DSRL icon indicating copy to clipboard operation
DSRL copied to clipboard

Published 20 hours ago • verified publisher icon Dootmaan

SISR

Open XinruiYuan opened this issue 4 years ago • 2 comments

Hello, maybe SISR is following "Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network" with Sub-Pixel convolution ?

XinruiYuan avatar Oct 28 '20 12:10 XinruiYuan

Hi there, thank you for your advice. I read the paper again and find out that they did mention that SISR follows the design of "Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network". Currently this repo only uses a simple combination of ConvTransposed and Conv for SR and that may be the reason why my results is not so good. Currently I am testing another implementation of FTM and i will update this repo when it's done.

Dootmaan avatar Oct 29 '20 03:10 Dootmaan

Hello, maybe SISR is following "Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network" with Sub-Pixel convolution ?

Hi there. The repo now have a new branch using SubPixel Conv in SISR. Thank you again for your suggestion.

Dootmaan avatar Dec 29 '20 04:12 Dootmaan