Dogway's Filters Packs

[rgr]

I noticed a long time ago and mentioned here that QTGMC+ simply works much worse than QTGMC.

[hello_hello]

Originally Posted by junah

Perhaps try without gpuid=-1 to make comparable examples.

I'm pretty sure they're comparable the way they are.

QTGMC+

# Core plugins:
# ResizersPack (for nnedi3=true)
# NNEDI3CL (for nnedi3=true and gpuid=0) (requires at least AVS+ v3.7.3)
# nnedi3 (for nnedi3=true and gpuid=-1)

Originally Posted by rgr

I noticed a long time ago and mentioned here that QTGMC+ simply works much worse than QTGMC.

It's odd because for fairly good quality video (interlaced or progressive) they seem fairly similar, but I think I'll go back to using the original QTGMC for the moment, just to be safe.

[rgr]

Originally Posted by hello_hello

It's odd because for fairly good quality video (interlaced or progressive) they seem fairly similar, but I think I'll go back to using the original QTGMC for the moment, just to be safe.

I tested it once and found that the + was faster, but too light. It seemed like its filters were much weaker. It's a pity, because at 10 bits it was significantly faster.

[hello_hello]

The sample I uploaded contains a lot of shimmering, although I'm not sure why (the video is only like that in a few places), so I assume it's QTGMC's shimmer repair doing most of the work stabilizing it, and maybe that isn't working correctly in the QTGMC+ version, or something....

[WAusJackBauer]

Hello all. I've used this SMDegrain for a while but after updating it, it seems to have changed a bit. I'm looking for recommendations (hopefully from Dogway) on what settings to use as a one size fits all thing even though I know it's better to fine tune settings depending on the source video.

My main priority is minimal ghosting/smearing/artifacts and my second priority is removal of most noise but a slight amount remaining isn't a big problem. The content I encode is raw bluray movies, so the videos are 1920x1080 and they're live action. My settings currently look like this:

SMDegrain(tr=3, thSAD=400, RefineMotion=true, contrasharp=true, plane=4, prefilter=6, chroma=true)

I'm wondering if anyone has any suggestions on different settings to use that would give better results as I don't have a lot of knowledge on the parameters.

[DTL]

thSAD setting depends only on the relative noise level in your content. It is generally only protection setting from too bad blends and smoothing. The main adjustment of linear temporal nosie reduction is tr-value. To get lowest level of distortions you typically need first select your working tr-value (balance between performance and degrain level) and slowly increase thSAD (from very low values where it is not degrain at all) to adjust balance between degrain and smoothing. If you lucky you have some range of thSAD where degrain already enough and details smoothing is not yet started. If not - you need to adjust thSAD between either better degrain or better details saving.

Some time ago I found there is a simple API call in AVS+ allow to print debug data on output frame - it is very highly wanted feature for adjusting lots of params for mvtools (MAnalyse + MDegrain). But still not used.

[TR-9970X]

Originally Posted by WAusJackBauer

Hello all. I've used this SMDegrain for a while but after updating it, it seems to have changed a bit.

SMDegrain(tr=3, thSAD=400, RefineMotion=true, contrasharp=true, plane=4, prefilter=6, chroma=true)

I'm wondering if anyone has any suggestions on different settings to use that would give better results as I don't have a lot of knowledge on the parameters.

I have one specific question - I recently looked into thSAD and was thinking of changing it to 500 because it removes slightly more noise than 400 which I feel is nicer for some extra noisy scenes but I'm worried that it will cause ghosting/smearing. Is thSAD 500 fine to use for my type of content?

Here's one that use a LOT, but does get changed depending on the footage quality.

video=SMDegrain(video,tr=1,thSAD=100,thSADC=50,contrasharp=true,prefilter="MinBlur3",str=5.0,refinemotion=true)

video=SMDegrain(video,tr=2,thSAD=200,thSADC=100,contrasharp=true,prefilter="MinBlur3",str=5.0,refinemotion=true)

etc, etc

I use RipBot264, but for most other app's, you'd remove the video from the call.

