QTGMC and Srestore Bouncing

[lomaidala]

I've recently noticed that using QTGMC when the Srestore plugin has also been loaded causes certain "stationary" areas of certain shots to "bounce" instead of staying stationary. This seems to only happen when the shot itself is predominately stationary (no panning or zooming, and relatively little change in the frame itself), and even then, only sometimes. Otherwise, the majority of the video looks perfect (or at least, as perfect as it can look). Changing the QTGMC presets has no effect.

Does anyone know how to keep the stationary parts of these shots from bouncing?

My current script:

LoadPlugin("C:\DGDecode.dll")
LoadPlugin("C:\Program Files (x86)\AviSynth\plugins\masktools2.dll")
Import("C:\Program Files (x86)\AviSynth\plugins\Srestore_27h.avsi")
clip1 = MPEG2Source("C:\DVD\1\VTS_01_1.d2v")
clip2 = MPEG2Source("C:\DVD\2\VTS_01_1.d2v")
clip2.Trim(21,2716).FadeOut0(19) + clip1.Trim(2696,0)
QTGMC( Preset="Placebo" )
Srestore(Frate=23.976)

Video source:
It's encoded interlaced and has interlaced frames, and there's field blending.

Software:
VirtualDub-1.10.4
AviSynth 2.6
Srestore 2.7h
Masktools2 2.2.14
MVTools2 2.7.25 with-depans
NNEDI3 0.9.4.48
QTGMC 3.357
RgTools 0.96
SMDegrain 3.1.2.93s
FFT3DFilter 2.4
GRunT 1.0.1

Specs:
Windows 7 Professional, 64-bit
AMD Phenom(tm) II X6 1090T Processor 3.21 GHz
AMD 6870
AMD 6850
8.00 GB RAM

[SaurusX]

use some of the deshimmering settings in QTGMC. Like tr2=1 or tr=2 or 3 if it's really bad.

[jagabo]

Maybe the wrong field order. Try AssumeTFF() or AssumeBFF() before QTGMC(). Or maybe SwapFields(). Or post a sample of the source.

[hello_hello]

Probably nothing to do with the problem, but wouldn't it be better to de-interlace and deblend the clips before combining them?

clip1 = MPEG2Source("C:\DVD\1\VTS_01_1.d2v").QTGMC( Preset="Placebo" ).Srestore(Frate=23.976)
clip2 = MPEG2Source("C:\DVD\2\VTS_01_1.d2v").QTGMC( Preset="Placebo" ).Srestore(Frate=23.976)
clip2.Trim(21,2716).FadeOut0(19) + clip1.Trim(2696,0)

Of course the frame numbers for trimming would change, but appending scripts can't mess with SRestore's deblending that way.

[lomaidala]

Maybe the wrong field order. Try AssumeTFF() or AssumeBFF() before QTGMC(). Or maybe SwapFields(). Or post a sample of the source.

Thanks for those suggestions, but none of them made the issue better, and even more of the shot bounces with SwapFields(). Attached is a sample of the source.

Probably nothing to do with the problem, but wouldn't it be better to de-interlace and deblend the clips before combining them?

clip1 = MPEG2Source("C:\DVD\1\VTS_01_1.d2v").QTGMC( Preset="Placebo" ).Srestore(Frate=23.976)
clip2 = MPEG2Source("C:\DVD\2\VTS_01_1.d2v").QTGMC( Preset="Placebo" ).Srestore(Frate=23.976)
clip2.Trim(21,2716).FadeOut0(19) + clip1.Trim(2696,0)

Of course the frame numbers for trimming would change, but appending scripts can't mess with SRestore's deblending that way.

Thanks! In what way might I notice Srestore's deblending being messed with? I only only combined the clips after QTGMC and Srestore because skipping around in the video to find the right spots for the cuts is a lot easier that way in VirtualDub. But I would much rather do things the proper way than the easy way.

use some of the deshimmering settings in QTGMC. Like tr2=1 or tr=2 or 3 if it's really bad.

Thanks for the suggestion, but TR2=1 and TR2=2 have no effect in my case.

