Dogway's Filters Packs

[tormento]

Originally Posted by Selur

I just called FFT3DFilter with the default parameters.

Someone knows how to extract the noise frequencies from an image and selectively remove them?

I think it would be really useful and fast with postproduction digital noise, if simple enough.

[DTL]

There were some FFTW-based plugins like DeFreq (?). But it works about good only if you distortion is narrow bandwidth enough (like colour subcarrier). Natural noise is much more complex mixed with useful data and can not be as simply separated. When you run SMDegrain and it blends different frames blocks - it may be equal to special temporal-convolution frequency filter to pass only narrow frequencies and suppress out of band noise. May be same or close idea is implemented in FFT3D plugin via FFT.

[tormento]

Originally Posted by DTL

Natural noise is much more complex mixed with useful data and can not be as simply separated.

That is not natural noise.

That anime was 3D rendered and digital noise added later, to simulate film effect. The bandwidth, as far as my eyes can see, is really limited and mostly visible in darker area, even if light ones are mildly affected too.

[DTL]

Typical examples for narrow band distortions are Aerial interference with analog broadcast or some analog interference at ADC board/device to convert analog to digital. If random special noise is added - it most probable have too wide (flat) spectrum and can not be easily filtered out with narrow band-stop filters.

[tormento]

Originally Posted by kedautinh12

Choose what you like. For personal, i prefer DPIR +UniScale CartoonRestore-lite + AnimeJanai's result

Would you please address me to where to download them?

Do you know of some model that can keep 1x scale and deblur/antialiasing, to have some anime restoration instead of scaling?

[tormento]

Tried to use Oyster_super.

I have to Crop(128,128,-128,-128) and resize because the image become enlarged and corrupted along the sides and doubles.

Plus, the comment says ## Input: 444 YUV/RGB at any bitdepth (output is float) but it needs to be 32 bits.

Anyone else?

[takla]

LWLibavVideoSource("INPUT.mkv")
ConvertBits(32).ConvertToPlanarRGB()
AItrashHere()

[rgr]

Has anyone compared Stabilization Tools Pack with Deshaker 3.1?

[tormento]

The call from xClean:

ConvertFormat(cconv, fmt_out="444", kernel_c=chroma, tv_in=!fulls, cplace_in=cplace)

causes

Script error: NNEDI3RESIZE does not have a named argument "src_width"
(D:/Programmi/Media/AviSynth+/plugins64/TransformsPackMain-2.2.1~Dogway.avsi, line 969)
(D:/Programmi/Media/AviSynth+/plugins64/TransformsPackMain-2.2.1~Dogway.avsi, line 971)
(D:/Programmi/Media/AviSynth+/plugins64/TransformsPackMain-2.2.1~Dogway.avsi, line 972)
(D:/Programmi/Media/AviSynth+/plugins64/TransformsPackMain-2.2.1~Dogway.avsi, line 973)
(D:/Programmi/Media/AviSynth+/plugins64/TransformsPackMain-2.2.1~Dogway.avsi, line 973)
(D:/Programmi/Media/AviSynth+/plugins64/TransformsPackMain-2.2.1~Dogway.avsi, line 1323)
(D:/Programmi/Media/AviSynth+/plugins64/xClean-beta10~mysteryx93.avsi, line 232)

and I can't sort out the bug in the TransformsPackMain-2.2.1.

Any idea?

[tormento]

It happens that anime are getting more and more "clean" in the general appearance but if you zoom in the dark areas you can see lot of "pixel fluctuations", perhaps coming from a poor AVC compression strategy, even from BDs. Sometimes it appears on light areas too but it's more rare.

Wanting to cleanup those lurking pixels without ruining the overall quality what do you suggest me?

I have tried with high tr and thsad numbers but all I obtain is to ruin the details without cleaning up the dark areas.

Someone tried it before?

[poisondeathray]

Originally Posted by tormento

It happens that anime are getting more and more "clean" in the general appearance but if you zoom in the dark areas you can see lot of "pixel fluctuations", perhaps coming from a poor AVC compression strategy, even from BDs. Sometimes it appears on light areas too but it's more rare.

Wanting to cleanup those lurking pixels without ruining the overall quality what do you suggest me?

I have tried with high tr and thsad numbers but all I obtain is to ruin the details without cleaning up the dark areas.

Someone tried it before?

Filter the dark areas with stronger filtering than the brighter areas (or no filtering for the bright)

You can use the LumaMask in the MasksPack

[tormento]

Originally Posted by poisondeathray

Filter the dark areas with stronger filtering than the brighter areas (or no filtering for the bright)

You can use the LumaMask in the MasksPack

Thank you, nice idea.

And what about “subtle pixel noise” in light areas?

It’s way less noticeable but still there.

The problem is that from film grain noise, digital workflow switches that easily fixable rumor to tiny pixel fluctuations than you can really notice on larger screens.

