Ah, I didn't know there is Topaz Photo AI.
Codeformer only work with face, it do better with larger face as it
provide more details. With face too small or blur Codeformer may create
something totally wrong.
The background is handled by Real-ESRGAN, it is included in the command
line and combine with faces generated by Codeformer. Real-ESRGAN do
sharpen and noise reduction but no option setting, it usually overdone.
It handle the image block by block, in many cases you will see different
results on some areas of the same images.
But sometimes it do work nicely:
Bejing 1984 OM-10 Sigma 28-70mm
https://accura.com.hk/temp/1984_Bejing-12F.jpg
https://accura.com.hk/temp/1984_Bejing-12FRE.jpg
On the other hand I found Codefromer+Real-ESRGAN do very well for small
digital images:
https://accura.com.hk/temp/Staysafe.jpg
https://accura.com.hk/temp/staysafe_CF.jpg
It handle compression artifacts very well.
C.H.Ling
On 1/2/2023 5:59, Wayne Shumaker wrote:
At 1/31/2023 07:51 AM, C.H.Ling wrote:
I have been using Topaz Sharpen AI for some time, processed over two thousands
images, mostly in Motion Blur mode. Topaz is okay but just can't compare with
Codeformer for face recovery.
C.H.Ling
Thanks for showing this tool. I will have to try it out also...
This was using the new Topaz Photo AI (latest version 1.1.9) not Topaz Sharpen.
I noticed that the texture in the shirt (3rd from right), for instance, is
better with Codeformer. Hence, Codeformer seems to do better, with not just
faces, than Photo AI. I went back and tried adjusting things in Photo AI to no
avail. I could never get the shirt right.
I tried all the Topaz tools on my cat image and Photo AI had artifacts in the
fur.
https://flic.kr/s/aHBqjAouQUI
I later reduced the resolution and got some better results, but overall I am
disappointed.
I have notice with Topaz AI tools that the image resolution can make a
difference. For larger images, such as A7Riii or A7Riv, the results have a lot
of artifacts but if the image size is reduced, the artifacts become less. I was
wondering if the AI model training does not include such high res images. And
if the image is already fairly sharp, Denoize Clear seems to do better and is
much faster.
I'm curious if Codeformer has a similar issue with resolution?
WayneS
--
_________________________________________________________________
Options: http://lists.thomasclausen.net/mailman/listinfo/olympus
Archives: http://lists.thomasclausen.net/mailman/private/olympus/
Themed Olympus Photo Exhibition: http://www.tope.nl/
|
