[Det] Environment Setup

Ubuntu Setting

• Ubuntu 20.04 Download: https://ubuntu.com/download/desktop

• 우분투 기본 설치
  • sudo apt update && sudo apt upgrade -y
  • sudo apt install build-essential cmake git curl

• NVIDIA driver: sudo ubuntu-drivers autoinstall

• CUDA Download: https://developer.nvidia.com/cuda-toolkit-archive
  • Currently(2021.02), download CUDA 11.0 for Tensorflow 2.4 and Pytorch 1.7
  • chmod a+x cuda_<version>.run
  • sudo ./cuda_<version>.run
• cuDNN: https://developer.nvidia.com/rdp/cudnn-archive

  • NVIDIA 계정 로그인

  • Select “cuDNN-xxx for CUDA 11.0”

  • Download “cuDNN Library for Linux (x86_64)”

  • tgz 압축해제

  • 헤더와 라이브러리 복사

    sudo cp <extracted directory>/cuda/include/* /usr/local/cuda/include
    sudo cp <extracted directory>/cuda/lib64/* /usr/local/cuda/lib64
    

  • ~/.profile에 아래 내용 없으면 추가

    export LD_LIBARARY_PATH="/usr/local/cuda/lib64:${LD_LIBARARY_PATH}"
    export PATH="/usr/local/cuda/bin:${PATH}"
    

• Pyenv: https://github.com/pyenv/pyenv-installer

  • pyenv prerequisites: https://github.com/pyenv/pyenv/wiki/Common-build-problems

  • curl https://pyenv.run | bash

  • add the three lines to ~/.bashrc

    export PATH="$HOME/.pyenv/bin:$PATH"
    eval "$(pyenv init -)"
    eval "$(pyenv virtualenv-init -)"
    

• PyCharm: sudo snap install pycharm-community --classic

• 한글 설치

  • Settings -> Region & Language -> Add “Korean(101/104 key compatible)” -> Manage Installed Languages -> Update
  • sudo apt install fcitx-hangul
  • Settings -> Region & Language -> Manage Installed Languages -> Keyboard input method system: “fcitx”
  • 재부팅
  • 한/영키 등록: 오른쪽 위 키보드 아이콘 -> Configure -> “+” -> “Hangul” 추가 -> Global Config 탭 -> Trigger Input Method
• Naver Whale: https://whale.naver.com/ko/download

Pyenv Setting

Pyenv를 이용한 파이썬 설치

pyenv install 3.8.x (latest)

Setup Virtual Environment

pyenv virtualenv 3.8.x py38dl
mkdir -p ~/workspace/detlec
cd ~/workspace/detlec
pyenv local py38dl
pip install --upgrade pip

pip install numpy opencv-python scikit-learn matplotlib pydot
sudo apt install graphviz

pip install tensorflow==2.4

pip install torch==1.7.1+cu110 torchvision==0.8.2+cu110 \
torchaudio===0.7.2 -f https://download.pytorch.org/whl/torch_stable.html

Test TF2 and Pytorch

• Open PyCharm at ~/workspace/detlec

• Set pycharm interpreter at ~/.pyenv/versions/py38dl/bin/python

• Test tensorflow by

  ```python import tensorflow as tf x = tf.random.uniform((2, 3, 4)) print(tf.reduce_sum(x, axis=(1, 2)))

Output: tf.Tensor([4.5362177 5.611371 ], shape=(2,), dtype=float32)

- Test pytorch by

  ```python
  import tensorflow as tf
  x = tf.random.uniform((2, 3, 4))
  print(tf.reduce_sum(x, axis=(1, 2)))
  
  ---------------------------
  Output:
  tensor([8.4671, 4.6608])

Troubleshooting

우분투 반응 느림

• RTX 2070 이 장착된 랩탑에서 Ubuntu 20.04를 설치했더니 자꾸 반응이 느려지는 현상 발생

• 키보드나 마우스 입력 후 3~4초 후에 반응함
• Ubuntu 18.04에서는 이상없음
• Ubuntu 18.04에서 20.04로 업그레이드를 해도 같은 현상 발생
• software-properties-gtk에서 그래픽 드라이버 버전을 460에서 450으로 수정했더니 다시 빨라짐
• CUDA 11.1까지는 450 버전에서 작동