Need help TIVTC'ing/Deinterlacing a clip...

[Ultimist]

*** Edited to implement manono's suggestions for uploading the files. Thanks manono! ***

I have a vob of a music video, ripped from DVD, that I'm trying to get into progressive format to keep on my PC hard drive.

The music video is on film, originally in 23.975 fps, but obviously converted to 29.97 fps for DVD. It's got some interesting problems. It seems double interlaced or something. The problem is really hard to describe.

I've tried everything from TIVTC to AnimeIVTC (even though it isn't intended for live action material) to QTGMC, with mixed results, and never exactly what I want, which is a perfectly progressive clip.

Even though tools like AnimeIVTC aren't intended for live action material, I've had success using it to fix problems with other live action vobs. As with everything AVISynth, sometimes scripts can be used successfully for purposes other than what was intended.

But in this case, I can't find anything that works. I can get part of the video looking progressive, but then the motion is very choppy... or I can get the motion looking ok, but the clip remains combed or interlaced in places, and so it looks terrible.

I've uploaded 2 clips in original m2v format (no sound) with no deinterlacing or pulldown removal applied, demuxed and cut from the unaltered vob file.

The first is the start of the video, where a native 29.97fps "intro" has been attached to the beginning, as they sometimes do on DVD compilations. I want to cut off the intro, and have the rest at a progressive 23.976fps. You'll see in the video, when the camera begins to pan across a beach with people holding hands.. I've tried everything and the clip ends up in either of 2 states:

1. Individual frames look progressive, but the motion clip shows a lot of annoying shimmering of the rocks on the beach as the camera pans across... or

2. Each frame IS progressive, and the beach doesn't shimmer, but the resulting clip suffers from some kind of severe motion judder problem. The smooth pan of the camera is transformed into a herky-jerky start and stop affair as the camera pans. Each individual frame is perfectly progressive in this case, but there seem to be frames missing, frames that would smooth out the motion as it should be.

The 2nd clip, after a few seconds, has the camera panning across a brick wall, down to some people standing in the street. That brick wall is wreaking havoc on any sort of deinterlacing/decombing/de-ivtc'ing I've tried. The wall shimmers instead of showing the bricks properly, or I get this weird cross-hatch shimmering effect coming from the wall. I've been hoping to fix this clip for years.

A couple of years ago I did something with a combo of animeivtc and virtualdub and got a perfectly progressive clip from start to end, but the whole thing suffered from that herky jerky motion judder.

I was just wondering if some helpful soul out there could take a look at these clips, perhaps figure out what the problem is, exactly, and offer suggestions on how to fix it?

My problem, partly, is that I don't have a full understanding of how pulldown removal, ivtc, deinterlacing, are supposed to work... the proper order of operations, etc..

So any help would be SO greatly appreciated.

Here are the 2 clips, uploaded to mediafire, to demonstrate my problem:

http://www.mediafire.com/watch/zjlrra068scoea0/Clip1.m2v

http://www.mediafire.com/watch/a75uzxdfs1v191d/Clip2.m2v

If someone can help me figure out what to do with this clip, I will be the most grateful person on the planet.

Thanks!

[manono]

If your source is a DVD why are you making available reencoded x264 in AVI? Upload either a section of the VOB (and smaller than 136 MB) or the M2V. DGIndex can be used to cut M2Vs from a VOB source.

[Ultimist]

Thanks manono, for your suggestions about the format and upload... Fixed!!

On further inspection, this video seems to have a strange pattern of 2 progressive frames, followed by 2 interlaced, followed by 2 progressive again, then another 2 interlaced...

I'm wondering if that pattern can give someone a clue as to what's going on with this video. I don't know how to interpret these results since as I said, I don't have a full understanding of the whole interlaced/progressive/pulldown thing.

[manono]

Much better. I only looked at the first one. That opening part is progressive 29.97fps and I usually encode that sort of thing separately (for DVD) and rejoin it during the authoring in Muxman. I think you said you're planning on just cutting it out. The rest of it is standard hard telecine easily IVTC'd. Except for the dissolves which are a mess. I assume there's very little of that in the movie so I wouldn't do anything special with them and just put up with the few seconds of jerkiness. Maybe others have some better ideas.

