first draft of unet training script

[hajeressef]

PR to add training script of Unet architecture for instance segmentation. The abstract_trainer used can be found in this PR

This is a sample experiment for a unet experiment on picsellia.

[picsalex]

Thanks for the PR @hajeressef ! Some details I've noticed on the structure:

• The main folder training-unet-instance-segmentation/... is not consistent with the repo (captured-...)
• There's a remaining pyproject.tom file, I guess this one can be removed since we don't use poetry within the base image.
• There's a .ipynb file that can also be removed from the PR.

[hajeressef]

Hi @picsalex . Thank you so much for your feedback.

• for the naming I think we will rethink the renaming of all repos. I removed the captured just to seperate the new refactored directories from the old ones. Since there was also a yolov8 one that I refactored before so it needed a different name.
• we are adding poetry files for local testing purposes only
• @PN-picsell and I discussed adding jupyter notebooks for future reference, in case we need to do future experiments. We could keep them here or think of another location (an outside folder) where we group them all.

[picsalex]

• Okay, makes sense. I guess we could work on another PR to update the folder names.
• Okay, so can I guess we can just add them in the .gitignore?
• I think we should put them somewhere else because right now they are included in the Docker images during the build process. Maybe at the root in training-unet-instance-segmentation/?

[hajeressef]

1. We could perhaps keep the old ones for now with the old naming, and slowly clean up/delete the "captured" repos as they get reworked/refactored into new repos with new names.
2. no I believe we need it to be pushed. The idea is that whoever pulls this repo to experiment more or test without a docker image, can use this poetry file in their environment
3. Makes sense. I will move it to root directory as you mentioned.

[picsalex]

• Okay, maybe can you update the README to mention this ? Or create an issue so we don't forget what we discussed about that matter.
• Same thing, maybe explain this in the README? I think we must be careful here since the dependencies are defined at 2 different places (requirements.txt and pyproject.toml).
• 👍

[PN-picsell]

Looks really nice to me, @picsalex @hajeressef we can meet online to discuss the matters adressed here. Just one think before we merge, can you update the notebook so it looks exactly like the main.py file ? This way it's clear where to start the research back