QTGMC Deinterlacing Script (v3.384)

[Boulder]

Originally Posted by mark0077

So do you just mean via not using SetMTMode at all?

Yes, that's what I mean. When I was an Avisynth user, I always split the QTGMC encode into multiple parts with Trim and encoded with multiple simultaneous ffmpeg calls. Crash-proof at least, and faster too.

[mark0077]

Great idea, thanks very much I'll try that!

[real.finder]

you can try MP_Pipeline, it's stable too, and ofc fast and mt base on every block

[trevorjharris]

There use to be a multithreaded plugin pack tuned for multithreading. Please can you tell me if an uptodate version is available or are the current plugins thread safe.

[real.finder]

Originally Posted by trevorjharris

There use to be a multithreaded plugin pack tuned for multithreading. Please can you tell me if an uptodate version is available or are the current plugins thread safe.

the updates one is safe too

[Trademark]

I've gotten a new capture card for my playstation 2 footage. I used to demux .ts files into separate audio and video streams, then load them into avisynth that way, then ran the qtgmc script as seen below without problem.

Now I'm using uncompressed .avi files, and am trying to use the same script:

AviSource("C:\Users\Owen\Desktop\2.avi")
Crop(0,6,0,0)
ConvertToYV12(interlaced=true)
ConvertAudioTo16Bit()
A2 = last
A2 = A2.QTGMC(Preset="slower")
TDeint(mode=1, order=1, field=1, edeint=A2, emask=TMM(mode=1, order=1, field=1),slow=2,chroma=false,MI=80,full=false,cthresh=9)


But the tdeint using qtgmc no longer works, and the footage remains interlaced. Using QTGMC slower by itself works just fine, but for some reason, I'm having a conflict between the new avi files, and this tdeint qtgmc script. Does anybody have any ideas? I can try to upload a short sample clip of the footage I'm working with if it helps.

[LemMotlow]

Never mind.

[StainlessS]

Originally Posted by Trademark

AviSource("C:\Users\Owen\Desktop\2.avi")
Crop(0,6,0,0)
ConvertToYV12(interlaced=true)
ConvertAudioTo16Bit()
A2 = last
A2 = A2.QTGMC(Preset="slower")
TDeint(mode=1, order=1, field=1, edeint=A2, emask=TMM(mode=1, order=1, field=1),slow=2,chroma=false,MI=80,full=false,cthresh=9)

Perhaps LemMotlow already posted on this, then changed his mind, so maybe I'm wrong too ???, but

We dont really like the crop, on interlaced should only be cropped vertical in multiples of 4 otherwise you are trashing the interlacing (whether colorspace is YV12 or anything else).

EDIT:
Oops, ignore above Too !!!, Interlaced non YV12 can of course be cropped in multiples of 2 guess thats why LM changed his mind,
guess we both got out of bed way too early


EDIT: However, result after cropping should be a multiple of 4 if later converting to YV12 with Interlaced=true, otherwise who knows what problems would occur.

[Music Fan]

Yes, and in my opinion, if your avi is in yuy2, keep it as is for de-interlacing and convert to YV12 after.
And having 2 A2 in your script is maybe problematic, I would try this ;

AviSource("C:\Users\Owen\Desktop\2.avi")
ConvertAudioTo16Bit()
Crop(0,6,0,-2)
A1 = last
A2 = A1.QTGMC(Preset="slower")
TDeint(mode=1, order=1, field=1, edeint=A2, emask=TMM(mode=1, order=1, field=1),slow=2,chroma=false,MI=80,full=false,cthresh=9)
ConvertToYV12()#not interlaced anymore after TDeint

But I'm afraid this way your video is first de-interlaced with the line A2 = A1.QTGMC(Preset="slower"), then TDeint actually gets 50p (or 60p), as if you had ;