[Ultimist]

The rest of it is standard hard telecine easily IVTC'd.

This confuses me, though. As said I have tried a gazillion methods to get it back to progressive and non-motion juddered... and I haven't had any luck. It always ends up either progressive-looking (on inspection of individual frames) but motion juddered when played, or interlaced-looking with smooth motion, no matter what I've tried. I've tried the standard tfm with tdecimate, animeivtc, qtgmc, various deinterlacing plugins in both avisynth and vdub.

What IVTC plugin or script (and what settings) are you using that makes it "easily IVTC'd" ?

the 2 prog/2 interlaced pattern I noticed also led me to look a bit closer. It's almost as if 2 frames are laid on top of each other, most of the time...

Also, how can I tell precisely where the 29.97 footage ends, and the 23.976 rate begins?

[ChiDragon]

Ewwww!

First off, what source filter are you using to bring the MPEG-2 file into Avisynth? With MPEG2Source, there is no pattern of 2 progressive/2 interlaced. It's plain old 3 progressive/2 interlaced.

Secondly, are there no better sources for these music videos? They're atrociously converted. Screenshot for those reading but who didn't download the samples:

http://s1.postimg.org/rw3xqbxmn/Ultimist_clip2.png

This is from one of the progressive frames. Straight lines aren't even straight and there is egregious edge enhancement. It looks like the fields are misaligned. I've seen similar-looking garbage but I've never known terms for all the artifacts, or the cause.

Anyway, it's normal for a brick wall like that to trigger interlace detection and cause TFM and others to interpolate instead of just field-matching as it should. If the fields are correctly matched, you can just turn off its post-processing option. Manually IVTCing by using DoubleWeave().Pulldown(0,2) works for the entirety of Clip2, but it will probably break at a different point in the music video because the telecine pattern does change at each dissolve.

[Ultimist]

Ewwww!

First off, what source filter are you using to bring the MPEG-2 file into Avisynth? With MPEG2Source, there is no pattern of 2 progressive/2 interlaced. It's plain old 3 progressive/2 interlaced.


I know exactly what you mean by "Ewwww!" - I've wanted to get this particular video into shape for a while now.

If I don't TIVTC I get 2 interlaced then 2 progressive (at least it appears that way) when the file is stepped through frame by frame in VirtualDub.

I'm using DGDecode_mpeg2source on a d2v generated by DGIndex with "Honor Pulldown" ticked. Would I get a different result using mpeg2source directly on the vob? Not sure what the differences are.

Secondly, are there no better sources for these music videos? They're atrociously converted.

Unfortunately, for this particular music video, this is the only DVD source I've ever found. It's Michael Jackson's "Cry" and it isn't on any of his other DVD's. This is from the "Michael Jackson's Vision" DVD set released in 2010 (or 11).

Honestly, at this point I would be willing to manually match each and every field, if such a tool exists where I could manually combine fields, check for artifacts, and individually "commit" my choices for each set of fields. It would be time consuming, but at this point I'm willing to do it if any such tool exists to manually compare and make changes to field alignment on a frame by frame basis. The video is only a few minutes long, so it wouldn't take forever. And it might actually be fun if the result is some sort of reasonably acceptable version of the video.

Thankfully, this is the only music video in the set that exhibits this particular problem. There are other problems with some of the others (this IS a horribly compiled set, clearly rushed), but I've worked all those out.

[Ultimist]

I found my "best attempt" of restoring this video. I did it a couple of years ago. It took a LONG time, over a week tinkering with animeivtc, qtgmc, some plugin for virtualdub that I found, called Smart Deinterlace, and a LOT of experimenting. Most of the interlacing artifacts are gone, with some small artifacts remaining in people's hair in certain scenes.. The whole video suffers from some kind of slight motion judder problem that I'd ultimately like to fix, if possible, but not at the expense of undoing the interlacing fixes.

Some people might look at this "fixed" sample and say to me "Looks like you don't need our help." but I really got lucky making this look as good as I did. I can't remember what settings I used or changed. Like I said, it is the result of a lot of tinkering and experimentation, and it's still not perfect.

