Convert Interlaced video to Progressive?

[ziggy1971]

Hello all,

I have a bunch of short video clips that are all interlaced, DGIndex says that the Frame Type is interlaced and I can clearly see the interlacing throughout the video. The clips do not come from a DVD source so I don't think you can restore the original progressive video.

I've been trying different .avs scripts with VirtualDubMod but I find that I get different issues with different scripts. For example: even though a script may do a good job of deinterlacing the video, I still see duplicate frames when I step through the video with VirtualDubMod.

Here are a couple of the scripts I've been trying:
Telecide(order=1,guide=0)
Decimate(cycle=5, mode=2)

or:
TomsMoComp(1,6,1)

I've been reading a lot of different threads on the forums on some tutorials on avs scripts, but I still can't seem to find a good combination.

Any advice on this is greatly appreciated,
Thanx for now,
ziggy1971

heart%20-%20alone2.mpg

[jagabo]

Try Yadif():

Load_Stdcall_plugin("C:\Program Files\AviSynth 2.5\plugins\yadif.dll")
MPEG2Source("C:\Documents and Settings\John\Desktop\heart - alone2.d2v")
Yadif(order=1, mode=0)
TDecimate()

But music videos typically go through all kinds of speedup and slowdown so you won't be able to get a nice clean jerk-free 23.976 fps video.

[45tripp]

well for the most part it's vanilla telecine.
aside from 3 scenes that aren't vanilla. pretty freaky.
and the fact that there's combing on the fields.
vinverse() that.

for the most part it'll be smooth.

pretty shitty though otherwise.

[ziggy1971]

Hey guys,

Thanx for taking a look at the video for me and the suggestions. I'll have to give them a try and see what I can come up with.

Yeah, I know the video is pretty crappy to start off with, I thought that someone might have come across a good way of dealing with it.

Perhaps this video will be a better example because they all come from the same source and maybe get better results.

Thanx again,
Cheers,
ziggy1971

ashanti_-_good_%20good.mpg

[jagabo]

ashanti_-_good_%20good.mpg

You could do this:

src = MPEG2Source("ashanti_-_good_good.d2v").ConvertToYUY2(interlaced=true).As sumeTFF()

v1=Trim(src, 1,18).SeparateFields().SelectEvery(10, 0,1, 2,3, 4,5, 7,8).Weave()
v2=Trim(src, 19,73).SeparateFields().SelectEvery(10, 0,1, 2,3, 5,6, 8,9).Weave()
v3=Trim(src, 74,108).SeparateFields().SelectEvery(10, 1,2, 4,5, 6,7, 8,9).Weave()
v4=Trim(src, 109,224).SeparateFields().SelectEvery(10, 0,1, 3,4, 6,7, 8,9).Weave()
v5=Trim(src, 225,260).SeparateFields().SelectEvery(10, 0,1, 3,4, 6,7, 8,9).Weave()
v6=Trim(src, 261,286).SeparateFields().SelectEvery(10, 0,1, 2,3, 5,6, 8,9).Weave()
v7=Trim(src, 287,297).SeparateFields().SelectEvery(10, 1,2, 4,5, 6,7, 8,9).Weave()

return(v1+v2+v3+v4+v5+v6+v7)

[45tripp]

aha,
why?

looks perfectly regular to me.

tfm()
tdecimate()

also got rainbows, ghosting, blocking, ringing.
nice

[jagabo]

Mostly I was showing another option. I wouldn't bother doing this type of manual IVTC with a video like this. There are too many telecine pattern breaks. But watch the door closing:

TFM:


manual:

[45tripp]

yes on every cut.

i'd not think it ever worth it.


i don't see what you're showing me.
i've got no aliasing.
got a bobber in there?

and in fact,
taking you're showing me frame 60,
with your posted script, frame 59 is still interlaced


tripp

[jagabo]

I didn't bother "fixing" frame 59 because a) It's not my video and I don't care, b) it contains orphaned fields, c) a single interlaced frame at a transition isn't critically important. I also skipped the first frame because it contains an orphaned field, and the last two frames because the sequence was short and I was only demonstrating the method.

I do manual IVTCs when there are only a few telecine breaks. It's far cleaner than any of the automated algorithms. TFM left several badly IVTC frames like the one I posted. I used the exact script you posted TFM().TDecimate().

