This was briefly mentioned during the last ivadomed meeting.
The inference script is a nice wrapper for nnUNet. It's hard to see how we could avoid duplicating it in model repositories, but it would be nice to include it because it makes these repos usable in a standalone fashion. 3rd parties can still access the model checkpoint by simply downloading the release asset, but providing an inference script is great for external users. Also, nnUNet is not yet supported in ADS but in the meantime I need a way to share models with collaborators. What happens it that we get variations of the script everywhere, i.e.
A way to centralize everything and to have only 1 script to maintain would be to actually package this small wrapper and list it as a requirement in model repositories. There's even already a requirements file!
valosekj
commented
11 months ago
This was briefly mentioned during the last ivadomed meeting.
The inference script is a nice wrapper for nnUNet. It's hard to see how we could avoid duplicating it in model repositories, but it would be nice to include it because it makes these repos usable in a standalone fashion. 3rd parties can still access the model checkpoint by simply downloading the release asset, but providing an inference script is great for external users. Also, nnUNet is not yet supported in ADS but in the meantime I need a way to share models with collaborators. What happens it that we get variations of the script everywhere, i.e.
A way to centralize everything and to have only 1 script to maintain would be to actually package this small wrapper and list it as a requirement in model repositories. There's even already a requirements file!