Tensorflow, OpenCV, cvlib with CUDA Docker image

A Dockerfile to build a Docker image which includes:

• Tensorflow with CUDA support,
• OpenCV with CUDA support, cuDNN support and Python 3 bindings,
• opencv-contrib,
• cvlib, a high-level and easy-to-use Computer Vision library for Python.

This image is based on the official NVIDIA Tensorflow Docker image. The Dockerfile is based on a modification of opencv-cuda-docker by Julian Aßmann.

Pre-requisites

• Latest NVIDIA GPU drivers (v.510 at the time of writing)
• Docker Engine
• NVIDIA Container Runtime / nvidia-docker2

Note: You will also need some available storage (more than 30GB is recommended).

How to build

cd <Dockerfile_directory/>
sudo docker build -t cvlib_cuda .

Building the image may take one to several hours depending on your specs.
Multiple warnings may pop up during the compilation of OpenCV.

How to run

sudo docker run -it --gpus all cvlib_cuda

Or, if you want the correct time to be reported in the container (tested on Ubuntu 18.04):

sudo docker run -it --gpus all \
	-v /etc/timezone:/etc/timezone:ro \
	-v /etc/localtime:/etc/localtime:ro \
	cvlib_cuda

How to update

If you wish to change/update the version of Tensorflow used in this image, you may change the version of the base image (line 1 of the Dockerfile).
If you wish to change/update the version of OpenCV used in this image, you may change the version specified after ARG OPENCV_VERSION=.
You may then re-build the image.