I'm trying to train pix2pix using with training set consisting of two types of medical images that have some small differences in image properties (primarily contrast and noise), but this does not seem to be working very well. The network performs well when the image types are used separately for training, but gives poor results when trained with both image types at the same time. Does anyone have any suggestions for handling this kind of problem? Perhaps some suggestions for additional pre-processing of the images?
I'm trying to train pix2pix using with training set consisting of two types of medical images that have some small differences in image properties (primarily contrast and noise), but this does not seem to be working very well. The network performs well when the image types are used separately for training, but gives poor results when trained with both image types at the same time. Does anyone have any suggestions for handling this kind of problem? Perhaps some suggestions for additional pre-processing of the images?