[Emulgator]

And may I add:
To whom prefilter=6 (KNLMeansCL) is too smeary (to me it was) I can suggest prefilter=8 (BM3D).
Well worth the effort.

[TR-9970X]

Originally Posted by Emulgator

And may I add:
To whom prefilter=6 (KNLMeansCL) is too smeary (to me it was) I can suggest prefilter=8 (BM3D).
Well worth the effort.

I kind of agree. I had forgotten what #6 was.

But doesn't BM3D work best with a hi end nVidia GPU ??

I tried it years ago, and I wasn't impressed, but I think I know why.

[Emulgator]

Yes, BM3D is resource hungry, RTX3080 here.
BM3D became my choice for the least harming treatment, keeping a filmic impression, subtly improving compressibilty.

[tormento]

Originally Posted by TR-9970X

But doesn't BM3D work best with a hi end nVidia GPU ??

1660 Super here.

Remember to disable BM3D fast parameter when used in combination with cpu filter, i.e. almost always.

[WAusJackBauer]

Originally Posted by Emulgator

And may I add:
To whom prefilter=6 (KNLMeansCL) is too smeary (to me it was) I can suggest prefilter=8 (BM3D).
Well worth the effort.

Yeah I heard prefilter 8 was a good one but I have an AMD GPU unfortunately.

I mainly used prefilter 6 because it apparently utilizes a GPU a little bit so it was slightly faster than the others. But if you believe a different prefilter would be better for my content, please let me know

[TR-9970X]

Originally Posted by WAusJackBauer

Yeah I heard prefilter 8 was a good one but I have an AMD GPU unfortunately.

I mainly used prefilter 6 because it apparently utilizes a GPU a little bit so it was slightly faster than the others. But if you believe a different prefilter would be better for my content, please let me know

That's unfortunate

You can use BM3D with a CPU, but what's the point in that.

So what app are you using ??

Have you tried my script suggestion ?

[LeXXuz]

Originally Posted by WAusJackBauer

Yeah I heard prefilter 8 was a good one but I have an AMD GPU unfortunately.

No shame in that. KNLMeansCL is a great denoiser if set up correctly.
I'd pick it over BM3D any day.
And it runs much better on AMD GPUs than on comparable Nvidia cards.

Keep in mind those prefilter presets are merely defaults for different kinds of noise. Those are for people who can barely create their own Avisynth script.
If you want to get the optimum out of SMDegrain, nothing beats your very own prefilter setup.

Just my 2 cents.

[tormento]

Originally Posted by LeXXuz

KNLMeansCL is a great denoiser if set up correctly. I'd pick it over BM3D any day.

Let’s agree to disagree

Show me a source and your scripts to prove me that BM3D, especially in temporal+spatial flavor, is worse than klnmeanscl.

[LeXXuz]

Originally Posted by tormento

Let’s agree to disagree

Show me a source and your scripts to prove me that BM3D, especially in temporal+spatial flavor, is worse than klnmeanscl.

Prove? There's nothing to prove here. I said I'd pick NLM over BM3D any day for my personal taste and needs.

Why do you want to do inferior temporal denoising on a clip you'll feed to an excellent temporal denoiser afterwards?

As a standalone filter, I agree with you because KNLMCLs temporal denoising is utter garbage. And that's why I'd never use prefilter-6 setting in SMDegrain.

Let SMDegrain do the temporal denoising and give it a little help with a custom spatial denoiser, if needed.

And here I pick NLM over BM3D because it does a good job in the spatial domain, it is fast and needs way less CPU resources and much less VRAM on AMD.

Which brings me back to my last post where I said it is no shame to have an AMD GPU. OpenCL apps run faster on AMD and need way less VRAM because Nvidia is too dumb to code proper memory management for OpenCL, especially on RTX GPUs.
People complain in NV forums for a long time, but they don't care because they want devs to use CUDA of course.