[poisondeathray]

Originally Posted by tormento

Thank you, nice idea.

Not really my idea - people have been doing this for >30 years, pre-avisynth with denoising images. You can find video mask scripts 20 years ago

And what about “subtle pixel noise” in light areas?

Use a "subtle pixel noise filter" for the light areas using the mask

The problem is that from film grain noise, digital workflow switches that easily fixable rumor to tiny pixel fluctuations than you can really notice on larger screens.

Then it wasn't produced correctly. Digital workflows add grain to blend everything in . e.g. CG added to film

[Z2697]

Originally Posted by poisondeathray

Then it wasn't produced correctly. Digital workflows add grain to blend everything in . e.g. CG added to film

Ray-tracing renderers can produce "real" grians by itself.
And most noises in digital era is added without good reason.

[poisondeathray]

Originally Posted by Z2697

Ray-tracing renderers can produce "real" grians by itself.

Yes, or not - it depends on the settings

Often the "grain" does not end up look like real film grain, but digital noise - not enough iterations and convergence during render

And most noises in digital era is added without good reason.

Yes, digital acquisition tends to be clean, especially with good lighting. The grain is added in post. "Dark" scenes are often shot brighter, then adjusted to look "dark" in post

But for film, the CG elements have grain emulation to blend into the film as stated above

[tormento]

Originally Posted by poisondeathray

Then it wasn't produced correctly. Digital workflows add grain to blend everything in . e.g. CG added to film

Some full CGI animes start to have good enough quality to be almost on par with drawn ones, i.e. Girls band cry.

The problem is that people are so used to grain that they artificially added it with no apparent reason.

Thus, denoising that digital "pixel fluctuation" makes faint details from CGI to disappear.

I uploaded a small clip from that anime, if you are curious to look at what I mean.

[Dogway]

@tormento, this is more or less the high freq grain, but I'm not sure you can do something with it.

a=last
STTWM(sw=0,tw=100,aw=100,sthres=0,tthres=2)
ex_makediff(a, metric="none", aug=128, dif=true, show=3)

This a simple filter (or use ex_FluxSmoothT() instead of STTWM):

pre=STTWM(sw=0,tw=100,aw=100,sthres=2,tthres=2).ex_sbr()
SMDegrain(2, 200, mode="RMDegrain", prefilter=pre,RefineMotion=true)

It's going to be hard to protect everything, since there are many low contrast lines, but it does the job.


You can also add some masks to protect details...

dt=ex_edge("kayyali",lo=40)
dtm=dt.ex_expand(2,mode="disk")
em=ex_edge("kroon")
gib=em.ex_expand(2).ex_inflate()


ConvertBits(16)
pre=ex_FluxSmoothST().ex_minblur(1)
msk=ex_logic(dtm,em,"max").ConvertBits(16)
pre=ex_merge(pre,last,msk)
SMDegrain(2, 200, mode="RMDegrain", prefilter=pre,RefineMotion=true)
ConvertBits(8,dither=1)

[Dogway]

Originally Posted by LouieChuckyMerry

I'm trying to put Deblock_QED to use in the DeblockPack.avsi and seem to actually have it functioning correctly, but wanted to make sure before I spend time testing it. My script is

Deblock_QED()
ConvertBits(Bits=16)
TemporalDegrain2()
CAS()
Neo_F3KDB(Y=100,Cb=100,Cr=100,GrainY=0,GrainC=0)

I found that putting the Deblock_QED call after converting to 16bits made for LSD flashback-style output; does this mean that it only accepts 8bit input?

Also, while gathering the needed dependencies for Deblock_QED in the DeblockPack.avsi, I encountered the plugin BlockDetect and wondered what it does. Well, I can read what it does, but don't understand why it would be used. Thanks.

I tested and Deblock_QED works fine in HBD.
The Deblock dependency has nothing to do with BlockDetect, its Deblock, here's the avisynth entry page. It's basically a deblock filter, Deblock_QED limits the deblocking to the block edges.

Maybe the issue is on the other filters?

[tormento]

Originally Posted by Dogway

@tormento, this is more or less the high freq grain, but I'm not sure you can do something with it.

Thank you and welcome back!

Please notice a fix for your scripts here.

P.S: MysteryX told me that he is not using Windows anymore and is not interested in further development. What about merging its source with ExTools or so?

P.P.S: Would be possible to have motion compensated BM3D_CUDA and KnlMeansCL?

[anton_foy]

Originally Posted by tormento

Thank you and welcome back!

Please notice a fix for your scripts here.

P.S: MysteryX told me that he is not using Windows anymore and is not interested in further development. What about merging its source with ExTools or so?

P.P.S: Would be possible to have motion compensated BM3D_CUDA and KnlMeansCL?

I am maybe wrong but does not knlmeanscl use some kind of motion compensation already? Try it with 'd=12' and see how it acts. BM3D is already using block matching.