human-pose-estimation

human-pose-estimation copied to clipboard

This is a first shot at providing a docker version of this project. The objective is to setup a container with cuda backend (using host GPU) and a iTorch notebook demonstration.

Since this is the first time I use docker and torch, I have had some troubles setting up the installation scripts. As a result the docker files are split over three stages, the first two are in my repo: https://github.com/pixelou/nvidia-torch and the pre-built images are at https://cloud.docker.com/app/pixelou/repository/docker/pixelou/nvidia-torch/general The last stage adds the source files from this project and is visible in Dockerfile in this PR.

I am not quite sure the prebuilt images I have pushed on docker cloud will run on any machine (maybe cutorch installation probes the cards during installation?). One solution would be to simply merge all the Dockerfile in one and regenerate the container from scratch for each user. Since it makes container generation slower, I suggest doing this last after all other issues have been investigated and fixed.

Could you please try and see if you can deploy the container yourself?

Also do you have any idea on how to improve the visualization part of the demo notebook?

Thanks a lot! I will have a look tomorrow on it.