Closed icoz69 closed 6 years ago
Hi @icoz69,
That is actually right. I didn't take inspiration directly from the original DenseNet paper but from https://arxiv.org/pdf/1611.09326.pdf, which uses a small variant for semantic segmentation. This avoid loosing information from the input, which I thought it may help for the case of Super Resolution too.
thanks for your reply. why dont you use the discriminator architecture from the original SRGAN paper?
Fun 😁
sorry? 😄
./train --cuda is it Super resolution for CIFAR-100 , or classification for CIFAR-100
hi, thanks for the code. i have a question about the densenet blocks. i noticed each block returns x which is a combination of previous inputs and y,the output of last conv. however, in this way N desenet blocks are actually one big densenet block. should it return y instead?