AviSource("C:\Users\Owen\Desktop\2.avi")
ConvertAudioTo16Bit()
Crop(0,6,0,-2)
QTGMC(Preset="slower")
TDeint(mode=1, order=1, field=1, edeint=A2, emask=TMM(mode=1, order=1, field=1),slow=2,chroma=false,MI=80,full=false,cthresh=9)
ConvertToYV12()

This could work ;

AviSource("C:\Users\Owen\Desktop\2.avi")
ConvertAudioTo16Bit()
Crop(0,6,0,-2)
A1=QTGMC(Preset="slower")
TDeint(mode=1, order=1, field=1, edeint=A1, emask=TMM(mode=1, order=1, field=1),slow=2,chroma=false,MI=80,full=false,cthresh=9)
ConvertToYV12()

[StainlessS]

Originally Posted by Music Fan

AviSource("C:\Users\Owen\Desktop\2.avi")
ConvertAudioTo16Bit()
Crop(0,6,0,-2) # <<<-------
A1 = last
A2 = A1.QTGMC(Preset="slower")
TDeint(mode=1, order=1, field=1, edeint=A2, emask=TMM(mode=1, order=1, field=1),slow=2,chroma=false,MI=80,full=false,cthresh=9)
ConvertToYV12(interlaced=true)

Small typo fix

[Music Fan]

Oops, thanks, edited !

[Trademark]

I had to put ConvertToYV12(interlaced=true) before the deinterlacing script because I was getting the error message: 'TemporalSoften: Scenechange not available on RGB32'

I also had to Crop only 6 pixels off the height of the video because the source footage is 486 pixels high, with black bars of 18 pixels on top and bottom. Cropping 6 of the black off the top made it so it's 480 pixels high, and a multiple of 4. If I crop more or less I get this: 'SeparateFields: YV12 height must be multiple of 4'


I'm afraid this doesn't work either.
A1=QTGMC(Preset="slower")
TDeint(mode=1, order=1, field=1, edeint=A1, emask=TMM(mode=1, order=1, field=1),slow=2,chroma=false,MI=80,full=false,cthresh=9)
ConvertToYV12()



Just to give some more context, this is where the script originally came from: https://forum.doom9.org/showthread.p...74#post1636574

Back then, we used it for the .ts files, and I wanted the deinterlacer to keep the same framerate. Nowdays, I'm doubling the framerate to 60progressive, hence I have the 1s for mode.


Here is the sample footage on google drive (500ish mb)

https://drive.google.com/file/d/0B1b...ew?usp=sharing


Thanks for helping so far guys!

[LemMotlow]

Had to change my mind about my original post, because the colorspace and other specs of the original source weren't clear. I figured more info would be needed.
The latest QTGMC with updated plugins can use YUY2. But I seldom use it, as most other plugins still want YV12, but often YUY2 does look cleaner.

[Trademark]

AviSource("C:\Users\Owen\Desktop\2.avi")
Crop(0,6,0,0)
ConvertToYV12(interlaced=true)
ConvertAudioTo16Bit()
AssumeTFF()
A2 = last
A2 = A2.QTGMC(Preset="slower")
TDeint(mode=1, order=1, field=1, edeint=A2, emask=TMM(mode=1, order=1, field=1),slow=2,chroma=false,MI=80,full=false,cthresh=9)


So I just tried adding AssumeTFF() to my deinterlacing script, and I've tried keeping the modes on tdeint and tmm to 1 and -1, and both yield the same results. It seems as though the tdeint with gtgmc edeint is now working, but the source-filter might not be frame accurate? I get a resulting video that spasms around back and forth.

Edit: I can add a .selecteven() after qtgmc, and change modes of tdeint and tmm to 0, and I get a smooth 30fps, but for some reason it's not agreeing with the bobbing for 60fps.

[LemMotlow]

Your sample is UYVY (4:2:2), bottom field first, and has 19 pixels top and bottom, not 18. You can get rid of 6 surplus lines to make it 480 height and keep the borders looking fairly even vertically using this in UYVY:

Crop(0,2,0,-4)

Not sure what you're trying to do. What's TDeint for? Deinterlace twice and you've got 199.88fps and duplicate frames. The TDeint line doesn't make much sense (for example if you have "mode=1," then "field=1" does nothing). I think you must be trying to clean up the ghosting and think it's "combing", but it won't work.

