A question about training results
xuejuhao opened this issue · comments
Hello, why did the results of training with the same configuration as in the paper turn out to be very poor?
The training configuration is as follows:
epochs = 300
batchsize = 1
patchsize = 256
Training images consist of 1000 pairs selected from LSRW/nikon/.
The testing images are also selected from LSRW/nikon.
May I ask what is the best training configuration. I would be appreciated if you could reply.
The training effect is shown in below with strong drag artifacts.
Maybe you can attempt with bigger patchsize, such as 400 or 512. Empirically, the bigger the patchsize is, the better performance will be.
Maybe you can attempt with bigger patchsize, such as 400 or 512. Empirically, the bigger the patchsize is, the better performance will be.
OK，fine. I will try it. Thank you.