[jagabo]

All those thin horizontal lines are confusing QTGMC. And there's essentially no motion in the clip making it hard to really analyze. Can you provide a clip with some real motion? Or, if this is the only part you're having problems with there are other ways to deal with it.

[lomaidala]

All those thin horizontal lines are confusing QTGMC. And there's essentially no motion in the clip making it hard to really analyze. Can you provide a clip with some real motion?

Yes, I can provide a clip with more motion, what kind of motion would be most helpful? A steady pan? If that's the case, it almost certainly won't be one of the shots that is giving me problems, which are all mostly still. Also, those thin lines are mimicking a monitor, so they're unique to that shot. But the problem I'm having does extend beyond that to shots without the lines.

Or, if this is the only part you're having problems with there are other ways to deal with it.

Unfortunately, this isn't the only part I'm having problems with. There are multiple DVDs I'm working with, each one with a handful of episodes. I haven't gone through and done a thorough check, but a cursory search indicates that there seems to be an average of one problem shot per episode. I suspect there are more though.

[jagabo]

Something with mixed motion would be best, moderate speed panning, character motion, etc.

That particular shot will probably have to be special cased. Something like this works because the only thing that needs deinterlacing is the bright white text:

# build a mask of bright white areas, sum frame before, frame, and frame after
mask = last.Trim(1,0).mt_binarize(180)
mask = Overlay(mask, last.mt_binarize(180), mode="lighten")
mask = Overlay(mask, last.loop(2,0,0).mt_binarize(180), mode="lighten")
mask = mask.mt_expand().Blur(1.0).Blur(1.0).Blur(1.0) # expand and soften edges

Overlay(last, Blur(0.0, 1.0), mask=mask) # vertical blur (cheap deinterlace) only the masked areas

In the end, no deinterlacer is perfect, nor can there ever be one.

[lomaidala]

Something with mixed motion would be best, moderate speed panning, character motion, etc.

Okay thanks, see attachment.

That particular shot will probably have to be special cased. Something like this works because the only thing that needs deinterlacing is the bright white text:

# build a mask of bright white areas, sum frame before, frame, and frame after
mask = last.Trim(1,0).mt_binarize(180)
mask = Overlay(mask, last.mt_binarize(180), mode="lighten")
mask = Overlay(mask, last.loop(2,0,0).mt_binarize(180), mode="lighten")
mask = mask.mt_expand().Blur(1.0).Blur(1.0).Blur(1.0) # expand and soften edges

Overlay(last, Blur(0.0, 1.0), mask=mask) # vertical blur (cheap deinterlace) only the masked areas

Thank you. I tried integrating your code into my script, but I came up with this error:
VirtualDub Error
Avisynth open failure:
Script error: Invalid arguments to function "Blur"
(line 15)

Current script:

LoadPlugin("C:\DVD\DGDecode.dll")
LoadPlugin("C:\Program Files (x86)\AviSynth\plugins\masktools2.dll")
Import("C:\Program Files (x86)\AviSynth\plugins\Srestore_27h.avsi")
clip1 = MPEG2Source("C:\DVD\1\VTS_01_1.d2v").QTGMC( Preset="Placebo" ).Srestore(Frate=23.976)
clip2 = MPEG2Source("C:\DVD\2\VTS_01_1.d2v").QTGMC( Preset="Placebo" ).Srestore(Frate=23.976)
mask = ImageSource("C:\Images\1.png")
mask = mask.ConvertToRGB32
mask = clip1.Trim(6665,7300).mt_binarize(180)
mask = Overlay(mask, clip1.mt_binarize(180), mode="lighten")
mask = Overlay(mask, clip1.loop(2,0,0).mt_binarize(180), mode="lighten")
mask = mask.mt_expand().Blur(1.0).Blur(1.0).Blur(1.0)
clip2.Trim(11,2178).FadeOut0(15) + \
clip1.Trim(2169,6664) + \
clip1.Trim(6665,7300).Overlay(clip1, Blur(0.0, 1.0), mask=mask) + \
clip1.Trim(7301,0)

I'm also thinking another approach might be better in this case because there's actually more of the shot you haven't seen (not included in my previous sample), which includes animation that covers the entire frame. Because of this, and because the unintentional interlacing won't look too out of place in this case because the original shot is meant to mimic a monitor with scan lines, do you think a better approach would be to just lose QTGMC altogether for just this one shot? Perhaps using RemapFrames?

