search

xinntao / Real-ESRGAN-ncnn-vulkan

NCNN implementation of Real-ESRGAN. Real-ESRGAN aims at developing Practical Algorithms for General Image Restoration.
https://github.com/xinntao/Real-ESRGAN
Other
1.46k stars 182 forks source link

Images come out wrong when using -s 2 #19

Open Ktr4ks opened 2 years ago

Ktr4ks commented 2 years ago

While using the default scale of 4x, the output images come out correctly, but while using 2x something weird is happening. Have a look:

Input Tom

Output 2x outputtom2x

Output 4x outputtom2x

NaokiSato102 commented 2 years ago

me too

x1 (orig)
x1

x2
x2

x3
x3

x4
x4

NaokiSato102 commented 2 years ago

I would like to have 1x increase (to do noise reduction only) while this is being addressed.

ghost commented 1 year ago

when i use realesrgan-x4plus-anime with -s 8 it also gets messy!

AlphaHasher commented 1 year ago

Can confirm this is still happening

The -s argument only works for the default realesr-animevideov3 model and not for anything other that is provided

all other models are stuck at 4x scale

Steps to recreate:

realesrgan-ncnn-vulkan.exe -i input.jpg -o output.png -n realesrgan-x4plus -s 2

or

realesrgan-ncnn-vulkan.exe -i input.jpg -o output.png -n realesrgan-x4plus-anime -s 2

Output (for both): output

AlphaHasher commented 1 year ago

I narrowed this issue down to being an issue. It is because the model being used for is a 4x model which means that it can only be used for 4x scale

zetyquickly commented 1 year ago

I also created a duplicate issue https://github.com/xinntao/Real-ESRGAN-ncnn-vulkan/issues/39

The fix for me was to use -f JPG