I'm very aware that this might be the best I'll ever get using this source, and that I might just have to live with the remaining artifacts. If that is the case, I guess I can accept it until (hopefully) some better source comes along.

I'm an amateur. I think this shows that someone more skilled with the proper knowledge and tools *might* be able to get this video looking even better than this. If someone like that could pass such information on to me, I'd forever be in their debt.

Here it is. The brick wall scene has been mostly fixed, except in a later scene (not in the sample) where it has a bit of that moire/crosshatching pattern again, but it's less noticeable. Most of the other interlacing and field misalignment problems have also been solved, resulting in the juddery (slight) motion.

http://www.mediafire.com/watch/b13v6p4yi40ns5i/Fixed_Sample.avi

I look at this clip and say "wow." I can't believe I was actually able to get it this good... I had no clue what I was doing, I just kept messing with options until the clip looked as "right" as I could get it.

Oh, and I also used NeatVideo after all other steps, to clean up the noise, and some other filter I can't remember, to reduce (some of) the oversharpening present in the original source.

So, am I stuck with this (it's not that bad) or can I get even better results somehow?

Edit: If watching this sample in a window, the motion judder is almost unnoticeable most of the time. It is far more apparent when watching in full screen.

[johnmeyer]

This confuses me, though. As said I have tried a gazillion methods to get it back to progressive and non-motion juddered... and I haven't had any luck. It always ends up either progressive-looking (on inspection of individual frames) but motion juddered when played, or interlaced-looking with smooth motion, no matter what I've tried. Manono is exactly correct in his assessment: all you need for the portion of the video that was original 24p film is standard IVTC. I just used this off-the-shelf script, and it worked perfectly:loadPlugin("c:\Program Files\AviSynth 2.5\plugins\TIVTC.dll")
AVISource("E:\fs.avi").killaudio()
AssumeTFF()
tfm()
tdecimate()
As for the "motion judder when played," that is what happens when you pan a 24p camera horizontally. Many people think that judder happens because of pulldown fields, and indeed that does make the motion look worse, but the main reason for the judder is that 24 progressive frames a second actually is not enough to provide true "persistence of vision," and your eye is still retaining the previous frame while watching the current frame, and this makes the image appear to jump, even when that is not happening.

There is nothing wrong with the IVTC process.

If you don't like that effect, then one solution would be to convert everything to 30p, which is the cadence of the opening part of the clip. You can do this using MVTools2. All you need is one of the MFLowFPS scripts given in the MVTools2 documentation, using a reasonably large block size. 30p is still going to have some judder compared to 24p, but it will be better.

Here is your clip, with the opening unaltered, but the remaining clip first IVTC'd using the above script, and then converted from 24p to 30p, in order to match the opening cadence:

test_(30p).m2v (http://www.mediafire.com/watch/4skfr5vdkghubfv/test_(30p).m2v)

[ChiDragon]

Your "Fixed Sample.avi" has judder because every 5th frame is a duplicate. It's 29.97 and should have been decimated down to 23.976.

[Ultimist]

Your "Fixed Sample.avi" has judder because every 5th frame is a duplicate. It's 29.97 and should have been decimated down to 23.976.

Yes, I recall now that doing it this way was how I got such a clear result, for the most part. Decimating the frames before other processing only led to (seemingly) insurmountable interlacing, shimmering, and other problems. When I had it all finished and looking this good (not perfect) I must have forgotten that the frame rate remained at 29.97

So, in Avisynth, how would I go about removing... Edit: I figured it out. This newly decimated and proper 23.976 clip doesn't look half bad. Decimating those repeated frames definitely cleared up some of the motion judder. If it can't be gotten any better than this, it may be good enough until I can find a better source.

Thank you, ChiDragon, for pointing out what I should have seen even before uploading it. It's been a couple of years since I created the file from which I extracted the "Fixed Sample" so I forgot about it needing decimation. I do appreciate your comments, and they have already helped me a great deal.

