IVTC Not Working?

[spudboy]

I recently finished backing up the NTSC "Crouching Tiger, Hidden Dragon" with DVD2SVCD. The backup completed without errors, but I noticed something strange when I examined the resulting mpeg files. If I load the file in a viewer and step through it frame-by-frame, I see frame after frame that contains interlacing artifacts. I had chosen the IVTC option in DVD2SVCD, so I assumed these artifacts would be eliminated. Does anyone know how to do a backup with DVD2SVCD that does not produce interlacing artifacts?

- spudboy

[daxab]

This is a big topic but I'll take a stab at it.

The first thing to do, before doing IVTC is to verify that your source is telecined. An easy way to do this is to load up the DVD2AVI project file and hit Preview. Look at the FILM/PROGRESSIVE/INTERLATED indicator. If it is FILM or FILM xx%, where xx is above 90 or so, you have telecined material (film stuff at 23.976 fps that has been manipulated so as to produce 29.97 fps -- but it's still 23.976 fps of actual unique frames).

If you have film source, your best bet is actually just to select force film. This is preferable to any form of IVTC. If you don't have film source, or if for some other reason you want to IVTC (e.g., hybrid source, like anime), then you need to feed all the frames (i.e., don't force film) from DVD2AVI and IVTC them afterwards.

No IVTC method is perfect. The IVTC Avisynth plugin that the DVD2SVCD program uses is no exception. It works very well, but it can be fooled. So whenever you do IVTC you may see interlacing.

One option is to tweak the 3 parameters to the IVTC command. This is a little involved; if you want to try this you should check out the IVTC plugin threads on the Avisynth board.

Personally I prefer GreedyHMA. This is also not perfect but I've had excellent results. (You can also use TMPGEnc to do IVTC but that's a little involved -- probably more work than you want to do.) If you want to read up on GreedyHMA there is a long and informative thread on it in the Avisynth forum.

If you want to switch from the standard IVTC plugin to GreedyHMA, it's easy. Just click the button to edit your .avs file. When the edit window comes up, delete the IVTC command and add the GreedyHMA command. You will also of course have to add a command to load the GreedyHMA plugin. This can be read about further on the Avisynth board, of if you want some examples I can be more specific -- but you probably get the idea by now.

My $0.02.

[dvd2svcd]

In the next build there will be support for Donald Grafts new Decomb method. I have tested it on Sakura Diaries and it works very. You could try that when I release it.

[spudboy]

daxab and dvd2svcd -

Thanks for taking the time to reply!

The NTSC "Crouching Tiger, Hidden Dragon" is hybrid in that the opening credits are completely interlaced while the actual movie is 23.976 fps progressive through-and-through.

My old way of backing up this type of movie was to save a DVD2AVI project with no ivtc or forced film, load in TMPGEnc and run ivtc, save the project and convert to .avi with VFAPI, then encode in CCE with chapter points. I always examine the frames chosen by TMPGEnc ivtc and I must say that I have never, ever seen it make a mistake, although theoretically I know it could.

Just so I have it straight, could you please clarify again which options you would choose when processing this type of movie? If I choose forced film in the DVD2AVI tab, wouldn't that make the opening credits look funky?

- spudboy

[daxab]

If I choose forced film in the DVD2AVI tab, wouldn't that make the opening credits look funky?I think that's correct. GreedyHMA will deinterlace such frames if you use mode 4, the fully automatic decimation mode, but I think you might still get some jumpiness, if the material is really 29.97 fps -- you'd have to try it and see.

Haven't heard of Decomb -- looking forward to trying it!

[sandchar99]

Recently backed up Crouching Tiger, Hidden Dragon (NTSC, Region 1). DVD2SVCD, DVD2AVI, etc did not see it as telecined. ForceFilm worked and final MPEG2 and SVCD discs look great.

[daxab]

I just did a side-by-side test of GreedyHMA vs. Decomb on some typical, pretty easy non-anime footage. At the first "sign of trouble" -- a typical scene with almost no motion except for an actor's mouth, during dialog, something that trips up a lot of IVTC algorithms -- GreedyHMA reconstructed all progressive frames correctly (this is without any GreedyHMA filtering). Decomb flubbed it and let through a string of interlaced frames.