[45tripp]

c) a single interlaced frame at a transition isn't critically important.

by the same token,
neither are a few frames with some aliasing

I do manual IVTCs when there are only a few telecine breaks. It's far cleaner than any of the automated algorithms. TFM left several badly IVTC frames like the one I posted. I used the exact script you posted TFM().TDecimate().

found it.
but a bad match is a bad match.
there are counters.
i'll call it the neutral aproach, rather than the cleaner. 8)


call me lazy :P


edit:
more palatable
http://www.mediafire.com/?x7zfckvgngk

[jagabo]

c) a single interlaced frame at a transition isn't critically important.

by the same token,
neither are a few frames with some aliasing ;)

And if that was the only problem I wouldn't mind either. But you'll often get longer errors. For example that whole shot with the closing door might have a buzzing line (it didn't happen in this case). A shot with a skyscraper (thin horizontal lines) in the background will have buzzing moire artifacts. Jerks during Pans because a frame was repeated and another was dropped. Etc. A small object moving in the distance will jerk or still be interlaced, or blend deinterlaced unnecessarily, etc.

I don't mind spending 10 minutes setting up a manual deinterlace to get a cleaner IVTC.

[ziggy1971]

Hey guys,

Clearly you guys have a lot more experience cleaning up this kind of video because I have little or no idea what to look for when doing this.

I looked at the clip 45tripp uploaded and it looks really good.

Now I just have to figure out how to do that on my own, so I'm wondering where you guys learnded how to see things like "...buzzing moire artifacts, jerks during pans because a frame was repeated and another was dropped, etc." and how to deal with the different situations.

I don't mind spending some extra time setting things up right if I get good results in the end, but at this point I don't know how to even analyze footage properly (at least not after this), but it seems that the result is well worthwhile.

Could someone direct me to some good tutorials and how to analyze the video and what to do with each situation, or do you just know how to do this with experience?

For instance, where or how do you come up with a script like this?

src = MPEG2Source("ashanti_-_good_good.d2v").ConvertToYUY2(interlaced=true).AssumeTFF() 

v1=Trim(src, 1,18).SeparateFields().SelectEvery(10, 0,1, 2,3, 4,5, 7,8).Weave() 
v2=Trim(src, 19,73).SeparateFields().SelectEvery(10, 0,1, 2,3, 5,6, 8,9).Weave() 
v3=Trim(src, 74,108).SeparateFields().SelectEvery(10, 1,2, 4,5, 6,7, 8,9).Weave() 
v4=Trim(src, 109,224).SeparateFields().SelectEvery(10, 0,1, 3,4, 6,7, 8,9).Weave() 
v5=Trim(src, 225,260).SeparateFields().SelectEvery(10, 0,1, 3,4, 6,7, 8,9).Weave() 
v6=Trim(src, 261,286).SeparateFields().SelectEvery(10, 0,1, 2,3, 5,6, 8,9).Weave() 
v7=Trim(src, 287,297).SeparateFields().SelectEvery(10, 1,2, 4,5, 6,7, 8,9).Weave() 

return(v1+v2+v3+v4+v5+v6+v7)

(OK, that didn't turn out the way I expected. )

As for the 2nd clip I uploaded, the quality of the video may have been degraded somewhat because I had to render a short clip at less that 1/4th of the original bitrate to get something small enough I could upload to this site. (The original is about 128 MB's)

Thanx again guys,
Cheers,
ziggy1971 8)

[jagabo]

Finding all the telecine pattern breaks and determining the proper SelectEvery() sequence is a bit hard to explain. I posted an explanation on a newsgroup a year or so ago but it's probably gone by now. I don't know of any guides. It works for video with normal 3:2 pulldown but is impractical with odd pulldown patterns (like the first video you posted).

Given the difficulty of the manual IVTC and the large number of pattern breaks you're probably best off using the simple TFM().TDecimate() script posted by 45tripp. You can also play with some of the other deinterlace/IVTC filters like Yadif(), TDeint(), TempGaussMC_beta1, etc.

[45tripp]

And if that was the only problem I wouldn't mind either. But you'll often get longer errors. For example that whole shot with the closing door might have a buzzing line (it didn't happen in this case). ...

worst case scenarios.
if you have something particularly ugly on hand,
i'll take a sample.

ivtc is tweakable.

sure there'll always be the case you're practically forced into manual corrections,
whatever it is you might be doing,
and i tend to do most things manually too (foolishly; then again there's zero volume to what i actually do),
but at some point you've got to draw the line and let the blessed machine earn it's keep.

i think i'm also getting less demanding as i go along.
and i'd not palm complexity to a noob.

I don't mind spending 10 minutes setting up a manual deinterlace to get a cleaner IVTC.

when you make it sound simple,
it's hard to argue

how many breaks, is a few?

