20.01.2025, 21:04
Resizers are mainly meant to resize.
Resizers based on machine learning models like RealESRGAN (which itself offers different models) do not just resize, but also change lots of stuff. Depending on what content they were trained on and what they produce.
So depending on the source and the model used in RealESRGAN resizer, it might to more or less.
VSGAN and VSMLRT offer to tons of different models, Hybrids add-ons come with some of them, but there are way more, for example https://openmodeldb.info/.
Since most of these models have no temporal component, they are often not suited to remove temporal noise or artifacts.
For such stuff you can use DPIR, SCUNet, BasicVSR++ and similar stuff.
Aside from these, there are also tons of other denoisers, etc. but those are not supported in Hybrid, since nobody ported them to Vapoursynth or they are simply to slow to be practical for video.
a. does a good job
b. can be used in Vapoursynth
let me know and I might add them.
For example atm. I'm thinking about adding support for https://github.com/TensoRaws/ccrestoration which support some machine learning model architectures, VSGAN and vs-mlrt do not support. The problem is that most of them are really slow and often not practical for video usage.
Cu Selur
Resizers based on machine learning models like RealESRGAN (which itself offers different models) do not just resize, but also change lots of stuff. Depending on what content they were trained on and what they produce.
So depending on the source and the model used in RealESRGAN resizer, it might to more or less.
VSGAN and VSMLRT offer to tons of different models, Hybrids add-ons come with some of them, but there are way more, for example https://openmodeldb.info/.
Since most of these models have no temporal component, they are often not suited to remove temporal noise or artifacts.
For such stuff you can use DPIR, SCUNet, BasicVSR++ and similar stuff.
Aside from these, there are also tons of other denoisers, etc. but those are not supported in Hybrid, since nobody ported them to Vapoursynth or they are simply to slow to be practical for video.
Quote:On github I see other project...If you find a project, which:
a. does a good job
b. can be used in Vapoursynth
let me know and I might add them.
For example atm. I'm thinking about adding support for https://github.com/TensoRaws/ccrestoration which support some machine learning model architectures, VSGAN and vs-mlrt do not support. The problem is that most of them are really slow and often not practical for video usage.
Cu Selur
----
Dev versions are in the 'experimental'-folder of my GoogleDrive, which is linked on the download page.
Dev versions are in the 'experimental'-folder of my GoogleDrive, which is linked on the download page.