frankierr/openpose_containers repository overview

⁠ ⁠

This repository builds Docker and Singularity images for OpenPose compiled with NVCaffe.

It currently has two bases:

1. bionic: Based upon the nvcr.io/nvidia/caffe:20.03-py3 Ubuntu 18.04 LTS (bionic) image provided by NVidia. More info about the base is available from NVidia.⁠ It contains dependencies of OpenPose including NVCaffe and OpenCV. It has been upgraded with a newer versions of some packages and otherwise tweaked to make a viable OpenPose build environment. It has the following versions:

• CUDA 10.2
• Python 3.7 (upgraded version from universe)
• OpenCV 3.4.1 (upgraded version from PPA)

1. focal: Based upon the ubuntu:focal image. It compiles NVCaffe with patches to make it compile in this newer environment. It has the following versions:

• CUDA 10.2
• Python 3.8 (default)
• OpenCV 4.2.0 (default)

Upon these bases OpenPose there are multiple images built. For example, for focal these are the following images:

• focal_cpu: Contains OpenPose built against an Intel MKL optimised CPU Caffe
• focal_nvcaffe: Contains OpenPose built against an NVidia optimised NVCaffe
• focal_multi: Contains both of the above versions, as well as normal GPU Caffe. The entrypoint script automatically configures the correct version of OpenPose.

Upon these Docker bases, Singularity images are runtime are made available.

The multi OpenPose container concept is based upon this blog post by Peter Uhrig⁠.

⁠Get the images

Get the images from Docker Hub⁠ or Singularity Hub⁠.

⁠Rebuilding from templates

To rebuild Singularity files:

1. Directly make your changes to one of the *_template files or a file in snippets.
2. Run make.
3. Add and push both all changes generated files so they can be build on Docker and Singularity Hub

For older versions see: https://github.com/frankier/gsoc2020/commits/master/openpose_singularity⁠