Auto3dSeg model ensemble combiner

[che85]

Hello,

I am wondering if there is a way to combine all folds of the auto3dseg training into one model with the best rankings combined. Using the model ensemble functionality is tricky if you want to deliver models only to a customer and individual models perform significantly worse than the model ensemble output.

From what I have seen, the current implementation requires the auto3dseg work_dir which has all training information stored in order to make the ranking decisions.

https://github.com/Project-MONAI/tutorials/blob/main/auto3dseg/docs/ensemble.md