Zerocost4DL - Running 3D UNET on local GPU

Hello folks,

I was wondering if there are any tutorials on running google colab - 3D UNET of Zerocost4DL with my local GPU resources. Currently I have been using the server based GPUs offered by google Colab, but I would like to train/predict using my local GPU as it has more computing capabilties (higher CUDA cores). Any suggestions on how to do this would be very much appreciated!

@Guillaume_Jacquemet

Thanks!

Hi Praveen, it’s nice that you get these working!
Local runtime makes a lot of sense for 3D U-Net. The way forward would be to set up local runtime from Colab following this link:
https://research.google.com/colaboratory/local-runtimes.html

For this you’ll need to have installed Python, CUDA Toolkit and cuDNN.
Create a conda environment:

You can use the requirements.txt from this notebook:

And follow the Colab’s info for local runtime. We’re also working on getting a guide written for this so watch our GitHub as well!

I hope this helps.

Best,

Romain

Great thanks Romain. I will have a go at this today. Great tool by the way!

Hi Romain,

I get an error when I tried to create a conda environment as suggested in the Github instructions. Do you mind help me with this?

Here is the error message:

ERROR: Could not find a version that satisfies the requirement astropy==4.2.1
ERROR: No matching distribution found for astropy==4.2.1

Ok, think I figured it out. Astropy 4.2.1 needs Python 3.7 or higher. So I just ran

conda create -n myzerocostenv python=3.7

instead of

conda create -n myzerocostenv python=3.6.9

and now it installed Astropy. I hope using Python 3.7 does not interfere with anything else.

I also get another error as we go down the requirement list. This time it is:

ERROR: Could not find a version that satisfies the requirement en-core-web-sm==2.2.5
ERROR: No matching distribution found for en-core-web-sm==2.2.5

What could be the issue here?

Hi Praveen,
We now have some simplified requirements.txt files, for 3D -U-Net, please use the one below:

This should solve your issues.
Best,

Romain