microsoft
Query regarding the code
Can you please tell me where in the code teacher model is being updated by the EMA of the student model?
See here https://github.com/microsoft/SoftTeacher/blob/bef9a256e5c920723280146fc66b82629b3ee9d4/ssod/utils/hooks/mean_teacher.py#L60
@MendelXu thank you very much for the prompt reply. I had another question. Can you please tell me where in the code should I make changes to apply strong augmentation to the labelled data and pass it through the student model?
You don't have to change to code, just change the config file. Changing the pipeline for labeled data and wrapping it with MultiBranch, in which one branch is a weak augmentation pipeline and another one is a strong augmentation pipeline. (Similar to the pipeline of unlabeled data and make sure that assign the same tag sup to both branch with ExtraAttrs.
Do I need to apply any transformation matrix on the output of the student model to keep the output aligned with the strong augmented input data? What part of the code should I need to change then?
If I understand your situation correctly, you just want to compute loss on labeled data which is augmented with both weak and strong augmentation, right? Then you don't have to change anything in the model as the ground truth of an augmented image is always aligned with its prediction ( The alignment has been done in the augmentation pipeline).