For instance, where or how do you come up with a script like this?

you have a 2:3 pattern
http://en.wikipedia.org/wiki/Telecine
so you take each frame apart into it's 2 fields,
{ separatefields() }
for every 10 fields, you want to keep 8, discarding each third duplicate.
{ SelectEvery(10, 0,1, 2,3, 5,6, 8,9) }
meaning: select out of each sequence of 10, the following...
and you see how he matches up pairs of fields.
then put them back together again.
{ weave() }

but the pattern breaks, like with edited stuff,
and with music videos, expect a break at each cut,
and not all scenes to be 2:3, and other uglies...

but if it's all 2:3,
just cut at each pattern break, (kill/handle orphans)
and reset

something like that.

Could someone direct me to some good tutorials

experience,
yeah.

some good starting point tuts are:
stuff from here: http://www.doom9.org/index.html?/capture/start.html
and here: http://www.animemusicvideos.org/guides/avtech/

and maybe find enthusists on a forum that work with the same type of stuff you usually do.


heart alone:
http://www.mediafire.com/?q1yxsnzu1ai



tripp

[jagabo]

I don't mind spending 10 minutes setting up a manual deinterlace to get a cleaner IVTC.

when you make it sound simple,
it's hard to argue :)

how many breaks, is a few?

Sometimes a movie (recorded off a channel with no commercials, or ripped from DVD/laserdisc/VHS) will have no pattern breaks at all. Sometimes just one break in the middle. Or maybe a break at each commercial. If it looks like it will go over a dozen or so breaks I'll switch to an automated IVTC. Exactly where I draw the line depends on the how much I care about a particular video, how easy it would be to get a clean version otherwise (is it available on DVD encoded progressively with pulldown flags? will it likely be soon?, etc), and how enthusiastic I'm feeling that day.

[manono]

I used to do similar things with complete movies on DVD before there were decent unblenders for unusual framerates. A good one had 20-30 breaks. The bad ones went up into the hundreds - almost every scene change sometimes - and I quickly gave up.

If you're going to check out every frame for aliasing and otherwise poor matches when using automatic IVTCs, it seems to me that override files are the way to go. Both Decomb and TIVTC support them.

[45tripp]

kissed momonster yet?

[manono]

kissed momonster yet?

That's funny. I was doing this with silent films with any kind of framerates. RePAL was available for PAL2NTSC stuff, but that's about it. Then later came Restore24, the first decent unblender for any framerate, and then later Momonster's work with CDeblend and MRestore.

This thing was called UnblendPattern and was an invention of scharfis_brain:

http://forum.doom9.org/showthread.php?p=406949#post406949

And it would drive you nuts. You'd find a working pattern and then check scenes every so often until you found it broken. Then you'd have to back up and find the exact frame or scene change where the pattern changed, and then try and pick out the new pattern. I did a few complete movies with it, each taking many hours of drudge work to finish. And there were others with so many pattern changes that I'd give up after a while.

So yes, I've often felt like, if not kissing him, at least praising him for making the unblending work much easier now.

[ziggy1971]

Hey guys,

For those who would like to see the full videos that I previously had clips for, please feel free to download them here:

http://www.zaxx-designs.com/temp/

Perhaps this will help in determining the best course of action.

Thanx again for all the help.
Cheers,
ziggy1971

[45tripp]

http://forum.doom9.org/showthread.php?p=406949#post406949

saw that,
but have not had the pleasure.

So yes, I've often felt like, if not kissing him, at least praising him for making the unblending work much easier now.

I have no qualms, so i might do it for you

For those who would like to see the full videos that I previously had clips for, please feel free to download them here:

the samples were representitive.
what's the target btw?


tripp

[ziggy1971]

Hey guys,

Sorry for not being clear on the target format. After reading through the tuts at doom9 I found that it's supposed to be one of the first things to decide, oh well, noobs!!!

The target format that I want is 512x384 with the Xvid codec and mp3 audio. So I guess it leaves a little room for resizing the original video.

I've been looking at some of the other videos I have and some of them I know will be much more difficult to deal with as they show no real pattern at all.

I'll try to post a sample ASAP.
Thanx again for all the help,
Cheers,
ziggy1971 8)

[45tripp]

heart alone
http://www.mediafire.com/?mm1dvmmnmmz

ashanti
http://www.mediafire.com/?woydxamimn0


done with tfm()
with 'alone' you can see some noticeable jerk in one zoom.
and with 'ashanti' you can see it's not plain 3:2 all through,
there's some crap too... not really worth note, but still indication
of music video uglies.


tripp