Consulting about details for training noiseprint++

[TronForever]

Hello,
Your work has greatly inspired me, but I encountered some issues when replicating your noiseprint++ module. My model struggles to learn how to extract noise. I would like to consult you about the training details of the noise extractor, especially the contrastive learning loss function you used, the proportion of various image augmentations (image editing) used during training, and the data scale after you perform image augmentations (image editing).
Thank you!