For our own projects, we make a lot of use of Docker, to better manage the challenging requirements for CUDA and Python libraries. You can obtain the images from two sources:

In-house Registry

The Docker images are hosted on campus on our own infrastructure, with any base images cached here as well. From within New Zealand, this could result in faster downloads.

More...

Docker Hub

Using Docker's hosting facility, this approach will give you access to our Docker images anywhere in the world.

More...

For some of the frameworks, we have tutorials available on how to use them in practice. This includes data preparation, training and making predictions. Check out our Applied Deep Learning site for more details.