Closed mike199212 closed 7 years ago
Yes, and it is depends on density of target ref image. I m also not sure about appropriate value, but I used 100 ~ 500 for leak[1]
Thank you for your reply! Could you tell me about your leak[0] and leak[1] to train the second network cnn_x2, i.e. the tuple "leak" in the constructor of Image2ImageDatasetX2?
in my opinion, 8 ~ 16 and 64~128 is ok for x2 leak[0] and leak[1] . but 2nd layer is still not highly optimized.
Thank you again. I will try using these values in training second network.
In the training of cnn_128, the default value for "leak" of Imae2ImageDataset is set to (0,0). However, one needs to set leak[1] to some positive value to let cnn_128 to be able to respond to "ref" (hint colors). What value should I use, or how can I find appropriate value?