In the end, no deinterlacer is perfect, nor can there ever be one.

Good to know. I'm definitely happier using a deinterlacer with its imperfections than I am leaving the video interlaced.

[jagabo]

The problem with Blur() in your script is that you aren't specifying what video to blur. AviSynth uses the name "last" whenever you don't specify a clip for input and output. So a script like this:

Mpeg2Source("filename.d2v")
Blur(1.0)

really means:

last = Mpeg2Source("filename.d2v")
last = Blur(last, 1.0)
return(last)

In your script, the line:

clip1.Trim(6665,7300).Overlay(clip1, Blur(0.0, 1.0), mask=mask)) + clip1.Trim(7301,0)

Blur() is trying work on last but you have explicitly named all the streams prior to that so there is no clip named last. There also a problem with the use of cllip1.Trim() and Overlay(clip1...). clip1 in the Overlay function is not the clipped version, but the original version of clip1. I recommend you split that into something like (I think this is what you're trying to do):

special = clip1.Trim(6665,7300)
special = Overlay(special, Blur(special, 0.0, 1.0), mask=mask)
special + clip1.Trim(7301,0)

There are other problems with your scirpt. You create a clip called mask from ImageSource() and convert it to RGB. Then you created a clip called mask (discarding the previous one) from clip1.Trim(...)... I'm not sure what you're trying to do there.

The new clip you provided is in really poor shape. If you look at individual fields you'll see that many are torn -- the top part of the field comes from one field, the bottom part from another (usually a blend of two fields). Normally one would use TFM().TDecimate() on a film source like this but this problem requires something different. Fortunately, it looks like there are enough clean fields that you can do this:

Mpeg2Source("VTS_01_1.demuxed(1).d2v", CPU2="xxxxxx", Info=3) 
QTGMC(preset="fast")
SRestore(frate=23.976)

SRestore() will preferentially remove frames where it sees blending, leaving only clear frames in this case.

There's also a lot of ghosting from an overly strong temporal noise reduction filter. But there's not a lot you can do about that.

[vipren369]

hey So i have an issue with the Srestore.avsi script... as i'm using it with the d2v it keeps giving me an error "No Function named GScriptClip" thinking may not be the correct sRestore Script??

[jagabo]

You need GRunt: http://avisynth.nl/index.php/GRunT

SRestore requires other filters that you may not have: http://avisynth.nl/index.php/Srestore#Required_Plugins

[lomaidala]

Thank you jagabo for pointing out my mistake with Blur(). And thank you for pointing out CPU2="xxxxxx" and Info=3! I had no idea my source video could benefit from those.

My latest script now loads without an error, but something isn't right because the video looks the same as before.

Current script:

LoadPlugin("C:\DVD\DGDecode.dll")
LoadPlugin("C:\Program Files (x86)\AviSynth\plugins\masktools2.dll")
Import("C:\Program Files (x86)\AviSynth\plugins\Srestore_27h.avsi")
clip1 = MPEG2Source("C:\DVD\1\VTS_01_1.d2v", CPU2="xxxxxx", Info=3).QTGMC( Preset="Placebo" ).Srestore(Frate=23.976)
clip2 = MPEG2Source("C:\DVD\2\VTS_01_1.d2v", CPU2="xxxxxx", Info=3).QTGMC( Preset="Placebo" ).Srestore(Frate=23.976)
mask = ImageSource("C:\Images\1.png")
special = clip1.Trim(6665,7300)
special = Overlay(special, Blur(special, 0.0, 1.0), mask=mask)
mask = special.Trim(6665,7300).mt_binarize(180)
mask = Overlay(mask, clip1.mt_binarize(180), mode="lighten")
mask = Overlay(mask, clip1.loop(2,0,0).mt_binarize(180), mode="lighten")
mask = mask.mt_expand().Blur(1.0).Blur(1.0).Blur(1.0)
clip2.Trim(11,2178).FadeOut0(15) + \
clip1.Trim(2169,6664) + \
special + clip1.Trim(7301,0)

There are other problems with your scirpt. You create a clip called mask from ImageSource() and convert it to RGB. Then you created a clip called mask (discarding the previous one) from clip1.Trim(...)... I'm not sure what you're trying to do there.

