Any docker images can be used to build a demo. For example, one can use a python base image or a debian base image.

However, IPOL also proposes a few prebuilt images for Python demos (recommended). See below of the descriptions of the images.

v2

• registry.ipol.im/ipol:v2-py3.9
• registry.ipol.im/ipol:v2-py3.9-pytorch
• registry.ipol.im/ipol:v2-py3.9-tensorflow
• registry.ipol.im/ipol:v2-py3.10
• registry.ipol.im/ipol:v2-py3.10-pytorch
• registry.ipol.im/ipol:v2-py3.10-tensorflow
• registry.ipol.im/ipol:v2-py3.10-tensorflow-gpu
• registry.ipol.im/ipol:v2-py3.11
• registry.ipol.im/ipol:v2-py3.11-gpu
• registry.ipol.im/ipol:v2-py3.11-pytorch
• registry.ipol.im/ipol:v2-py3.11-pytorch-gpu
• registry.ipol.im/ipol:v2-py3.11-tensorflow
• registry.ipol.im/ipol:v2-octave

ipol:v2-py

Each image is based on the dockerhub python:3.x-bookworm image. The GPU images are based on nvidia/cuda:11.4.3-cudnn8-devel-ubuntu20.04.

The list of preinstalled system packages is available here v2/packages.txt.

The list of preinstalled python packages is available here v2/requirements-*.txt (it is the same for all Python versions).

-pytorch

This flavor adds the following packages:

• torch==2.2.0+cpu (from https://download.pytorch.org/whl/cpu)
• torchvision==0.17.0+cpu (same)
• torchaudio==2.2.0+cpu (same)
• pytorch-lightning==2.2.0

-tensorflow

This flavor adds the following package:

• tensorflow==2.15.0.post1

-gpu

The GPU variants are a bit special. They are built on top of nvidia/cuda:11.4.3-cudnn8-devel-ubuntu20.04 with Python installed from ppa:deadsnakes/ppa.

variants:

• py3.11-gpu: python 3.11, cuda 11.4 (from the base image)
• py3.11-pytorch-gpu: python 3.11, cuda 11.8, same pytorch as described above (2.2.0) but installed with cuda support (from https://download.pytorch.org/whl/cu118), and with xformers==0.0.24
• py3.10-tensorflow-gpu: python 3.10, cuda 11.8, tensorflow[and-cuda]==2.14.1

ipol:v2-octave

This image is based on debian:bookworm with the following octave packages:

Package Name  | Version | Installation directory
--------------+---------+-----------------------
     control  |   3.4.0 | /usr/share/octave/packages/control-3.4.0
       image  |  2.14.0 | /usr/share/octave/packages/image-2.14.0
          io  |   2.6.4 | /usr/share/octave/packages/io-2.6.4
       optim  |   1.6.2 | /usr/share/octave/packages/optim-1.6.2
      signal  |   1.4.3 | /usr/share/octave/packages/signal-1.4.3
  statistics  |   1.5.3 | /usr/share/octave/packages/statistics-1.5.3
      struct  |  1.0.18 | /usr/share/octave/packages/struct-1.0.18

Octave is at version 7.3.0. The packages are not loaded by default, so you must use pkg load image in your code for example.

---

v1

• registry.ipol.im/ipol:v1-py3.7
• registry.ipol.im/ipol:v1-py3.7-pytorch
• registry.ipol.im/ipol:v1-py3.7-tensorflow
• registry.ipol.im/ipol:v1-py3.8
• registry.ipol.im/ipol:v1-py3.8-gpu
• registry.ipol.im/ipol:v1-py3.8-pytorch
• registry.ipol.im/ipol:v1-py3.8-pytorch-gpu
• registry.ipol.im/ipol:v1-py3.8-tensorflow
• registry.ipol.im/ipol:v1-py3.8-tensorflow-gpu
• registry.ipol.im/ipol:v1-py3.9
• registry.ipol.im/ipol:v1-py3.9-pytorch
• registry.ipol.im/ipol:v1-py3.9-tensorflow
• registry.ipol.im/ipol:v1-octave

ipol:v1-py

Each image is based on the dockerhub python:3.x image, which itself is based on Debian bullseye. The images also contain quarto v0.9.106.

System packages

The following packages are preinstalled in every IPOL images, in addition to what is already installed in the python:3.x base image:

cmake
libtiff5-dev
libjpeg-dev
libpng-dev
libfftw3-dev
liblapack-dev
libblas-dev
libopenblas-base
libopenblas-dev
libblas3
liblapacke-dev
liblapacke
libconfig9
libconfig-dev
libconfig++-dev

py3.7

Includes the following additional Python packages from PIP:

pip==22.0.4

numpy==1.21.5
scipy==1.7.3
scikit-image==0.19.2
scikit-learn==1.0.2
opencv-contrib-python-headless==4.5.5.64

pillow==9.0.1
iio==0.0.3
imageio==2.16.1
imagecodecs==2021.11.20

jupyter==1.0.0
matplotlib==3.5.1
plotly==5.6.0
pandas==1.3.5

papermill==2.3.4

py3.8

Includes the following additional Python packages from PIP:

pip==22.0.4

numpy==1.22.3
scipy==1.8.0
scikit-image==0.19.2
scikit-learn==1.0.2
opencv-contrib-python-headless==4.5.5.64

pillow==9.0.1
iio==0.0.3
imageio==2.16.1
imagecodecs==2021.11.20

jupyter==1.0.0
matplotlib==3.5.1
plotly==5.6.0
pandas==1.4.1

papermill==2.3.4

py3.9

Includes the following additional Python packages from PIP:

pip==22.0.4

numpy==1.22.3
scipy==1.8.0
scikit-image==0.19.2
scikit-learn==1.0.2
opencv-contrib-python-headless==4.5.5.64

pillow==9.0.1
iio==0.0.3
imageio==2.16.1
imagecodecs==2021.11.20

jupyter==1.0.0
matplotlib==3.5.1
plotly==5.6.0
pandas==1.4.1

papermill==2.3.4

-pytorch

This flavor adds the following packages (from https://download.pytorch.org/whl/cpu/torch_stable.html):

• torch==1.11.0+cpu
• torchvision==0.12.0+cpu
• torchaudio==0.11.0+cpu

-tensorflow

This flavor adds the following package:

• tensorflow-cpu==2.8.0

ipol:v1-octave

This image is based on debian:bullseye with the following packages:

octave
octave-image
octave-signal
octave-control
octave-io
octave-optim
octave-statistics 

Octave is at version 6.2.0. The packages are not loaded by default, so you must use pkg load image in your code for example.