otput of SR get much smaller size than input (not 8x problem)
Athenanna opened this issue · comments
thank you for sharing your code as well as your dataset.
i would like to do the SR implementation. however, given an input of size 19201072, the output will be 19201072, while when input is 38402144 (satisfied 8x ), the output will be of size 21121168 instead， W/H also changed .
What are the options to get the same resolution as output? thanks!
Hi! Would you like to share the command you run as well as the file structure where you are running the command?
We haven't tested with input of that high resolution, but the behavior with different resolution (as long as 8x) should be consistent.
Also, I'm wondering what's the memory size of your machine? 1920 x 1072 is very high resolution as SR input...
Thanks!
@anchen1011 ,thank you for your reply. I use th demo.lua -mode sr -inpath ../data/xxx to do the SR, and the input is 3840x2144 (RGB, bicubic upsampling from 1920x1072), the expectde output is 3840x2144, while it is 2112x1168 instead。I notice that the size of input is limited in the loader.py: local npMax = 2500000, when it set to local npMax = 8500000, it crashed. however, i tried cutting the input images into four squares, and SR each square respectively. Is there any better solution?
The reason why it crashed should be memory outage, since 1920 x 1072 is very high resolution as SR input...
Cutting it into smaller squares should be able to generate a reasonable input. However, this might cause abnormal artifacts in boundaries. Introducing some overlapping, averaging, and smoothening might help.
Closed due to inactivity. Feel free to reopen if there is further question~