00. 환경구성 - pineland/object-tracking GitHub Wiki

01. Docker 기반 Nvidia GPU enabled Tensorflow & Jupyter Notebook 설치

01.01 Dockerfile 작성

FROM nvidia/cuda:10.1-cudnn7-devel-ubuntu18.04

RUN apt-get update && yes | apt-get upgrade
RUN apt-get install -y apt-utils build-essential
RUN apt-get install -y python3 python3-pip openjdk-8-jre wget

RUN ln -s /usr/bin/python3 /usr/bin/python
RUN ln -s /usr/bin/pip3 /usr/bin/pip

RUN pip install --upgrade pip

WORKDIR /root/

# install dependency libraries of Tensorflow Object Detection API.
RUN apt-get install -y protobuf-compiler python3-pil python3-lxml git
RUN pip install setuptools --upgrade
RUN pip install tensorflow-gpu==2.2

RUN git clone https://github.com/tensorflow/models.git /root/tensorflow/models

WORKDIR /root/tensorflow/models/research

RUN protoc object_detection/protos/*.proto --python_out=.
RUN export PYTHONPATH=$PYTHONPATH:`pwd`:`pwd`/slim

WORKDIR /root

RUN pip install Jupiter
RUN pip install matplotlib

RUN jupyter notebook --generate-config --allow-root
RUN echo "c.NotebookApp.password = u'sha1:6a3f528eec40:6e896b6e4828f525a6e20e5411cd1c8075d68619'" >> /root/.jupyter/jupyter_notebook_config.py

EXPOSE 8888

# 아래와 같이 해서 컨테이너 생성하면서 바로 주피터 노트북 서버 데몬을 띄워도 되고, 컨테이너 생성 시 끝에 /bin/bash를 붙여 들어가서 해당 디렉토리에서 아래의 명령을 실행해도 된다. 
CMD jupyter notebook --allow-root --ip=0.0.0.0 --port=8888 --no-browser

01.02. 이미지 만들기

$ docker build -t nvidia-cuda10.1-cudnn7-tensorflow-gpu2.2-ubuntu18.04:1.0 .

01.03. 도커 컨테이너 생성

$ docker run --gpus all -it --rm -p 8888:8888 -v `pwd`:/root/workspace nvidia-cuda10.1-cudnn7-tensorflow-gpu2.2-ubuntu18.04:1.0

01.04. 주피터 노트북 실행

웹브라우저를 열고 http://localhost:8888 에 접속한다. 이때, 패스워드는 Dockerfile에서 설정한대로 root 이다.

01.05. tensorflow-gpu 가능여부 확인

# 버전 확인
import tensorflow as tf

print(tf.__version__)
2.2.0

# 그래픽카드, CUDA 등 설치 상태 확인
!nvidia-smi

Thu Jul 23 01:55:03 2020       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 435.21       Driver Version: 435.21       CUDA Version: 10.1     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce GTX 1060    Off  | 00000000:01:00.0 Off |                  N/A |
| N/A   56C    P0    26W /  N/A |    441MiB /  6078MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+                                                                             
+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
+-----------------------------------------------------------------------------+

# 텐서플로우에서 GPU 가용여부 확인
from tensorflow.python.client import device_lib
device_lib.list_local_devices()

incarnation: 14796649298003808976
physical_device_desc: "device: XLA_CPU device",
name: "/device:XLA_GPU:0"
device_type: "XLA_GPU"
memory_limit: 17179869184
locality {
}
incarnation: 17109582847013019115
physical_device_desc: "device: XLA_GPU device",
name: "/device:GPU:0"
device_type: "GPU"
memory_limit: 5475903744
locality {
  bus_id: 1
  links {
  }
}
incarnation: 2258141267897666028
physical_device_desc: "device: 0, name: GeForce GTX 1060, pci bus id: 0000:01:00.0, compute capability: 6.1"]
⚠️ **GitHub.com Fallback** ⚠️