Maybe tell us a little history of how this video was created, and possibly how it was borked.

For the linked sample, which is just straight deinterlace to 59.94fps, I used this:

Crop(0,2,0,-4)
ConvertToYV12(interlaced=true)
QTGMC(preset="very fast",border=true)

Trademark_2sample.mp4 (~13 mb): https://www.mediafire.com/?v6gz5y2cwmouonn

[Trademark]

Here's a little discussion that lead to the creation of that script a while back: https://forum.doom9.org/showthread.p...74#post1636574

The link directs to poisondeathray's post, for which I got the script from. Essentially, I found that QTGMC alone smoothed the text on the video too much, while his script keeps most of benefits of QTGMC, plus keeping the text sharp.

I've been using that script without the SelectEven() and putting mode=1on tdeint and tmm to get smooth and buttery 60fps progressive-scan videos, but that was for raw h264 streams demuxed from .ts files. Now I'm recording videos in uncompressed 8-bit yuv with .avi containers.


Thanks for clarifying that the video is bottom field first, and not top field first. I've changed my script accordingly:

AviSource("C:\Users\Owen\Desktop\2.avi", audio=false)
AssumeBFF()
Crop(0,2,0,-4)
ConvertToYV12(interlaced=true)
A2 = last
A2 = A2.QTGMC(Preset="slower")
TDeint(mode=1, order=0, field=0, edeint=A2, emask=TMM(mode=1, order=0, field=0),slow=2,chroma=false,MI=80,full=false,cthresh=9)

This seems to work now! However, if anybody has any suggestions for deinterlacing scripts that can yield even better visual results, I'm always interested.

[mbcd]

Tried version 3.57 some weeks ago, but recognizes many issues with it. So went back to 3.32 for a while.

Today I got still many issues with v3.57, mostly temporal problems.
I am using exact the same parameters as with 3.32, v3.32 still works without any problems since a very long time.



So it cant be my chain, its definitively version 3.57.
Downloaded the latest plugins-pack for 3.57 including the 2 updated plugins.

This are my settings, they work fine for 3.32 since many many months, source is still progressive.

QTGMC(Preset="Medium", EdiThreads=6, InputType=1, TR2=2, NoiseProcess=1, NoiseRestore=0.0, StabilizeNoise=true, DenoiseMC=true, Denoiser="dfttest", Sigma=0.5, Sharpness=2.0, SLMode=2, ForceTR=3, SourceMatch=3, MatchEnhance=0.75, FPSDivisor=1)

The whole movie look more like a bad norm-converted ... the problems are around everywhere, no special areas or scenes.

And NO, its not a ghostmovie ;-)

[LigH]

Originally Posted by mbcd

And NO, its not a ghostmovie ;-)

But can we be sure that it's not already norm-converted with blending in the "original" material?

We would need a scene from the original material for inspection.

I would recommend not to blame software before we are sure that nothing else is to be blamed.

[mbcd]

I dont have the source anymore, but I can take ANY source I want, because it is always persistant after I switch from 3.32 to 3.57, doesnt matter what I use as source.

I am for 101% sure that sources are progressive, also why are these problems gone if I switch the same source from 3.57 to 3.32 without changing parameters?
Is 3.32 with InputType=1 not also "progressive" ? Why this difference in output?

I think that my comment 100% prooves that there are some flaws in 3.57 while ALL is fine with 3.32.
Still wondering why I am the only one with that massive problems, so I think it might be my parameters, thats why I posted them too.
As none-native american sometimes its hard to understand what some parameters do.

[poisondeathray]

I can reproduce mbcd's "ghosting" on any random progressive source with 3.357 using those settings. Drop in 3.32 and ghosts vanish.