QTGMC with interlaced PAL

[ziggy1971]

I haven't used QTGMC before, so I'd like some advice for high quality settings while keeping the encoding time reasonable.

I have a TV series on DVD. The format is PAL and it varies from year to year on how (or whether) interlacing was used.
The first season is fine at all progressive. From there, it's basically a mess.

What would be some good settings to start off with for this SD video and what would be simply overkill?
Currently, I'm giving this a try:

SetMTMode(5, 12)
mpeg2source("H:\My Videos - Originals\_TV Series\Star Trek Voyager\Voyager S1D1\S01-E01_-_Caretaker.d2v", cpu=4, info=3)
ColorMatrix(hints=true, threads=0)
SetMTMode(2)
QTGMC( Preset="Slow", SourceMatch=3, Sharpness=1.2, SLMode=1 )
SelectOdd()

I'm using SelectOdd() because it's Top-Field First

I'm getting about 40 fps and have no idea what the quality will look like.
Any suggestions?

[jagabo]

With "interlaced" PAL, try using just TFM() rather than QTGMC().SelectOdd().

[ziggy1971]

Using the following results in a softer picture:

TFM(order=1, slow=2)

What's a good sharpening filter to use with this?

[jagabo]

Using the following results in a softer picture:

TFM(order=1, slow=2)

Because QTGMC is artificially sharpening, especially with sharpness=1.2.

But I should have specified to use TFM(pp=0) first and look for combing artifacts. If you don't see combing the underlying video is progressive. But instead of each frame containing fields from the same film frame they contain fields from two adjacent film frames. So instead of 1T+1B, 2T+2B you have 1B+2T, 2B+3T... With out-of-phase PAL video like that TFM() just recombines the fields to the restore original progressive frames. If that's the case there's no need for QTGMC followed by decimation by 1/2. Of course, some people use QTGMC anyway to get its noise reduction and edge enhancements. But in that case you can follow TFM() with QTGMC(InputType=1...).

What's a good sharpening filter to use with this?

Nobody can tell without seeing a sample. Sharpen(), LSFMod(), aWarpSharp2()...

[manono]

I'm using SelectOdd() because it's Top-Field First

I may be misunderstanding you here, but if you want to choose all the bobbed top fields then you want SelectEven. That way you'll get the first, third, fifth... frames. In other words, you'll get frame numbers 0, 2, 4, 6...

[ziggy1971]

Here is a section of one clip.

[manono]

It's not really interlaced, just phase shifted. Field matching it will fix it without degrading it by deinterlacing it:

TFM()

TFM is a part of TIVTC. And no one needed 9 minutes and 310 MB to figure out what was going on. Ten well chosen seconds would have been plenty.

[ziggy1971]

Sorry about the file size/length; better too much than not enough. I'll remove the video, if possible, or should I leave it for others to see/compare?
I just wanted to show the amount of changes throughout the video. Some parts seem to be interlaced every frame while others very randomly.

So, by just using

TFM()

the frames can be restored?
What can I use to sharpen the video, or should I just leave it as is?

I know what phase shifting is in audio, but I've never heard about it in video before. Is it common or is it just unlucky that this video has it? If it's common I'll have to read up on it.

Thanx for helping out...

[manono]

It's fairly common with PAL sources, both broadcast and DVD, less so with NTSC DVDs. Everything I saw was phase shifted, although I don't pretend to have gone over the whole thing thoroughly. If there were progressive sections, that's fine and TFM won't do anything to those.

It looked plenty sharp to me. When I sharpen, it's with LimitedSharpenFaster. It might be better to concentrate your efforts on fixing the levels. The whites are somewhat blown out in that sample.

[jagabo]

So, by just using

TFM()

the frames can be restored?

Yes. I took a look at the MPG file too. It does alternate between in-phase and out-of-phase sections. If the entire video was out of phase you could fix it with SeparateFields().Trim(1.0).Weave(). But when the phase switches back and forth the out-of-phase sections will become in-phase and the in-phase sections will become out-of-phase. TFM() will adapt to the phase changes leaving only progressive frames. It will also deinterlace any orphaned fields. (An orphaned field is a field for which the other half of the film frame is missing.)

What can I use to sharpen the video, or should I just leave it as is?

Try something like:

Santiag()
LSFMod(strength=100).

Adjust strength to your liking. Santiag reduces aliasing artifacts.

I know what phase shifting is in audio, but I've never heard about it in video before. Is it common or is it just unlucky that this video has it? If it's common I'll have to read up on it.

I already explained what it is in post #4. In short. pairs of fields from adjacent film frames are paired together in the video frames, rather than both fields of the video frame coming from the same film frame.

[ziggy1971]

If this "phase" issue is common among PAL sources I've probably edited/converted all my previous sources incorrectly. I've always treated them as "Interlaced" sources.

Many years ago I read through the tutorial here:
https://www.animemusicvideos.org/guides/avtech/index.html

I see now that many things have changed at that website.

For instance, one of the ways I was using to deinterlace was with this (should give an idea how long ago it was):
AMVDeint(mode=3)

So, now I'm left with more questions, like how did you determine that it was a "phase" issue and not interlacing?

I have more questions regarding color spaces, so no longer appropriate for this thread. Will start a new one.

[jagabo]

how did you determine that it was a "phase" issue and not interlacing?

A simple Bob() will show pairs of identical images (aside from aliasing artifacts of the bob). Or just try TFM(pp=0) and see if you get frames without any interlacing. Look especially at panning shots. Real interlaced PAL will show movement with every field. Out of phase progressive will show movement only every other field.

[ziggy1971]

I just tried Bob() and I can see what you mean, simple enough.

Could you help me with the color spaces, or should I start a new thread?

[jagabo]

I suggest you start a new thread. That will be more useful for others in the future.

[hello_hello]

I use QTGMC in progressive mode to "clean up" progressive video quite a bit. It's sharpening is fairly mild, similar to when using it for de-interlacing. So if a video is progressive.... or if it's progressive with fields out of phase.....

TFM() # only to fix out of phase fields
QTGMC(InputType=1, Preset=Medium, EzDenoise=1.5) # the de-noising is optional, depending on the source, but it helps with cleaning up artefacts
LSFMod(Strength=75) # I'd never sharpen video with artefacts, and I dislike video that looks "sharpened", but if I sharpen it's with LSFMod
GradFun3() # from the dither package to prevent banding for 8 bit encoding.

I encoded this a while ago, but I'm fairly sure it was encoded with the settings below, or very close to them (plus cropping and resizing to 960x540). It also included a conversion to RGB and back before QTGMC in the script so I could tone down the PAL "green" just a tad. Don't compare them without opening each full size in a new tab or saving them first as the forum software sharpens a lot when it resizes.

ConvertToRGB()
RGBAdjust(1.0,0.98,1.01)
ConvertToYV12()
QTGMC(InputType=1, Preset=Medium, EzDenoise=1)
LSFMod(Strength=75)
GradFun3()

Original


Encode


For cleaning up any noise in progressive mode I prefer the medium QTGMC speed preset, or faster. The logic being.... the slower speed presets tend to stabilise any remaining noise too much for my taste and it can result in a kind of "looking through a fly-screen" effect where any remaining noise appears "static" over the picture. The faster presets don't stabilise as much and any remaining noise stays more noise-like.