Data Augmentation using the batchgenerators-framework in nnUNet

[sabinevater]

Dear nnUNet-team,

I hope you are doing well today.

In your paper "nnU-Net for Brain Tumor Segmentation" from 2020 ( https://arxiv.org/abs/2011.00848 ) you made several changes to the standard parameters for data augmentation in nnUNet in order to improve performance:

• increase the probability of applying rotation and scaling from 0.2 to 0.3.
• increase the scale range from (0.85, 1.25) to (0.65, 1.6)
• select a scaling factor for each axis individually
• use elastic deformation with a probability of 0.3
• use additive brightness augmentation with a probability of 0.3
• increase the aggressiveness of the Gamma augmentation

You furthermore wrote that you used the batchgenerators - framework. I use nnunetv2 and wanted to ask, where exactly in the code could I apply these changes ( I did not find anything regarding the scaling factors for the axis, the additive brightness augmentation and the Gamma augmentation aggressiveness in the code)?

Kind regards