Could you please implement --tile option to your code?
zelenooki87 opened this issue · comments
Fantastic project but out of memorry for bigger pics.
cant upscale over 3MPX with 12GB GPU Ram
Thank you very much in advance
For the super-resolution of a large image with 2K or 4K pixels,
we can crop the input image into patches and perform SR on each of these sub-images and then merge them together.
You can refer this code to do so.
could you please help and more advice, cause I am not very familiar to python code.
could you please make .ipynb or .py file that can do this for me but in SAFMN project.
results are fantastic, I am very impressed, better than with swinir in most cases
(for inference without ground throuth images)
thanks
Yeah, but can it be a few days later?
I've had a paper going deadline recently.
I expect to add this part of the code around the 15th.
that is OK. keep hard work!
greetz
Hello, the code is available here
You can use this script to achieve memory-efficient forward, but the output may have a checkerboard effect.
python inference/inference_real_safmn.py --large_input
thank you very much