[Ultimist]

Manono is exactly correct in his assessment: all you need for the portion of the video that was original 24p film is standard IVTC. I just used this off-the-shelf script, and it worked perfectly:loadPlugin("c:\Program Files\AviSynth 2.5\plugins\TIVTC.dll")
AVISource("E:\fs.avi").killaudio()
AssumeTFF()
tfm()
tdecimate()
As for the "motion judder when played," that is what happens when you pan a 24p camera horizontally. Many people think that judder happens because of pulldown fields, and indeed that does make the motion look worse, but the main reason for the judder is that 24 progressive frames a second actually is not enough to provide true "persistence of vision," and your eye is still retaining the previous frame while watching the current frame, and this makes the image appear to jump, even when that is not happening.

There is nothing wrong with the IVTC process.

If you don't like that effect, then one solution would be to convert everything to 30p, which is the cadence of the opening part of the clip. You can do this using MVTools2. All you need is one of the MFLowFPS scripts given in the MVTools2 documentation, using a reasonably large block size. 30p is still going to have some judder compared to 24p, but it will be better.

Here is your clip, with the opening unaltered, but the remaining clip first IVTC'd using the above script, and then converted from 24p to 30p, in order to match the opening cadence:

test_(30p).m2v (http://www.mediafire.com/watch/4skfr5vdkghubfv/test_(30p).m2v)

I'm seeing the presence of interlacing issues remaining in that file you uploaded (shimmering in the ocean as the camera pans down, some kind of field offset as the camera pans across the people in the final scene of this clip, leaving jaggies on their faces and bodies).

[johnmeyer]

I'm seeing the presence of interlacing issues remaining in that file you uploaded (shimmering in the ocean as the camera pans down, some kind of field offset as the camera pans across the people in the final scene of this clip, leaving jaggies on their faces and bodies).Yes, I saw those artifacts, but those are in the original clip and are not a function of the inverse telecine process nor do they result from the motion estimation interpolation that I used to increase the frame rate from 24p to 30p.

I was mostly focusing on solving your judder issue, and also illustrating how to correctly inverse telecine your clip. You will get absolutely nowhere until you do that, and do it correctly.

If that line twitter in the waves is a problem, you can apply a little filtering. If you search the forum, I think you can find some code that Didée posted several years ago that dealt with this.

[creaothceann]

I'm seeing the presence of interlacing issues remaining in that file

One filter you could try is QTGMC in progressive mode.

[Ultimist]

I think you can find some code that Didée posted several years ago that dealt with this.

What would I call this problem when searching for Didée's code?

[ChiDragon]

Vinverse seems to mask some of the problems, but I don't think that's what johnmeyer was suggesting.

QTGMC in progressive mode doesn't help. The individual fields are aliased; maybe that's why?

TIVTC and other auto-IVTC'ers can't produce 100% smooth results during the dissolves when the telecine pattern changes. The semi-automated IVTC filter that I wrote many years and never released allows you to mark these areas for full motion handling. Sometimes fields need to be deinterlaced instead of field-matched for this to work, so spatial resolution is lost. If you're interested in using this, every single dissolve range needs to be manually marked in a text file. When I was working on the filter, I was encoding TV series that might have a few dissolves per episode and not music videos that used nothing but dissolves for minutes on end.

Before anything else, you need to get the video into Avisynth with the normal 3/2 pattern we see and not this 2/2 alternation only you've been getting!

[johnmeyer]

The clip has obviously been modified in some way other than simply being telecined. If you do a separatefields() on the clip you supplied, and just look at individual fields, you will see evidence of some attempt at deinterlacing and/or resizing. Take a look at the "teeth" artifacts in the girl's hair:

http://i177.photobucket.com/albums/w208/johnmeyer/Artifact_zpsaab6ae4d.jpg

I am pretty sure that this was not in the original video.

As for search terms, you are looking for a solution to something like, "improper deinterlacing," or "bad deinterlacing." You might also find things using the word "conversion" along with some modifier like "bad" or "improper."

As always, if you can get further back in the video food chain, and find something closer to the original, you will have a lot better luck.

[Ultimist]

Thanks, johnmeyer, for your suggestions. I'll definitely be looking into it.


As always, if you can get further back in the video food chain, and find something closer to the original, you will have a lot better luck.

Yes, I wish I could find a better source. However, according to my research, this is the one and only time this particular music video has ever appeared on DVD. As for other high quality sources, I've come up empty so far, but I'll keep looking.

