2:4 Pulldown

[supervehicle]

I'm currently working on a fan edit of Pink Floyd: Live at Pompeii, and it appears the NTSC DVD was transcoded from a PAL source, thus resulting in an odd 2:4 (2 clean, 4 interlaced) interlacing pattern. Does anyone know what an AviSynth script should look like if I plan to deinterlace this?

I've posted a clip of it online, if anyone's interested.

[jagabo]

Mpeg2Source()
Yadif(mode=1, order=1)
SRestore()

[supervehicle]

Mpeg2Source()
Yadif(mode=1, order=1)
SRestore()

Hmm, that didn't seem to do it. Some frames were doubled, resulting in a 28.74 frame rate. I'm trying to get back to 25p.

[jagabo]

It worked for me with your sample clip. Try specifying the frame rate in SRestore(frate=25.0).

[supervehicle]

It worked for me with your sample clip. Try specifying the frame rate in SRestore(frate=25.0).

It worked! Thanks a bunch, jagabo! There is, however, this new flickering line that appears at the bottom of the video every couple frames. Any idea on how to fix that?

[jamespoo]

i think i should know how to make love

[jagabo]

You mean the slightly darker line at the bottom of the frame -- even in the original source? You could mask it or try a strong temporal filter just at the bottom of the frame.

[supervehicle]

You mean the slightly darker line at the bottom of the frame -- even in the original source? You could mask it or try a strong temporal filter just at the bottom of the frame.

No, it only appears after applying the script.

[jagabo]

You mean the slightly darker line at the bottom of the frame -- even in the original source? You could mask it or try a strong temporal filter just at the bottom of the frame.

No, it only appears after applying the script.

I'm not seeing that. Try updating some of the AviSynth filters. I seem to recall the Yadif used to have a problem like that.

[manono]

Yeah, sometimes you get that bobbing effect. Sometimes this works:

Mpeg2Source()
AddBorders(0,8,0,8)
Yadif(mode=1, order=1)
SRestore()
Crop(0,8,0,-8)

Sometimes using a different bobber works (QTGMC). Sometimes you have to crop and resize or just mask it off by covering it with black.

[supervehicle]

Yeah, sometimes you get that bobbing effect. Sometimes this works:

Mpeg2Source()
AddBorders(0,8,0,8)
Yadif(mode=1, order=1)
SRestore()
Crop(0,8,0,-8)

Sometimes using a different bobber works (QTGMC). Sometimes you have to crop and resize or just mask it off by covering it with black.

Thanks manono. While that didn't fix it entirely, it's definetley less noticable now. But I'll try using a different bobber.

[manono]

Easiest might be to just cover the bottom with a couple of rows of black pixels:

Letterbox(0,2)

If your TV set has overscan you won't see it anyway.

[supervehicle]

I actually got better results with TomsMoComp. My script looked something like this:

LoadPlugin("TomsMoComp.dll")
AviSource()
ConvertToYV12()
TomsMoComp(-1,5,1)
SRestore(frate=25.0)

[supervehicle]

Thanks again, jabago and manono. Really appreciate your help.

[jagabo]

Which deinterlacer works best can depend on the nature of the source. You might also try nnedi3() or QTGMC() as manono suggested earlier. QTGMC() almost always gives the best results but it is very slow. A good compromise is QTGMC(preset="fast"). Your sample with QTGMC() attached.

[supervehicle]

Which deinterlacer works best can depend on the nature of the source. You might also try nnedi3() or QTGMC() as manono suggested earlier. QTGMC() almost always gives the best results but it is very slow. A good compromise is QTGMC(preset="fast").

Yeah, I gave QTGMC a try, definitely an improvement over Yadif. I personally don't mind if it runs slow, just as long as I'm satisfied with the end result.

My script looked something like this:

AviSource()
ConvertToYV12()
QTGMC(preset="fast")
SRestore(frate=25.0)

Thanks yet again, jagabo.

[jagabo]

If you don't mind long encoding times the default settings of QTGMC() will look even better.

[supervehicle]

So I did another test, this time on a clip with multiple scenes. It turned out alright. I'm assuming, however, that tweaking the scene change detection parameters in MVTools could give better results, although I'm not sure what exactly needs to be added to the script.

[supervehicle]

So I figured it out.

It's as easy as adding "QTGMC(ThSCD1= ,ThSCD2= )" to my script. Since the defaults are ThSCD1=180 and ThSCD2=98, I raised each parameter a bit (so I went with "ThSCD1=195" and "ThSCD2=113").

Unfortunately, nothing seemed to change.

I could raise the parameters even higher, but that probably isn't a good idea, as scene changes could be mistakenly detected.

[jagabo]

You must have an older version of QTGMC(). I have 3.0 now and the thscd1 and thscd2 parameters were removed. The only way to set them in 3.0 is to modify QTGMC.AVSI.

<edit>

No, I have the older one! LOL. I see that 3.31 has reinstated those parameters.

</edit>

[poisondeathray]

What was wrong with the scene changes ?

[supervehicle]

I figured I'd upload some screen caps to better illustrate my issue (this is what I get AFTER deinterlacing with QTGMC):

Here's a frame.



And the following frame (also prior to a scene change).



Next frame.



And the one after that.



Here's another example:

[manono]

That has nothing to do with the deinterlacer used. However you bob it you'll get that crap at scene changes. You can thank the PAL2NTSC standards conversion for that. I know of one workaround, but it involves duplicating the frames both before and after the scene change. Unless you want to go through the whole thing manually using a FreezeFrame on the scene changes where you have the problem. Anyway, here's an automatic way to fix it:

A=Last
prev = A.selectevery(1,-1)
next = A.selectevery(1,1)
SCclean = A.SCSelect(next,prev,A,dfactor=2.0) # 2.0 ~ 5.0
return(SCclean)

Thanks to Didée for the script. You'll need v0.9 of RemoveDirt for it to work:

[supervehicle]

jagabo, manono, thanks a million.

You guys are lifesavers.