pytorch-LapSRN icon indicating copy to clipboard operation
pytorch-LapSRN copied to clipboard

Published 20 hours ago • verified publisher icon twtygqyy

Reproducibility

Open xternalz opened this issue 7 years ago • 4 comments

I can't seem to reproduce the Set5 (x4) result reported, using the default training options with --cuda. Do you have any idea?

xternalz avatar Jun 28 '17 06:06 xternalz

@xternalz Were you able to reproduce x8 result? The training options should be same as x8 as I implemented. You can also test x2's performance when training x4

twtygqyy avatar Jun 28 '17 13:06 twtygqyy

I also can't reproduce the Set5 (x4) result reported, using the default training options with --cuda.

FYXX avatar Jul 08 '17 16:07 FYXX

Hi， I meet the same problem. I train the weight file using the the default training options with --cuda, the results on Set5(x4) were even worse than bicubic. But I can reproduce the result with the weight file you provided in folder model . Can you give me some advice. Thanks a lot.

ZhangDY827 avatar Sep 16 '17 11:09 ZhangDY827

@FYXX @CasdDesnDR Did you generate the training file with the script I provided here https://github.com/twtygqyy/pytorch-LapSRN/tree/master/data ?

twtygqyy avatar Sep 17 '17 01:09 twtygqyy