I'm going to stick with GreedyHMA for now, although maybe Decomb will improve in future versions.

[dvd2svcd]

What did your Decomb AVS file look like. Would you mind posting it here?

[daxab]

Sure. Here's the whole thing, showing both the Decomb and the GreedyHMA function calls:LoadPlugin("c:\app\avisynth\mpeg2dec-d.dll")
LoadPlugin("c:\app\avisynth\greedyhma.dll")
LoadPlugin("c:\app\avisynth\decomb.dll")
SegmentedAVISource("cap_33.avi")
AssumeFrameBased.ComplementParity
Letterbox(56,61)
a=GreedyHMA(1,0,4,0,0,0,0,0)
b=Telecide.Decimate15
StackHorizontal(a,b)I open the AVS in VirtualDub and drag the thumb around the timeline, looking for artifacts on either the left or the right.

I've spent more time looking at this, and now I've found some scenes where Decomb does the right thing, and GreedyHMA gets confused. So each one has its strengths and weaknesses.

I can't see any definite pattern yet, except for the fact that GreedyHMA does better with talking scenes (no movement except lips) -- Decomb 1.5 messes these up a lot.

[dvd2svcd]

No wonder you got bad results with Decomb.


LoadPlugin("c:\app\avisynth\mpeg2dec-d.dll")
LoadPlugin("c:\app\avisynth\greedyhma.dll")
LoadPlugin("c:\app\avisynth\decomb.dll")
SegmentedAVISource("cap_33.avi")
AssumeFrameBased.ComplementParity
Letterbox(56,61)
a=GreedyHMA(1,0,4,0,0,0,0,0)
b=Telecide.Decimate15.FieldDeinterlace
StackHorizontal(a,b)


FieldDeinterlace is recommended by Donald Graft, and it works well.

http://sauron.mordor.net/dgraft/decomb.html

[daxab]

I did try it with FieldDeinterlace, with the same result. That was actually the first thing I tried, since that's what the docs mention:a=GreedyHMA(1,0,4,0,1,0,0,0)
b=Telecide.Decimate15.FieldDeinterlaceThis gives both filters a chance to 'clean up' stray interlacing. I looked at it with MotionMask and the interlaced detection algorithm wasn't picking up the interlaced area.

The interlaced area is very small, just the edge of someone's face when the person is talking and their lips are moving. I'm surprised GreedyHMA can handle it; I assume it's succeeding because it's following an established telecine pattern (whereas Decomb is pattern-less).

From my limited experience so far, I'd say this:

- GreedyHMA is very good, in general, but can get very confused sometimes and allow large areas of interlaced video to get through. Seems to happen especially on dark scenes, or fade-in / fade-out sections. It's surprisingly good with "dialog" scenes.

- Decomb is very good, in general, and doesn't ever seem to let larget areas of interlaced video through. On the other hand, in "dialog" scenes, where the interlaced areas are tiny, it can get lost, since it doesn't use predictive patterns.

[spudboy]

Here is another question:

After experimenting with Crouching Tiger, Hidden Dragon, I found that the pulldown pattern was a simple "1010010100", etc. After reading the Avisynth reference more carefully, I noticed the Pulldown function. Would it work if I chose "IVTC" on the DVD2AVI tab and "edit avisynth script after DVD2AVI", and then replaced the IVTC function call with:

DoubleWeave
Pulldown(0,2)

Sincerely,
spudboy

[daxab]

If that is indeed the pulldown pattern, and if you have the syntax right, then that should work. Make sure you select the IVTC button in the DVD2AVI panel, so the program knows you are doing an IVTC in the Avisynth script (and performs a pulldown, and knows what frame rate to expect).

Decomb follow up: neuron2 has been working on the Decomb filter (see the Avisynth board if interested) and the latest version, 1.7, performs roughly on part with GreedyHMA, perhaps better overall.