search

NifTK / NiftyNet

[unmaintained] An open-source convolutional neural networks platform for research in medical image analysis and image-guided therapy
http://niftynet.io
Apache License 2.0
1.37k stars 406 forks source link

Feature Request: CONDA, Jupyter, Docker #352

Open agrawal123 opened 5 years ago

agrawal123 commented 5 years ago

NiftyNet is amazing, but getting it going on various computers is quite difficult. Especially since it does not use the latest Tensorflow (1.12 vs. 1.13), finding the appropriate CUDA and dependencies can be difficult.

Have you considered packaging it as a CONDA package? A Docker container for GPU/CPU would be convenient as well. A Jupyter Notebook would make deployment/inference easy. I would not be that much work, but would make this accessible to a much larger audience.

https://www.pugetsystems.com/labs/hpc/How-to-Install-TensorFlow-with-GPU-Support-on-Windows-10-Without-Installing-CUDA-UPDATED-1419/ https://towardsdatascience.com/tensorflow-gpu-installation-made-easy-use-conda-instead-of-pip-52e5249374bc

Thanks!

wyli commented 5 years ago

Thanks for the nice suggestions @agrawal123, actions on this ticket would be:

The existing demo (https://github.com/NifTK/NiftyNet/tree/dev/demos/PROMISE12) could be a good starting point.

At the moment niftynet (v0.5.0) depends on TF 1.12.x which requires CUDA 9, but pip install tensorflow will automatically install TF 1.13.x, requires CUDA 10.

agrawal123 commented 5 years ago

Thank you for the prompt response!

Yes, we internally created a NiftyNet docker built on top of the Nvidia docker for GPU. unfortunately, in Docker, we can only get acceleration in Linux.

We have trained some models and will be publishing two papers referencing your group. :)

We were looking for an easy deployment strategy (end users upload images, we give them segmentations). We tried to contact the DEEPinfer group multiple times with no response. I believe they also just use a Docker to interface with 3d slicer.

Take care! Sumit

Sumit Agrawal, MD, FRCSC Neurotology & Skull Base Surgery Associate Professor

Department of Otolaryngology - Head & Neck Surgery Western University London, Ontario Canada

Get Outlook for iOShttps://aka.ms/o0ukef

From: Wenqi Li notifications@github.com Sent: Tuesday, May 28, 2019 6:23:47 AM To: NifTK/NiftyNet Cc: agrawal123; Mention Subject: Re: [NifTK/NiftyNet] Feature Request: CONDA, Jupyter, Docker (#352)

Thanks for the nice suggestions @agrawal123https://github.com/agrawal123, actions on this ticket would be:

The existing demo (https://github.com/NifTK/NiftyNet/tree/dev/demos/PROMISE12) could be a good starting point.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHubhttps://github.com/NifTK/NiftyNet/issues/352?email_source=notifications&email_token=ADBKO77AZAVEUYIGE45QU4DPXUB3HA5CNFSM4HP7XEZKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODWLVRWI#issuecomment-496457945, or mute the threadhttps://github.com/notifications/unsubscribe-auth/ADBKO745UBPWFTDVO4ZLCT3PXUB3HANCNFSM4HP7XEZA.

agrawal123 commented 5 years ago

One more thing - thank you for that Google Colab link. If we create a Jupyter notebook from a working Conda install, we may be able to use it directly on Google Colab. Then CUDA, cdnn, etc should all be compatible and working.

Get Outlook for iOShttps://aka.ms/o0ukef

From: Sumit Agrawal sumit_kishore_agrawal@hotmail.com Sent: Tuesday, May 28, 2019 8:51:31 AM To: NifTK/NiftyNet; NifTK/NiftyNet Cc: Mention Subject: Re: [NifTK/NiftyNet] Feature Request: CONDA, Jupyter, Docker (#352)

Thank you for the prompt response!

Yes, we internally created a NiftyNet docker built on top of the Nvidia docker for GPU. unfortunately, in Docker, we can only get acceleration in Linux.

We have trained some models and will be publishing two papers referencing your group. :)

We were looking for an easy deployment strategy (end users upload images, we give them segmentations). We tried to contact the DEEPinfer group multiple times with no response. I believe they also just use a Docker to interface with 3d slicer.

Take care! Sumit

Sumit Agrawal, MD, FRCSC Neurotology & Skull Base Surgery Associate Professor

Department of Otolaryngology - Head & Neck Surgery Western University London, Ontario Canada

Get Outlook for iOShttps://aka.ms/o0ukef

From: Wenqi Li notifications@github.com Sent: Tuesday, May 28, 2019 6:23:47 AM To: NifTK/NiftyNet Cc: agrawal123; Mention Subject: Re: [NifTK/NiftyNet] Feature Request: CONDA, Jupyter, Docker (#352)

Thanks for the nice suggestions @agrawal123https://github.com/agrawal123, actions on this ticket would be:

The existing demo (https://github.com/NifTK/NiftyNet/tree/dev/demos/PROMISE12) could be a good starting point.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHubhttps://github.com/NifTK/NiftyNet/issues/352?email_source=notifications&email_token=ADBKO77AZAVEUYIGE45QU4DPXUB3HA5CNFSM4HP7XEZKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODWLVRWI#issuecomment-496457945, or mute the threadhttps://github.com/notifications/unsubscribe-auth/ADBKO745UBPWFTDVO4ZLCT3PXUB3HANCNFSM4HP7XEZA.

dzhoshkun commented 4 years ago

Hi all,

Re Docker

Just wanted to let you know that we are starting to use nvidia-docker in one of our projects for CI-testing GPU-accelerated capabilities on the same test server (GIFT-Little) that is used for NiftyNet's CI tests. From preliminary tests, it appears to be working reasonably well, albeit with a little bit of configuration effort.

I remember there was some discussion around containerising the NiftyNet CI tests. If this is still of interest, I can help set up a new Docker runner on the test server, which then might be useful for this particular feature request as well.