Icevision: Add Docker support with GPU

[oke-aditya]

🚀 Feature

Add support for containerizing mantisshrimp.
We should create a Dockerfile, and a .dockerignore file

[lgvaz]

That's a great idea, and it should make it very easy to install the lib, this would even solve the problem of pycocotools in windows right?

[oke-aditya]

Yes, it will solve Pycocotools problem. Can you add CI for docker build on Github. So that we can check if container builds with our dockerfile.

[lgvaz]

Can you make a PR for that?😄

[oke-aditya]

I will make PR for the docker build yaml file. But you need to add the secrets.
Check here

[oke-aditya]

Not yet solved with #104 . Will add a direct GPU supported NVIDIA container runtime docker itself.

It will solve all the problems and will not create a multiple dockerfile issue.

[ai-fast-track]

@oke-aditya, @lgvaz , Paperspace provides GPU-enabled container that are pre-loaded with both fastai2 library and the latest fastai course (V4). Check out the references here below, they might be adapted to mantisshrimp.

https://github.com/Paperspace/fastai-docker/tree/master/fastai-v4
https://github.com/Paperspace/fastai-docker/blob/master/fastai-v4/Dockerfile
https://github.com/NVIDIA/nvidia-docker

[oke-aditya]

Yes gonna adapt them 👍 We will have a docker file very similar to them.