adds Dockerfile

[miguelvr]

Added Dockerfile to build with CUDA

Example build script with defaults (CUDA_VERSION=9.0, CUDNN_VERSION=7):

nvidia-docker build -t maskrcnn-benchmark docker/ 

Example build script with other CUDA and CUDNN versions:

nvidia-docker build -t --build-arg CUDA_VERSION=9.2 --build-arg CUDNN_VERSION=7 maskrcnn-benchmark docker/ 

[facebook-github-bot]

Thank you for your pull request and welcome to our community. We require contributors to sign our Contributor License Agreement, and we don't seem to have you on file. In order for us to review and merge your code, please sign up at https://code.facebook.com/cla. If you are contributing on behalf of someone else (eg your employer), the individual CLA may not be sufficient and your employer may need the corporate CLA signed.

If you have received this in error or have any questions, please contact us at [email protected]. Thanks!

[facebook-github-bot]

Thank you for signing our Contributor License Agreement. We can now accept your code for this (and any) Facebook open source project. Thanks!

[fmassa]

I don't have much experience with Docker, but this looks great to me, thanks a lot!

Could you add a note in the bottom of the INSTALL.md file mentioning the docker installation as well?
Maybe with the details that you added in the comment of the PR?

Thanks a lot!

[miguelvr]

Sure thing! You're welcome, thanks for the awesome project!

[robpal1990]

Sure thing! You're welcome, thanks for the awesome project!

This looks great. Could you add one which does not use conda/miniconda as well?

[miguelvr]

Sure thing! You're welcome, thanks for the awesome project!

This looks great. Could you add one which does not use conda/miniconda as well?

why would you want that? It makes no difference because you're in a docker container.

I just used conda because I know it is the preferred method to install pytorch due to the MKL version.

[soumith]

yea i dont think non-conda based Dockerfile is needed.

[soumith]

If there's a way for one dockerfile to start from another dockerfile, we should use this as the base Dockerfile https://github.com/pytorch/pytorch/blob/master/docker/pytorch/Dockerfile

[miguelvr]

If there's a way for one dockerfile to start from another dockerfile, we should use this as the base Dockerfile https://github.com/pytorch/pytorch/blob/master/docker/pytorch/Dockerfile

You can start a docker from an image, not another dockerfile. If you have an image built and hosted on a dockerhub I would be able to do it. But I am seeing that pytorch's image is for CUDA 9.0 only.

[robpal1990]

If there's a way for one dockerfile to start from another dockerfile, we should use this as the base Dockerfile https://github.com/pytorch/pytorch/blob/master/docker/pytorch/Dockerfile

You can start a docker from an image, not another dockerfile. If you have an image built and hosted on a dockerhub I would be able to do it. But I am seeing that pytorch's image is for CUDA 9.0 only.

You can always paste the contents of the dockerfile you created at the end of the pytorch dockerfile that Soumith linked here. Not elegant but should work.

[miguelvr]

@fmassa please check if those instructions are enough

[fmassa]

I think this looks good, but I'll let @soumith comment as well because I don't have much experience with Docker.

[mingruimingrui]

The current Image would be built without NCCL which is needed for fast distributed training.
Might I suggest running
apt-get update && apt-get install -y --no-install-recommends libnccl2=2.3.5-2+cuda9.0
under # install basics>

[fmassa]

Thanks a lot!