The RGB conversion is because I thought RGB conversion was necessary for mask images to work properly. As for overriding last with last, I guess I'm still not clear on when a new line adds to last, and when it completely overrides it with a new last.

[jagabo]

thank you for pointing out CPU2="xxxxxx" and Info=3!

Note that Mpeg2Source's deblocking is a too strong and loses detail in some areas. I usually use deblock_qed() instead. But you'll have to tune the deblocking strength settings to your video.

Some general notes:

LoadPlugin("C:\DVD\DGDecode.dll")
LoadPlugin("C:\Program Files (x86)\AviSynth\plugins\masktools2.dll")
Import("C:\Program Files (x86)\AviSynth\plugins\Srestore_27h.avsi")

dll and avsi files in AviSynth's plugins folder should load automatically. So the latter two lines shouldn't be necessary.

mask = ImageSource("C:\Images\1.png")
special = clip1.Trim(6665,7300)
special = Overlay(special, Blur(special, 0.0, 1.0), mask=mask)

What's 1.png? What is it's purpose? You're using it to blur parts of the frame but isn't that what the mask below is for?

mask = special.Trim(6665,7300).mt_binarize(180)
mask = Overlay(mask, clip1.mt_binarize(180), mode="lighten")
mask = Overlay(mask, clip1.loop(2,0,0).mt_binarize(180), mode="lighten")
mask = mask.mt_expand().Blur(1.0).Blur(1.0).Blur(1.0)

Now you are building a new mask -- and not using it for anything. The first line is attempting to build mask from frames 6665 to 7300 of the clip called special But special is only frames 6665 to 7300 of clip1. Ie, it is only 636 frames long. I'm not going to bother with the rest of the errors here.

What I think you're trying to do is apply QTGMC().SRestore() to the bulk of your videos but apply the special deinterlacing (blurring only bright white parts of the frame) to frames 6675 though 7500 of clip1, the portion where QTGMC causes parts of the picture to bounce up and down. Aside from the other errors in the script, you need to perform that special deinterlacing on the video before QTGMC has been applied.

LoadPlugin("C:\DVD\DGDecode.dll")
LoadPlugin("C:\Program Files (x86)\AviSynth\plugins\masktools2.dll")
Import("C:\Program Files (x86)\AviSynth\plugins\Srestore_27h.avsi")

rawclip1 = MPEG2Source("C:\DVD\1\VTS_01_1.d2v", CPU2="xxxxxx", Info=3)
clip1 = rawclip1.QTGMC( Preset="Placebo" ).Srestore(Frate=23.976)
clip2 = MPEG2Source("C:\DVD\2\VTS_01_1.d2v", CPU2="xxxxxx", Info=3).QTGMC( Preset="Placebo" ).Srestore(Frate=23.976)

special = rawclip1.Trim(6665,7300) # get the special part of rawclip1
mask = special.Loop(2,0,0).mt_binarize(180) # build a mask from the previous frame of special
mask = Overlay(mask, special.mt_binarize(180), mode="lighten") # add to the mask from the current frame of special
mask = Overlay(mask, special.Trim(1,0).mt_binarize(180), mode="lighten") # add to the mask from the next frame of special
mask = mask.mt_expand().Blur(1.0).Blur(1.0).Blur(1.0) # expand and blur the mask
special = Overlay(special, Blur(special, 0.0, 1.0), mask=mask) # deinterlace (blur) only the bright areas of special
special = special.TDecimate() #convert to 23.976 fps like the rest of the clips

# now append all the wanted segments
clip2.Trim(11,2178).FadeOut0(15) + \
	clip1.Trim(2169,6664) + \
	special + \
	clip1.Trim(7301,0)

I don't have the source clips so I haven't tested this script. So there might be some errors.