Question: mirroring feet data as input for nnUNet

[luuknoordam]

Dear Fabian and team,

I am currently using nnU-Net for a segmentation task on foot CT data, and I had a short question about mirroring augmentation.

I have also come across papers on foot and ankle imaging where researchers mention mirrored and non-mirrored cases, which made me wonder about the added value of explicitly including mirroring for this type of anatomy in the train and validation set (paired cases kept together in splits_final.json).

Since the data consists of feet, mirroring could easily create a larger and anatomically correct dataset. I noticed the mirror_axes lines in nnUNetTrainer.py and was wondering whether this already covers this type of mirroring as part of the standard data augmentation pipeline.

Thank you in advance for your time!