Closed AceCoooool closed 6 years ago
I used uniform map.
Thank you kai :+1:
Sorry to ask another question: (I want to achieve the training part) The loss function: (y_pred-y)^2/(noise_level) or [(y_pred-y)/(noise_level)]^2 (through my previous experiment in DnCNN-B, the first one can get similar results in your paper.)
I think pixel-level loss functions such as L1 and L2 can give the same final results. I used this loss function.
Thank you . However, If the batch-size is small. there is a considerable oscillation in loss. (Due to each batch's noise level is far away. ) Yeah, large-batch without this problem.~ Thank you again~ :smile:
Each training data with an uniform map? (or is there some training data with spatially variant noise map in training stage?)
thank you :smile: