deinterlace filters in megui vs qtgmc

[codemaster]

MeGUI includes the following deinterlacing filters: Yadif, Yadif (with Bob), TDeint, TDeint (with Bob), TDeint (with EDI), TDeint (with EDI + Bob), LeakKernelDeint, TomsMoComp, Field Deinterlace, Field Deinterlace (no blend).

Is QTGMC with it's default settings and parameters, a better quality deinterlacer than those included with MeGUI ?

Is it worth it to switch from Yadif to QTGMC ?

[sanlyn]

Yes.

[manono]

Is it worth it to switch from Yadif to QTGMC ?

Only you can be the judge of that - if the quality gain is worth the longer encoding time. I use Yadif for stuff I don't much care about. It's usually 'good enough'. For stuff for which I want the best result I use QTGMC.

[codemaster]

So QTGMC with it's defaults, is a higher quality deinterlacer than all other deinterlacers that exist for AviSynth ?

[codemaster]

Is it worth it to switch from Yadif to QTGMC ?

Only you can be the judge of that - if the quality gain is worth the longer encoding time. I use Yadif for stuff I don't much care about. It's usually 'good enough'. For stuff for which I want the best result I use QTGMC.

Ok, I'll use QTGMC only for the important stuff. Good idea.

[manono]

So QTGMC with it's defaults, is a higher quality deinterlacer than all other deinterlacers that exist for AviSynth ?

Without a doubt. Of course, you have to be sure a deinterlacer is called for. Many times people, for one reason or another, think their source needs deinterlacing when, in reality, it might need an IVTC, unblending, field-aligning, or nothing at all (encoded as interlaced, with progressive content). And QTGMC by default is a bobber (doubles both the framecount and the framerate) so be sure that's what you want, too. To make it single-rate, add a 'SelectEven()' afterwards.

[codemaster]

But, if mediainfo says that source is 25fps interlaced top field first, and QTGMC makes 50 fps, then all frames will get 1/2 of the vertical resolution ? Because, in order to maintain an aspect ratio close to source, I don't want to upscale, I only downscale, to avoid interpolation. And if I add SelectEven(), I'll have 25fps, and all frames will have 1/2 of the vertical resolution, while Yadif keeps the vertical resolution intact ?

[sanlyn]

No, all frames would use half the vertical resolution and interpolate a full image from there. Yadif used in the same way will do the same thing. Think about it. Even if you keep all the frames, deinterlacers separate fields. Fields have 1/2 the vertical resolution of the original frame. A full image is interpolated from that 1/2 image to get a full image, even if you keep all the frames. That's why people often ask: why are you deinterlacing?

[manono]

But, if mediainfo says that source is 25fps interlaced top field first...

Almost all PAL stuff is encoded as interlaced, both true interlaced material as well as progressive sources. If it's a movie it almost certainly doesn't need deinterlacing. If you don't see any interlacing, then it's not interlaced. And even if you do then it still might not be. Any sources for which you have questions post unprocessed samples here.

And if I add SelectEven(), I'll have 25fps, and all frames will have 1/2 of the vertical resolution, while Yadif keeps the vertical resolution intact ?

What makes you think Yadif is any better at this than is QTGMC? Yadif has a bob mode too. But it's the default for QTGMC and you have to turn on Mode=1 in Yadif to use it. The idea for all good deinterlacers is that frames or parts of frames without movement return the full resolution. QTGMC process everything to a degree, but in frames with movement it leaves behind far fewer artifacts than does Yadif. Have a look at a diagonal line in something where Yadif was used. Then compare with the same line processed with QTGMC.

In any event, it's always better not to deinterlace progressive material.

[jagabo]

But, if mediainfo says that source is 25fps interlaced top field first, and QTGMC makes 50 fps, then all frames will get 1/2 of the vertical resolution ? Because, in order to maintain an aspect ratio close to source, I don't want to upscale, I only downscale, to avoid interpolation. And if I add SelectEven(), I'll have 25fps, and all frames will have 1/2 of the vertical resolution, while Yadif keeps the vertical resolution intact ?

Just try it. You'll find there's no comparison between QTGMC and all those other deinterlacers. Deinterlacing is the attempt to create a full pictures from half pictures. The two half pictures in an interlaced frame can be made into one full picture or two full pictures.

[hello_hello]

I agree. I use MeGUI myself, but these days I mainly de-interlace with QTGMC and try not to spend too much time regretting the number of old encodes where I didn't.

If a picture is worth a thousand words, I guess a video sample must be worth a few more. Have a look at the ones I posted in this thread (posts 8 and 20) comparing QTGMC to Yadif at both 25 and 50fps. In the first samples, keep your eye on the blinds as the camera moves.

Even though it's primarily a de-interlacer, it has a progressive mode, so I often use it when re-encoding badly de-interlaced progressive video and even as a pseudo noise filter. In progressive mode it tends to "stabilise" the picture, including the noise, rather than remove it as such, but once again that means it doesn't blur as noise filters often do.

If you're interested (getting off the actual topic of de-interlacing a little) I posted some sample encodes here comparing standard noise filtering to runing QTGMC in progressive mode. It didn't seem to attract much interest, but I've been using QTGMC as a "psuedo" noise filter more and more. Currently I'm using it in progressive mode while re-encoding some old TV captures and I'm very pleased with the result. For most I've not used additional noise filtering, while for the really noisy ones I'm using QTGMC's EZDenoise.

[sanlyn]

I caught your noise reduction demo earlier. Nice post. I find QTGMC's script to be a useful learning tool for studying ways that many of its support files can be used as stand-alone plugins. Some of it is a bit cryptic for most users (including myself), but it's still a good source of info. Thanks for sharing.

[hello_hello]

I caught your noise reduction demo earlier. Nice post. I find QTGMC's script to be a useful learning tool for studying ways that many of its support files can be used as stand-alone plugins. Some of it is a bit cryptic for most users (including myself), but it's still a good source of info. Thanks for sharing.

Apologies to the OP for wandering off topic a little......

I ran some more comparison encodes between QTGMC in progressive mode, and the two script-based denoisers I've been playing with, TemporalDegrain, and FastDegrain. I used the default settings for each, but also adding QTGMC's EZDenoise function into the mix. The source videos (old TV captures) range from average quality to very poor quality with lots of noise.

So far.....
TemporalDegrain's default settings and QTGMC+EZDenoise4.0 seem to produce almost identical results (also confirmed by the almost identical output file size). Occasionally when comparing the two frame by frame I thought I preferred the TemporalDegrain encodes, but the differences were very slight and could possibly be due to x264's compression (I didn't check if I was comparing B frames with B frames etc, although I used the same encoder settings each time). Over-all though, I preferred QTGMC+EZDenoise4.0 as the picture itself was a little more stable (once again though the difference was very small) however QTGMC+EZDenoise4.0 repaired more of the "bad-deinterlacing" nastiness. And in single threaded mode, it seems to be marginally faster than TemporalDegrain (only around 1fps faster for me).

FastDegrain seems to be pretty similar to QTGMC+EZDenoise=2.0 when it comes to noise removal, and it's much faster (QTGMC in single threaded mode) but at the same time doesn't come close to QTGMC's ability to stabilise the picture.

If the slow speed doesn't bother you, for removal of lighter noise, QTGMC in progressive mode without EZDenoise enabled is definitely my preference. It removes most of the noise while stabilising the picture and doesn't blur it as FastDegrain and TemporalDegrain do. If anything, it'll sharpen fine detail a little. For heavier noise sometimes just "stabilising" it isn't enough as it doesn't look natural, so it's not a magic fix which eliminates the need for noise removal, but even with it's own noise removal enabled, I still prefer QTGMC.

[codemaster]

I did some tests on my source. I got satisfactory results only with

QTGMC(Preset="Placebo", FPSDivisor=2)

Almost no artifacts. Very few, but very small and almost unnoticeable. With

QTGMC(Preset="Slower", FPSDivisor=2)

there are a little bit fewer artifacts than with

Yadif(order=1)

which I used in the past and won't go back to from now on.

[hello_hello]

I did some tests on my source. I got satisfactory results only with

QTGMC(Preset="Placebo", FPSDivisor=2)

Almost no artifacts. Very few, but very small and almost unnoticeable. With

QTGMC(Preset="Slower", FPSDivisor=2)

there are a little bit fewer artifacts than with

Yadif(order=1)

which I used in the past and won't go back to from now on.

The way I understand it, without "ShutterBlur" and/or other motion blur settings enabled,
QTGMC(Preset="Placebo", FPSDivisor=2)
is the same as
QTGMC(Preset="Placebo")
SelectEven()

I probably should check that to make sure there aren't any differences, but either way you're de-interlacing to "half frame rate".

If I'm correct regarding the above, then
QTGMC(Preset="Slower", FPSDivisor=2)
Would be the same as
QTGMC()
SelectEven()
Because "slower" is the default speed preset.

Have you tried de-interlacing to "full frame rate" instead?
I think the idea behind FPSDivisor and ShutterBlur etc is to add a little motion blur when de-interlacing to "half frame rate" to make motion appear a little smoother, but in my opinion "full frame rate" is the way to go. I haven't played around with the motion blur settings as once I discovered full frame rate de-interlacing...... Most/many current devices should play 50fps/60fps video (both the Bluray players in this house do) and when using CRF encoding it actually doesn't increase the file size all that much, and full frame rate de-interlacing looks so much better.

[hello_hello]

As a quick experiment, because I had some PAL concert footage handy (lots of flashing lights and motion, hard to de-interlace well) I ran two encodes using a 3 minute section of the video. Same cropping and resizing each time.

QTGMC(Preset="Fast", FPSDivisor=2)
Produced exactly the same output as
QTGMC(Preset="Fast")
SelectEven()
At least as best as I could tell, but the output bitrate was the same and the final file sizes only differed by about 1KB. (CRF18 and default x264 settings each time).

[jagabo]

I did some tests on my source. I got satisfactory results only with

QTGMC(Preset="Placebo", FPSDivisor=2)

Different material may require different settings. If there is a lot of motion and a lot o very thin near horizontal lines or edges it can be very difficult deinterlace. Most material looks fine with the default (slower) setting.

[codemaster]

Have you tried de-interlacing to "full frame rate" instead?
I think the idea behind FPSDivisor and ShutterBlur etc is to add a little motion blur when de-interlacing to "half frame rate" to make motion appear a little smoother, but in my opinion "full frame rate" is the way to go. I haven't played around with the motion blur settings as once I discovered full frame rate de-interlacing...... Most/many current devices should play 50fps/60fps video (both the Bluray players in this house do) and when using CRF encoding it actually doesn't increase the file size all that much, and full frame rate de-interlacing looks so much better.

Yes, I realize that with original frame rate, motion blur should probably be tweaked. But I'll choose doubled frame rate, because motion looks much more fluid and smooth, and even the old 2002 pc can play it back without dropping frames.

If choosing doubled frame rate, will this make it unnecessary to compensate with motion blur tweaks ?

[jagabo]

If choosing doubled frame rate, will this make it unnecessary to compensate with motion blur tweaks ?

Yes. See 24v30v60.avi in this post:
https://forum.videohelp.com/threads/307004-Best-framerate-conversion-%28eg-23-97-to-30-...=1#post1888926