[Ultimist]

johnmeyer,

Not sure if this was the code you meant, but I found this bit, posted by Didee in 2010:

LoadPlugin("C:\Program Files\AviSynth 2.5\plugins\nnedi2.dll")
LoadPlugin("C:\Program Files\AviSynth 2.5\plugins\mt_masktools-25.dll")
# LoadPlugin("C:\Program Files\AviSynth 2.5\plugins\mt_masktools-26.dll") # "-25" is for Avisynth 2.5.x, "-26" for Avisynth 2.6.
LoadPlugin("C:\Program Files\AviSynth 2.5\plugins\RemoveGrain.dll")
LoadPlugin("C:\Program Files\AviSynth 2.5\plugins\Repair.dll")
Mpeg2Source("L:\Videos L\my video.d2v")
ConvertToYV12()
o = last
NNEDI2(field=-2)
dbl = Merge(SelectEven(),SelectOdd())
dblD = mt_MakeDiff(o,dbl,U=3,V=3)
shrpD = mt_MakeDiff(dbl,dbl.RemoveGrain(11),U=3,V=3)
DD = shrpD.Repair(dblD,13)
dbl.mt_AddDiff(DD,U=3,V=3)

I modified it slightly to fit my source and include the TIVTC'ing (plugin paths removed because they reveal personal info. If you wish to try it on the sample clips, you'll have to enter your own plugin paths):

LoadPlugin("nnedi2.dll")
LoadPlugin("mt_masktools-25.dll")
LoadPlugin("RemoveGrain.dll")
LoadPlugin("Repair.dll")
LoadPlugin("DGDecode.dll")
DGDecode_mpeg2source("F:\MJVobs2\Film\Cry (MJ Vision) Intro Clipped.d2v", info=3)
LoadPlugin("ColorMatrix.dll")
ColorMatrix(hints=true, interlaced=true, threads=0)
LoadPlugin("TIVTC.dll")
tfm(order=0).tdecimate(hybrid=1)
ConvertToYV12()
o = last
NNEDI2(field=-2)
dbl = Merge(SelectEven(),SelectOdd())
dblD = mt_MakeDiff(o,dbl,U=3,V=3)
shrpD = mt_MakeDiff(dbl,dbl.RemoveGrain(11),U=3,V=3)
DD = shrpD.Repair(dblD,13)
dbl.mt_AddDiff(DD,U=3,V=3)

This does seem to solve most of the ocean shimmering, and softens the jaggies in people's hair... but introduces other problems (or leaves them untouched) later in the video. The brick wall is still a mess with this code, and various scenes have an absolutely intolerable amount of shimmering, jaggies, etc. The cross-dissolves between scenes seem even more messed up with this code.

BUT, like I said, it fixes certain things.

I can keep the fixed sections and lablel them as "solved". Then, if I can use some other code snippets to repair the problems in the remaining footage, I can just combine the best result for each scene into a reconstructed representation of the original.

I don't mind mixing and matching clips, if there is no all-in-one fix for the entire video. If I can find solutions for each scene individually, I'm more than willing to chop up the video and use the best fixes for each individual one.

For example, if there's a bit of code that fixes JUST the cross dissolves between scenes... I can extract the corrected dissolves as clips and keep them... and if there's something that fixes JUST the brick wall problem, I can cut out that fixed portion and use it in the final version.

I do have a couple of questions:

1. Does DGDecode_mpeg2source() produce precisely (pixel for pixel) the same output as mpeg2source() if used with the same d2v?

2. The reason the hybrid setting is in with the decimation, is that is what was recommended in MeGUI's "AVS Script Creator" when pressing the "Analyze" button - I use this because I don't know another way of analyzing the video and setting up proper TIVTC/decoming/deinterlacing. However, if that hybrid setting is causing problems, I can remove it. I guess I've never been certain that MeGUI's "Analyze" always produces the correct settings.

3. Is there any sort of manual deinterlacing utility, where I can view frames side by side, combined, shift one field or the other forward or back in time, up and down, etc... to manually deinterlace? As I said, this clip is only a few minutes long, so manually doing it frame by frame would be a bit time consuming, but far from impossible.

