How to deal with this frame blending situation?

[xuguang_he]

{ Original }


{ After Using QTGMC }


This blending frame is very stubborn and even QTGMC couldn't decimate it completely.
So, may I ask what will be the best solution to deal with this kind of fast-moving scene?

BTW, comparing to TIVTC, QTGMC is more efficient. But, as far as I was concerned, it do some frame interpolation which TIVTC doesn't do. And for some sources which are not "really interlaced", I've been told to follow such rules:

(1) For PAL interlace: yadifmod(edeint=nnedi2())
(2) With NTSC DVDs, if they are 100% film: tfm(pp=0).tdecimate()
(3) If they are not 100% film (for example, 99.99% film or less), choose from these two for the more efficient one:
tfm().tdecimate() or tfm(order=1).tdecimate()

Now, since I knew QTGMC was efficient but slow, I want to use it all the time without caring about the percentage number given by the d2v file. Is this a good solution? Could I use the following code all the time to deal with interlacing and blending (Obviously, in this example, QTGMC is not enough)?

WILL QTGMC DO HARM TO THE SOURCE? (That is to say, apart from decimating interlacing and blending, will it do something harmful to the source other than restoring?)

AssumeTFF()
Import("D:\Program Files\Megui\tools\avisynth_plugin\QTGMC-3.32.avs")
QTGMC()
Merge(SelectEven(),SelectOdd())

[xuguang_he]

This is another movie, needless to say it's another totally different source. But the situation is resembled.
Take a look at Gregory Peck's action.

{ Before }


{ After }

[xuguang_he]

Another comparsion. (Before & After using QTGMC)

{ Before }


{ After }

[xuguang_he]

I felt that, sometimes, QTGMC makes the source's color more thick and the picture looks a bit like a oil painting, which is not to my like.

{ Before }


{ After }


Or maybe it's due to the fuzzy original source.

Also, take a look at the walking person in background. The original source was blended a bit, but after QTGMC' filtering function, it has been decimate perfectly.

[xuguang_he]

{ Before }


{ After }

[xuguang_he]

EEDI2().Spline36Resize(width,height).turnleft().EEDI2().Spline36Resize(height,width).turnright()

Is this the best solution for anti-aliasing?

[poisondeathray]

This blending frame is very stubborn and even QTGMC couldn't decimate it completely.
So, may I ask what will be the best solution to deal with this kind of fast-moving scene?

Sometimes you need to apply different settings to different sections, different filter, or different approach entirely . QTGMC is a temporal filter and known to cause blends

BTW, comparing to TIVTC, QTGMC is more efficient.

There is probably misinterpretation in "efficient"
1) These are not comparable functions, they have different purposes (field matching vs. deinterlacing)
2) Efficiency deals with measuring inputs and outputs e.g. a cost to get a certain effect

But, as far as I was concerned, it do some frame interpolation which TIVTC doesn't do. And for some sources which are not "really interlaced", I've been told to follow such rules:

Again, these aren't comparable functions. Apples to oranges.

But you can tell TIVTC to use QTGMC during comb detection with the clip2 parameter. Have a look the instructions

Now, since I knew QTGMC was efficient but slow, I want to use it all the time without caring about the percentage number given by the d2v file. Is this a good solution? Could I use the following code all the time to deal with interlacing and blending (Obviously, in this example, QTGMC is not enough)?

Not a good idea

WILL QTGMC DO HARM TO THE SOURCE? (That is to say, apart from decimating interlacing and blending, will it do something harmful to the source other than restoring?)

Yes it alters the original source, doesn't retain original fields. It applies other processing like denoising and sharpening. At default strengths, often too high for most sources (think "oil painting" look). All the settings are adjustable - read the instructions. Even for deblending (srestore) , it can cause more problems than other deinterlacers (but other deinterlacers may leave more deinterlacing artifacts behind)

[poisondeathray]

EEDI2().Spline36Resize(width,height).turnleft().EEDI2().Spline36Resize(height,width).turnright()

Is this the best solution for anti-aliasing?

Maybe in some , not other situations

There are dozens of AA functions. Some work better on some types of content, but not on others

You have to try them out, adjust the settings. There is never "1 filter to rule them all" - we don't live on "Middle Earth"

[xuguang_he]

Apart from QTGMC, there seems no perfect solution to this source:
https://forum.videohelp.com/threads/355377-Interlacing-problem

But QTGMC do sth else that give the source a "oil painting" look while it also make the small interlacing gone. (TIVTC is not efficient when dealing with this source)

[xuguang_he]

Sometimes you need to apply different settings to different sections, different filter, or different approach entirely . QTGMC is a temporal filter and known to cause blends

And may I ask how to apply different settings to one source? (Searching through the forum, I've learnt to use Trim(), but the preview in MeGUI seems miserable, the ending of the film comes to the middle and the length of the film increased to 5 hours!!! Totally horrible.)

I'm ripping movies, so there is audio included. Is Trim() still available for this kind of situation?

[xuguang_he]

Sometimes you need to apply different settings to different sections, different filter, or different approach entirely . QTGMC is a temporal filter and known to cause blends

But apparently, in some of the examples I gave above, QTGMC decimate the blends. How to explain it may I ask?

[jagabo]

To remove field blending from PAL/NTSC conversions use:

Yadif(mode=1)
SRestore()

or

QTGMC()
SRestore()

Be sure the field order is set correctly first.

As usual, you should post a short sample of your source.

[xuguang_he]

So, is there any solution that could both decimate the interlacing artifacts completely and not to make the picture "oil painting" look?

