Zhedong Zheng
Zhedong Zheng
It is unreasonable to use a too large stride step like 14. It means every time you sample a pixel then you need to skip 14 pixel to sample the...
Or do you mean padding 14?? You can do it by change the pad setting like [7,7,7,7]. It will pad 14 pixels in width and height.
Since we will cut the edge 8+4+4=16 pixels, I think `9:end-8` is beter. Noted that output size = (input size - filter size)/stride+1
@star4s You can find it in https://github.com/layumi/2016_super_resolution/blob/master/train_SRnet.m#L78
You may refer to https://github.com/layumi/2016_super_resolution/blob/master/SRnet.m I use 9-1-5. Note that if you change to 9-5-5, you should change the output size in the training code, since the output size will...
@star4s I just used the learning Rate setting in my github code. How does your curve look like? Is it similar to mine? (P.s. Different machines may have different random...
Hi @star4s I just use one GTX1080 GPU card although there are 4 GPU cards on my school machine. I use `opts.train.gpus=3` to select the third card to run rather...
weightDecay is regularization term. If you have learned SVM, you will know the loss function is: `(y-Wx)^2 + gama*||W||` weightDecay is gama. In CNN, it works in a similar way...
@star4s 1. The network is hard to interpret. In fact, I just found the loss converged in 4 epochs so I stop it. If there is a reason for better...
@star4s Do you change any training code? The result seems like only show the green channel.