Docker#

WARNING: as for the moment, docker images are not supported / up-to-date

Docker images are provided with the correct software stack:

For FCM pipelines: - PyTorch - Detectron2 - JDE

For VCM pipelines: - PyTorch - CompressAI - Detectron2 - fiftyone - This library (CompressAI-Vision) - VTM

Please take a look at docker/, you will find for each FCM/VCM case:

  • docker-driver.bash : run this file once in order to be able to use nvidia GPUs with docker

  • Dockerfile.* dockerfiles for building images

  • make_image.bash helper script

  • run_image.bash helper script for creating and running a container.

To create image compressai_vision:vcm using Dockerfile just use:

./make_image.bash 1

You should create your own copy of run_image.bash that suits your needs: you can use it to run an interactive notebook, create bind mounts to local directories, run batch jobs, etc.

The default demo version starts a python notebook and tensorboard:

./run_image.bash 1 gpu

File versions#

List of docker images and software versions:

Dockerfile

PyTorch

Detectron2

CUDA

VTM

FFMpeg

OS

Description

Dockerfile.FCM

2.0.0

1.0

11.3

12.0

4.2.7

Ubuntu 20.04

All inclusive image, run with run_image.bash

Dockerfile.VCM

1.8.2

0.4

11.1

12.0

4.2.7

Ubuntu 20.04

Hot-reload CompressAI-Vision (for compressai-vision development), run with run_image.bash.