[poisondeathray]

Sometimes you need to apply different settings to different sections, different filter, or different approach entirely . QTGMC is a temporal filter and known to cause blends

And may I ask how to apply different settings to one source? (Searching through the forum, I've learnt to use Trim(), but the preview in MeGUI seems miserable, the ending of the film comes to the middle and the length of the film increased to 5 hours!!! Totally horrible.)

I'm ripping movies, so there is audio included. Is Trim() still available for this kind of situation?

Search on how to apply different filters to different sections, there are many approaches and they are discussed in great detail with examples. e.g trim() with aligned splice, replaceframessimple() , clipclop() to name a few. The lenghts are unchanged when you use them properly


Trim works on audio as well , if you include in your script e.g. audiodub()

vid=mpeg2source()
aud=nicac3source()
audiodub(vid,aud)
trim(500,1000)

This would return frames 500-1000 with audio



e..g

orig=whateversource()

orig
trim(0,100)
filter1()
a=last

orig
trim(100,0)
filter2()
b=last

a++b

This would apply filter1 to 0-100, filter2 to 100-end, same length as original

[xuguang_he]

To remove field blending from PAL/NTSC conversions use:

Yadif(mode=1)
SRestore()

or

QTGMC()
SRestore()

Be sure the field order is set correctly first.

As usual, you should post a short sample of your source.

SRestore() is really destroyable.

{ Just Using QTGMC }


{ Using Both QTGMC and SRestore }


So, could I give the conclusion that SRestore add blended artifacts to the frame?

[jagabo]

Whether SRestore() will work or not depends on your source. Sample required.

[xuguang_he]

Whether SRestore() will work or not depends on your source. Sample required.

I've used DGIndex to cut the clip, but up till now, I could only cut out the first few seconds of a VOB file by demuxing it for a few seconds and exit the program as soon as possible (Then I get a m2v file of a VOB, but just the first few seconds of it). How could I just cut out a few frames which I intended to cut?

[xuguang_he]

Whether SRestore() will work or not depends on your source. Sample required.

https://forum.videohelp.com/threads/355377-Interlacing-problem
The Sample is cut from the same source as the sample posted in this thread.

[poisondeathray]

I've used DGIndex to cut the clip, but up till now, I could only cut out the first few seconds of a VOB file by demuxing it for a few seconds and exit the program as soon as possible (Then I get a m2v file of a VOB, but just the first few seconds of it). How could I just cut out a few frames which I intended to cut?

Navigate to the spot you want using the slider, then use the "<" and ">" buttons to jump to nearest keyframe. Then use the "[" and "]" buttons to mark in/out points respectively, file => save project and demux video

[jagabo]

Whether SRestore() will work or not depends on your source. Sample required.

https://forum.videohelp.com/threads/355377-Interlacing-problem
The Sample is cut from the same source as the sample posted in this thread.

That sample doesn't show the same problem. And it doesn't show the same blending problem.

[xuguang_he]

Here it is.

Apparently the original source is not blended, but somehow QTGMC plus SRestore made it so.

[jagabo]

The video doesn't have blended fields so there's no use for SRestore().

[xuguang_he]

But after using QTGMC, it do cause some new blended frames which original source don't have. This one is not included, but there are others.

[jagabo]

Then upload a sample of your source that shows the problem after QTGMC().

[poisondeathray]

But the original does have bended fields . From your first post "original" . So post a sample that has blends

SeparateFields()

This is the even field


This is the odd field

[xuguang_he]

I'll post it tomorrow, now is late night in China, sleep is the first matter...

[xuguang_he]

Here it is, the clip, which contained both blended and interlaced artifacts, and couldn't be decimate by srestore or deblend.

[jagabo]

There is no field blending in that clip. It's the same as the other clips but has a phase shift in some frames (some frames contain fields from two different film frames). Using Merge(SelectEven(),SelectOdd()) after QTGMC() is causing the blended frames when the two fields are out of phase.

For example, when using:

AssumeTFF()
QTGMC()
Merge(SelectEven(),SelectOdd())

frames 0 to 10 are in phase and show no blending artifacts. The rest of the frames are out of phase and show blending artifacts.

Using:

AssumeTFF()
QTGMC()
Trim(1,0)
Merge(SelectEven(),SelectOdd())

shows frames 0 to 9 with blending, the rest with no blending.

An automated way around this might be to use TFM() first, then follow with QTGMC() and Merge(...). It worked for this clip but could cause problems with others.

AssumeTFF()
TFM()
QTGMC()
Merge(SelectEven(),SelectOdd())

[Selur]

A2 = last
A2 = A2.QTGMC(Preset="slow", FPSDivisor=2)
TDeint(edeint=A2,emask=TMM(mode=0),slow=2,chroma=true,MI=80,full=false,cthresh=9)

looks fine here

[xuguang_he]

There is no field blending in that clip. It's the same as the other clips but has a phase shift in some frames (some frames contain fields from two different film frames). Using Merge(SelectEven(),SelectOdd()) after QTGMC() is causing the blended frames when the two fields are out of phase.

For example, when using:

AssumeTFF()
QTGMC()
Merge(SelectEven(),SelectOdd())

frames 0 to 10 are in phase and show no blending artifacts. The rest of the frames are out of phase and show blending artifacts.

Using:

AssumeTFF()
QTGMC()
Trim(1,0)
Merge(SelectEven(),SelectOdd())

shows frames 0 to 9 with blending, the rest with no blending.

An automated way around this might be to use TFM() first, then follow with QTGMC() and Merge(...). It worked for this clip but could cause problems with others.

AssumeTFF()
TFM()
QTGMC()
Merge(SelectEven(),SelectOdd())

Thank you for suggestion, it works fine. I may use Trim() just for this few seconds frame.