step)-----Environment configuration
- This article is based on Ubuntu 18.04, one of the Linux distributions used, and is reproduced under pytorch. Students who use Windows or reproduce on tensorflow will automatically skip it;
- This article is being continuously updated. If students have questions about environment configuration, they can leave comments in the comment area at any time. The blogger will give guidance after seeing it;
- The hardware graphics card required to run Deeplabv3 needs to be better. This article uses NVIDIA GeForce RTX-2060.
- Finally, you may encounter various problems when learning deep learning. If you encounter problems, just search for solutions in Baidu, Zhihu, Bilibili, csdn, and Google Chrome. I hope everyone can learn something. become.
Deeplabv3+ paper address: https://arxiv.org/pdf/1802.02611.pdf Github
Deeplabv3+ code address: https://github.com/VainF/DeepLabV3Plus-Pytorch
Environment for this article: Ubuntu18.04 cuda 9.0.176 cudnn7.6.5 Anaconda3-5.3.0 torch1.13.1
Table of contents
2.cuda9.0 & cuDNN7.6.5 installation
1. Verify whether the system has a CUDA-supported GPU
2.cuda 9 requires gcc>=6, g++>=6, make sure your system meets these requirements
4. Install other import packages
5 Download and install cuda9 toolkit and cuDNN7.6.5
Download the Anaconda installation package
4. Code environment configuration
1.Ubuntu18.04 installation
If there are too many problems during the deeplabv3+ environment configuration process, reinstall the system. Refer to my previous article:
[Ubuntu18.04] Installation and configuration problems solved----Continuously updated
2.cuda9.0 & cuDNN7.6.5 installation
1. Verify whether the system has a CUDA-supported GPU
Enter the following code on the command line:
lspci | grep -i nvidia
You can view your graphics card information as: GeForce RTX 2060
You can also check your graphics card information in settings.
Check the NVIDIA website https://developer.nvidia.com/cuda-gpus for products that support CUDA . If you can find your own graphics card, it proves that your graphics card can install CUDA.
2.cuda 9 requires gcc>=6, g++>=6, make sure your system meets these requirements
Enter in the terminal:
gcc -v
g++ -v
#如果不满足要求，下载安装
sudo apt install gcc-6
sudo apt install g++-63.Install NVIDIA driver
#Check your own graphics card and driver versions that can be installed
sudo ubuntu-drivers devices
Install according to the recommended driver version, mine is 470 ( CUDA9 requires driver version 384 and above, so it is OK ).
First remove the old NVIDIA driver:
sudo apt-get purge nvidia-*
then install
sudo add-apt-repository ppa:graphics-drivers/ppa
sudo apt-get update
sudo apt-get upgrade
#Install
sudo apt install nvidia-driver-470
Just restart .
4. Install other import packages
sudo apt-get install g++ freeglut3-dev build-essential libx11-dev libxmu-dev libxi-dev libglu1-mesa libglu1-mesa-dev
5 Download and install cuda9 toolkit and cuDNN7.6.5
- cuda 9 installation:
# Installation package download
wget https://developer.nvidia.com/compute/cuda/9.0/Prod/local_installers/cuda_9.0.176_384.81_linux-run
# Make the downloaded file executable
chmod +x cuda_9.0.176_384.81_linux-run
# Install
sudo ./cuda_9.0.176_384.81_linux-run --override
Answer the following questions when installation begins:
You are attempting to install on an unsupported configuration. Do you wish to continue? y
Install NVIDIA Accelerated Graphics Driver for Linux-x86_64 384.81? n
Install the CUDA 9.0 Toolkit? y
For the installation process, please refer to the following article: Installing cuda-9 on Ubuntu
- Install matching cuDNN 7.6.5 for cuda
To download cudnn, you must first register on the official website, official website address: https://developer.nvidia.com/cudnn
Register and log in, then find Download cuDNN v7.6.5 (November 5th, 2019), for CUDA 9.0 in Archived cuDNN Releases (cuDNN archived releases) and download it.
Extract cuda directory
# Switch to the download directory
cd ~/Downloads
# Unzip
tar -xzvf cudnn-9.0-linux-x64-v7.3.0.29.tgz
Copy the unzipped files to the cuda toolkit directory
sudo cp -P cuda/include/cudnn.h /usr/local/cuda-9.0/include
sudo cp -P cuda/lib64/libcudnn* /usr/local/cuda-9.0/lib64/
sudo chmod a+r /usr/local/cuda-9.0/lib64/libcudnn*
Set environment variables
# Add cuda-9/bin to the path
echo 'export PATH=/usr/local/cuda-9.0/bin:$PATH' >> ~/.bashrc
# Add cuda-9/lib64 to the library path
echo 'export LD_LIBRARY_PATH=/usr/local/cuda-9.0/lib64:$LD_LIBRARY_PATH' >> ~/.bashrc# Close the environment variable file .bashrc, and enter the environment variable update command
source ~/.bashrc in the terminal
Add symlinks for gcc and g++
sudo ln -s /usr/bin/gcc-6 /usr/local/cuda/bin/gcc
sudo ln -s /usr/bin/g++-6 /usr/local/cuda/bin/g++
Restart your computer to activate changes
Verify successful installation
nvidia-smi
# Check the cuda version, which is shown as 9.0.176
nvcc -V
# Check the cuDNN version, which shows 7.6.5
cat /usr/local/cuda/include/cudnn.h | grep CUDNN_MAJOR -A 2
3.anaconda3 installation
This article installs Anaconda3-5.3.0-Linux-x86_64.sh. Before installation, make sure that python3.7 is installed. If not, install python3.7 first. If pip is not installed, you can install it by following the link below: https://blog.csdn.net/hymanjack/article/details/80285400
# Install python3.7
sudo apt install python3.7
# Let both pip and python point to python3
gedit ~/.bashrc
#Add these two lines at the end of the file, save and exit
alias pip=/usr/local/bin/pip3.7
alias python=/usr/bin/python3.7# Update environment
source ~/.bashrc
Download the Anaconda installation package
Terminal download
wget https://repo.anaconda.com/archive/Anaconda3-5.3.0-Linux-x86_64.sh
# It may happen that wget is not installed. Install wget first and then download it again.
Web page download
Installation reference: Super detailed Ubuntu installation Anaconda steps + Anconda common commands-CSDN Blog
After the installation is complete, install the visual studio option and select no.
Seeing Thank you for installing Anaconda3! means that your installation has been completed.
4. Code environment configuration
Create a virtual environment:
# deeplabv3+ is the environment name, you can choose it yourself
conda create -n deeplabv3+ python=3.7
# Start the installation environment
conda activate deeplabv3+
In the virtual environment, use the command to install the required packages (pytorch, etc.)
#Create the folder requirements.txt on the desktop and put the following files there
touch requirements.txt
#Copy the following files and put them in requirements.txt
torch
torchvision
numpy
pillow
scikit-learn
tqdm
matplotlib
visdom
Installation (may take several hours depending on internet speed)
cd Desktop
pip install -r requirements.txt
After the installation is complete, you can see the packages you have installed below.
View the packages installed in the virtual environment
conda list
Verifytorch
At this point, the environment configuration is successful. Thank you everyone. If you have any questions, please feel free to contact me.