Avisynth sharpening filters in Hybrid for vhs capture?

[Leanoric]

still nothing

That's weird! You are running the 64bit EXE file? Check Help>About. The 64bit version is identified by "Compiled with MS Visual Studio 2008 for AMD64" in the blurb.

All sorted The vdf from the last folder worked, XP-SSE2.

[Alwyn]

[rgr]

Yes, I know.

It can help, sometimes, in combination with AVS/VS

1. You should never, ever use VEAI for denoising. It's just terrible.
2. But it can be used to correct already denoised images (after Avisynth or Neat) and the effects may (but do not have to) be quite good.

This water on VEAI looks nice in the photo, even very nice, but even in my video I saw that it is sometimes unnatural. VEAI unfortunately does not perform well on VHS, but that is not the fault either. Maybe in 10-20 years...

(If this were my video, I would probably try High Quality for the source file)

[rgr]

This is then a comparison between a full AviSynth/VapourSynth processing (yours) and Topaz VEAI (helped by AviSynth) as in my previous attempt.

The winner is evident (just look to the eyes, even if the background is more "defined" in Topaz)

https://imgsli.com/MjM5MTU3

The guy in Topaz is 10 years older

(never use Topaz to correct your face - noted )

[lollo]

1. You should never, ever use VEAI for denoising. It's just terrible.
2. But it can be used to correct already denoised images (after Avisynth or Neat) and the effects may (but do not have to) be quite good.

Which is exactly what I wrote and what I did

VEAI unfortunately does not perform well on VHS, but that is not the fault either. Maybe in 10-20 years...

We all know

The guy in Topaz is 10 years older
(never use Topaz to correct your face - noted )

[Leanoric]

@lollo Do you use the default settings for TemporalDegrain2?

[lollo]

Yes, except for degrainTR=3 instead of default at 1. Be careful that different revisions of TD2 have different defaults. Experiment a bit with the key parameters with your source to find the best settings.

[Leanoric]

Cheers, will do.

[lollo]

The main parameters are grainLevel, degrainTR, postFFT and postSigma.

From TD2 doc:

# - degrainTR (1), temporal radius of degrain, usefull range: min=default=1, max=fps/8. Higher values do clean the video more, but also increase probability of wrongly identified motion vectors which leads to washed out regions
# - grainLevel (2), if input noise level is relatively low set this to 0, if its unusually high you might need to increase it to 3. The right setting must be found using grainLevelSetup=true while all other settings are at default. Set this setting such that the noise level is lowest.
# - grainLevelSetup (false), only to be used while finding the right setting for grainLevel. This will skip all your other settings!
# - postFFT (0), if you want to remove absolutely all remaining noise suggestion is to use 1 (ff3dfilter) or for slightly higher quality at the expense of much worse speed 3 (dfttest). 2, 4 and 5 are GPU based versions.
# - postSigma (1), increase it to remove all the remaining noise you want removed, but do not increase too much since unnecessary high values have severe negative impact on either banding and/or sharpness

I rarely use post-processing, but option postFFT=4 (KNLMeansCL) is not too bad if additional denoise is needed.

[rgr]

2, 4 and 5 are GPU based versions.

A little more?

[lollo]

2, 4 and 5 are GPU based versions.

A little more?

If the question is specific to usage of GPU by TD2, better ask to ErazorTT here https://forum.doom9.org/showthread.php?t=175798.

If you want more options , just set TD2 postprocessing to nil and use your favourite (spatial) denoiser after.

If it was just a comment, then

[Leanoric]

I rarely use post-processing, but option postFFT=4 (KNLMeansCL) is not too bad if additional denoise is needed.

Ive had a play around and default settings seem to work for me. PostFFT makes it about 3x slower so I keep it off. I have sharpen off as well as I guess theres no need when using LSFmod.

What about QTGMC denoising, do you have that turned off when using TD2? Ive tried them both together and I cant work out which one looks better.

Ive added a deflicker filter now, theres still flicker there but I dont think its looking too bad.

[lollo]

What about QTGMC denoising, do you have that turned off when using TD2? Ive tried them both together and I cant work out which one looks better.

QTGMC denoises by itself, while rebuilding the frames from the fields, in accordance to the temporal data across them.
Many, myself included, tends to reduce the "extra" denoise operations not necessary to the previous to the minimum, because specific denoiser are more effective, and having 2 denoise steps often introduces plastic look.

One of the setting I use most is:

QTGMC(preset="slow", matchpreset="slow", matchpreset2="slow", sourcematch=3, tr1=2, tr2=1, NoiseTR=2, sharpness=0.1)

inspired to some indications from Skiller.

Once more, experiment the different combinations, the final result depend on the source, and there is no a "best" way always valid.

[rgr]

By the way:

Has anyone compared LSFplus with LSFmod?

### LSFplus features and changes (compared to LSFmod v2.193):
### - Frame properties support
### - Increased performance (+50~75%)
### - HD support
### - PC range aware
### - Better edge detection algorithm
### - Better preblur algorithm
### - Better supersampling kernels
### - Higher kernel support (5x5)
### - More limiting modes
### - More and optimized defaults/presets

And it works with YUV422 or YUV444 also at depth>8bit.

Note that the default strength value for LSF+ is 50 (and not 100 as for LSFmod)

Edit: link https://github.com/Dogway/Avisynth-Scripts/blob/master/MIX%20mods/LSFplus.avsi

[lollo]

Has anyone compared LSFplus with LSFmod?

On my "to do" list...

And it works with YUV422 or YUV444 also at depth>8bit

Not sure if internally there is a YV12 conversion, I need to check the code. >8bit is useless for VHS if you mean input stream to the filter, maybe for internal post-processing.

[rgr]

And it works with YUV422 or YUV444 also at depth>8bit

Not sure if internally there is a YV12 conversion, I need to check the code. >8bit is useless for VHS if you mean input stream to the filter, maybe for internal post-processing.

I use a depth greater than 8 bits (9, 10) with more filters. It avoids or minimizes rounding.

There are only 3 occurrences of "convert" (and only ConvertBits).

            pre       = pre.ConvertBits(8, dither=-1, fulls=!tv)

[...]
        cb1     = repair(pre.MCompensate(srchSuper, cb1, thSAD=220, thSCD1=round(6.25*8*8), thSCD2=round(2.03*8*8), mt=true), pre, 3, -1).ConvertBits(bi, fulls=!tv)
        cf1     = repair(pre.MCompensate(srchSuper, cf1, thSAD=220, thSCD1=round(6.25*8*8), thSCD2=round(2.03*8*8), mt=true), pre, 3, -1).ConvertBits(bi, fulls=!tv)

[lollo]

I use a depth greater than 8 bits (9, 10) with more filters. It avoids or minimizes rounding.

Yes, post-processing.

There are only 3 occurrences of "convert" (and only ConvertBits).

I'll check, original LimitedSharpen(Faster) version by Dideè was processing YV12 for a reason, I suppose....

[rgr]

I use a depth greater than 8 bits (9, 10) with more filters. It avoids or minimizes rounding.

Yes, post-processing.

Yes, and it is effective, especially in minimizing banding.

Edit: because I deleted it from the post --> And this is only with one filter: TemporalDegrain2 (+ApplyGradationCurve).

https://imgsli.com/MjQxNjcy

There are only 3 occurrences of "convert" (and only ConvertBits).

I'll check, original LimitedSharpen(Faster) version by Dideè was processing YV12 for a reason, I suppose....

There were probably limitations in the libraries that lsfmod uses. So far I haven't noticed any differences between lsfmod and lsfplus.