I'd do it if I had such a tool that can let me adjust each field/frame one by one. I'm always wiling to do the work to get the result I want, if I have the proper tools. All these auto-deinterlacers/etc. are great, but sometimes they can't pick up things that are plainly visible to the human eye, especially for someone like me with a very picky eye for detail.

[videoh]

DGDecode_mpeg2source() and mpeg2source() are just different names for the same thing.

Look into yatta-ivtc for manual IVTC.

[johnmeyer]

I have looked at several different versions of this video that are posted on YouTube and Vimeo. They all exhibit the problems in your DVD version.

Cry Video (http://www.youtube.com/watch?v=3teLshzC9P0)

It is amazing to me that the DVD would actually be this bad. I wonder if the source of this video is actually the original DVD, or if it came from a download from one of these online sources.

[ChiDragon]

Here is a short clip showing the semi-automated plugin I wrote vs TFM().TDecimate(), on one of the dissolves where the pattern changes in a way that causes TIVTC to lose temporal smoothness. Both of them have deinterlacing artifacts.

Ultimist Clip1, Dissolve 2 - SCIVTC vs TIVTC.avi on Mega.co.nz (1.4 MB) (https://mega.co.nz/#!wcEBCZKQ!oQpzumE6KMPKSxw4BHkiHFFmqQY-HjD6UvQkEX8P2vw)

As I said, I never released it. There's a bug that can lead to a crash which I never tracked down, and I was unable to get my head around another design flaw that causes certain combinations of patterns to fail. If you want to try it, I can PM you a link.

But come to think of it, is there a particular reason you want to IVTC the video? Simply running QTGMC in its regular mode produces a 60fps video with just the regular 3:2 judder we see all the time. If you don't mind its smoothed-over look, it clears away the little spatial artifacts and the frame rate is fast enough that its motion compensation errors can't be seen in playback. The brick wall itself is the only part of the brick wall frame that doesn't look better.

http://s1.postimg.org/rw3xqbxmn/Ultimist_clip2.pnghttp://s22.postimg.org/b2wh4wty9/Ultimist_clip2_QTGMC.png

[johnmeyer]

The problem is that after the IVTC, you find that the original fields have all sorts of problems, suggesting that some processing was done to the original video (improper deinterlacing or re-sizing) prior to it finding its way onto the OP's DVD.

[Ultimist]

It is amazing to me that the DVD would actually be this bad. I wonder if the source of this video is actually the original DVD, or if it came from a download from one of these online sources.

In my particular case it is ripped directly from the "Michael Jackson's Vision" DVD set that I own. I can't speak to those other versions you found. I, too, am surprised at the quality (or lack thereof) of the encoding. As I mentioned, there are other problems exhibited in other videos from that DVD set, but the other problems (overly noisy video, despite it being claimed that all of the videos have been 'restored," etc.) are more easily solved.

I've looked around and found some versions on youtube, including the one released on the official Michael Jackson Vevo channel, and it has the same problems...

[Ultimist]

Here is a short clip showing the semi-automated plugin I wrote <snip>

If you want to try it, I can PM you a link.


Yes, please do PM me a link. I would appreciate that more than I can say.


But come to think of it, is there a particular reason you want to IVTC the video?

No particular reason other than wanting to get the video back to its 23.976fps original state (without all the artifacts) so that I can clean it up and have a nice looking, smooth playing copy without all the issues.

[Ultimist]

Beyond the overly ambitious sharpening that has clearly been applied, there's also some weird kind of ghosting apparent in certain scenes, and some strange "ghosty" lines at the top and bottom of the frame (on the actual footage, not the black borders)... I'm not sure if that ghosting can give us any kind of clue as to what else was done to this video, or not.

The ghosting becomes more apparent behind the people in the first long shot of people holding hands atop a ledge at the beach, especially after using dehalo to get rid of the sharpening artifacts.

I haven't removed the black borders because I want to take care of the other problems, first.

