[Not an issue] Run sbb_pixelwise_segmentation in a docker image

[johnlockejrr]

If anybody is like me on CUDA 12.7 and can't train a model on keras v2 because is on keras v3 you can do it via a docker image:

docker pull nvidia/cuda:11.8.0-cudnn8-devel-ubuntu22.04

docker run --rm --name ubuntu22.04-nvidia-cuda11.8 --ipc=host -v E:/Docker/mount/sbb_train:/home/incognito/sbb_pixelwise_segmentation/train:rw -it --gpus=all nvidia/cuda:11.8.0-cudnn8-devel-ubuntu22.04 bash

• I added a user incognito where I cloned the sbb_pixelwise_segmentation repo, made an virualenv, installed all the dependencies, then all the training I did in /home/incognito/sbb_pixelwise_segmentation/train so will be saved when shutting down the container.

Here is the container I used: https://hub.docker.com/layers/nvidia/cuda/11.8.0-cudnn8-devel-ubuntu22.04/images/sha256-5079d0d0f36cc63050a0f5c010d769c68b8e959c2d2a45f8ec44dd7e5c1bf7f9

[cneud]

Hi @johnlockejrr - first of all, thanks for your interest and contributions in https://github.com/qurator-spk/sbb_pixelwise_segmentation/issues/17.

Our plan is to integrate the codebase from the unifying-training-models branch as train into the main codebase of Eynollah - e.g. to have the same (and more up-to-date) versions for dependencies and to reuse code. Once this is done (in a few weeks), there will automatically be also container builds available via docker pull ghcr.io/qurator-spk/eynollah:latest. Meanwhile, thanks for publishing a CUDA 12 compatible image.

[johnlockejrr]

Happy to hear that! I love your project, it is really good and has even much potential.