Closed jcohenadad closed 11 months ago
We're sticking with the following data augmentation methods:
I have noticed that adding more augmentations only increases the time per epoch with incremental improvements in performance.
EDIT: The transforms here are not up-to-date, please refer to the the next comment below.
Here are the data augmentation transforms for the final model:
https://github.com/sct-pipeline/contrast-agnostic-softseg-spinalcord/blob/1976cec36162f8dd1ce200cee13a3bf5493cbd9c/config_templates/meanGT_soft.json#L118
Ideas: artifacts, eg: