Preprocessing 3D Rotation transform


I have been searching through the documentation but haven’t been able to figure out- is there a transform which allows for variable rotation of training samples as a preprocessing augmentation? Ie. randomly rotate image -30->30deg on some axis.

EDIT: I came across ai4med.components.transforms package — Clara Train SDK v3.1 documentation

It seems this will apply the random rotation?


Thanks for your interest in clara train SDK
yes the transform you referred to will do rotation and also crop an ROI depending on the pos / neg ratio
There is also other transformations as RandomRotate2D and RandomRotate3D

We recommend to use the saveAsNiftito debug the augmentation please see the FAQ Question 2 Clara Train FAQ — Clara Train SDK v3.1 documentation

Also, please revert back to the original affine before the segmentation image is written out, if applicable. This is critical in a deploy pipeline where the segmentation image will be wrapped in a DICOM Seg OID referencing the original DICOM instances.