It's hard telling what exactly was done to the clip by the studio before its release, but it's definitely a crappy transfer job. Produced by the multi-billion dollar company, Sony, no less... This set was definitely rushed out to the public to capitalize on Michael's death. As I mentioned earlier, it was advertised as having fully restored, uncut versions of all Michael Jackson's short films, videos, etc... but none of them have actually been restored and have artifacts from poor noise reduction techniques, many of them are cut or altered in some way, and two (this one, "Cry" and "You Rock My World") have strange interlacing issues. The issues are different for "You Rock My World" and are more easily solved.

I don't know if someone very knowledgeable could look at this video and be able to tell exactly what has been done to it. It seems to me that knowing what was done might help me "undo" those things, perhaps. The first step in solving a problem is identifying what the precise problem is.

[Ultimist]

DGDecode_mpeg2source() and mpeg2source() are just different names for the same thing.

Look into yatta-ivtc for manual IVTC.

Thank you for this info. I've found Yatta and got it set up. It has a bit of a learning curve but this may be what I have been looking for.

I can't seem to find any comprehensive tutorials on yatta, though, so I'm struggling to figure out how to use it.

[videoh]

Another approach is to use a manual override file with Decomb or TIVTC.

[feisty2]

did you ever try "tfm (pp=0)" tfm always misdetects some perfect progressive frames as interlaced frames and post processes them then they look like shit

[cretindesalpes]

Field misalignment can be fixed with this code:
SeparateFields ()
Spline36Resize (Width (), 174)
Weave ()

TFM (pp=0)
TDecimate ()

You still have these ugly sharpening halos. I think they may be attenuated using convolution + masking.

Scene crossfades exhibit double-hard telecine. There are more or less sophisticated methods to deal with it. Ideally you check every pattern change and do the field matching and decimation manually to get clean pictures. There are only two cases of shift: ±1 or ±2 fields. Long time ago I came with some kind of formalization of these cases:
# Type 1: 2-field shift, use guided field matching
# ref: AA AB BC CC DD
# 2nd: EE FF FG GH HH
# do : NP <N PN P- NP
# dup: **
# tdc: + - + + +
# Type 2: 1-field shift, must deinterlace one every two frames
# ref: AA AB BC CC DD
# 2nd: EF FF GG HH HI
# do : >N P- D- NP >D
# dup: **
# tdc: + - + + +
# Legend:
# D : deinterlace field
# <> : use the result of the previous or next field
# NP : match fields (with next or previous one)
# - : don't use

So you have to identify all the pattern change and parameter your IVTC code appropriately. I have functions to do this but they are stuck in a complicated framework and it would be less hassle to rewrite them.

[ChiDragon]

Field misalignment can be fixed with this code:
That is fantastic! How did you come up with the right number?

Here it is resized up (and down) from 720x348 using nnedi3_rpow2(2, cshift="Spline36Resize", fwidth=720, fheight=480). Compare to the original, on the left.

http://s1.postimg.org/rw3xqbxmn/Ultimist_clip2.pnghttp://s27.postimg.org/ihibektv7/Ultimist_clip2_cretindesalpes_nnedi3.png

So you have to identify all the pattern change and parameter your IVTC code appropriately.
My plugin does a lot of this work... when it works. I wonder if you could help me fix the bugs. You seem extremely well-versed with interlacing issues.

[cretindesalpes]

The lines “less oblique than the actual lines” are generally caused by fields individually resized without taking their vertical positions into account. Fortunately the process is more or less non-destructive in the upscale case and can be reverted by resizing the fields to their original size. Finding the exact size is a matter of trial and error. When you get close, it shows some beating patterns that become larger and larger until they disappear completely when you hit the exact size.

[Ultimist]

cretindesalpes,

I apologize for my delayed reply. I've been busy with other things.

I have to say, your solution for the field misalignment issues is absolutely incredible!

Bar none, AVIsynth has to be the most powerful video processing software on the planet. I've seen other even more severe problems simply melt away with the proper script, and your short, simple script is another example of that sheer power. Thank you!

ChiDragon and everyone else, as well.... Thank you, too! Your input and advice has been so helpful. The knowledge and expertise so generously offered in these forums is simply phenomenal, and much appreciated.

With the incredibly hair-pulling field issues resolved, now to tackle the mismatches during the dissolves. I have a feeling ChiDragon's plugin will be very helpful in that regard once I get it running.

I am so pleased I inquired here first, for help. You people are amazing!