Regarding AutoGK and IVTC - behind the scene...

[kikker]

So, I've read up on many telecine FAQs, DGIndex, Avisynth scripts, Decomb and its components, etc.

What I'd like to know is what precisely does AutoGK do with its settings. If AutoGK deems a source to be >95% FILM, is the "Force Film" feature of DGIndex then enabled?

If the source is found to be hybrid, and neither IVTC nor deinterlace are checked in the advanced options, what script settings are generated? I would assume that the output would remain 30fps and that fields could be blended if similar enough.

If source is found to be hybrid and "Force normal IVTC" is checked in the advanced settings for AutoGK, is that the SAME as Force Film in DGIndex, or does that result in some use of the DECOMB filter for Avisynth?

If the source is found to be hybrid and the "Force deinterlace" is checked in the advanced settings, does that result in a "FieldDeinterlace" application?

Lastly, if the source is found to be NTSC/Video, then is any sort of "FieldDeinterlace" applied automatically?

Thanks for any answers!

[manono]

Hi-

What I'd like to know is what precisely does AutoGK do with its settings. If AutoGK deems a source to be >95% FILM, is the "Force Film" feature of DGIndex then enabled?

A lot of the answers can be gotten from an examination of the log and AviSynth scripts. This first question I'm not entirely sure of the answer to. That is, I'm not sure at what point the percentage (which has nothing to do with the film percentage in DGIndex, but from the AutoGK analysis) determines that a Force Film is OK, or a full IVTC is necessary. Either way, above 95% it'll be restored to 23.976fps. So it shouldn't worry you much. In fact, I'm not sure of the dividing line percentages between any of the categories, but your other questions don't really ask about them.

If the source is found to be hybrid, and neither IVTC nor deinterlace are checked in the advanced options, what script settings are generated?

It depends if it's Hybrid, mostly film or Hybrid, mostly video. In the mostly film case, it restores it to 23.976, but uses Decimate's Mode 3 to blend 2 frames of the video portions, to lessen the impact of the 30fps video portions (look it up in the DecombReferenceManual for more info). It uses Decimate's Mode 1 if it's mostly video, and adds a blended frame during the film portions to give a smoother playing 29.97fps.

If source is found to be hybrid and "Force normal IVTC" is checked in the advanced settings for AutoGK, is that the SAME as Force Film in DGIndex

Not at all. A true IVTC is much different than a Force Film. Using Force Film presupposes the existence of pulldown flags. If it's a hybrid, there either aren't any such flags at all, or only for a portion of the video. Force Filming such a video will leave you with an interlaced jerky playing mess.

If the source is found to be hybrid and the "Force deinterlace" is checked in the advanced settings, does that result in a "FieldDeinterlace" application?

No, it deinterlaces with the much better KernelDeint (or LeakKernelDeint, I forget). FieldDeinterlace is used only within Telecide, and not as a separate deinterlacer.

Lastly, if the source is found to be NTSC/Video, then is any sort of "FieldDeinterlace" applied automatically?

No, again, it uses KernelDeint for any deinterlacing chores. But it will be deinterlaced automatically.

[kikker]

Ah, OK thanks for the answers! Just to clear some fuziness, Force Film relies on proper flags. If the flags are improper, you will get bad results. And, IVTC, or the use of the DECOMB filter, is a more intensive process of field matching. BTW, I don't see any evidence of a Mode 3 in Decimate (only Modes 0-1).

Another curiosity about AutoGK and its analysis: It assigns a percentage to 4 variables.

1. % interlacing in motion areas
2. % telecined patterns
3. % progressive patterns
4. % interlaced patterns

In true telecined material, is it not correct to say that (1) should be about 40% due to the "3-2" pattern of progressive-interlaced frames?

Does (2) refer to this "3-2" pattern of progressive-interlaced frames?

As for (3) and (4), could you clarify what those mean in terms of patterns? Does it have something to do with the "0123401234012340" as a progressive pattern and "012340122222223401222222222" as an interlaced pattern?

[manono]

If the flags are improper, you will get bad results.

Which isn't to say it can't be IVTC'd. It just can't be Force Filmed.

BTW, I don't see any evidence of a Mode 3 in Decimate (only Modes 0-1).

Look harder.

If mode=3, instead of discarding the most similar frame, Decimate() will apply different decimation strategies for 3:2 pulldown material (film) and for pure video (nonfilm) cycles. Film cycles are decimated in the normal way. Nonfilm cycles are decimated by applying a frame blending decimation of the cycle from 5 frames to 4 frames. The resulting frame rate is the film rate (decimated by 1/5).

In true telecined material, is it not correct to say that (1) should be about 40% due to the "3-2" pattern of progressive-interlaced frames?

Yes, but static/motionless frames won't show interlacing, so it has to allow some leeway for that.

Does (2) refer to this "3-2" pattern of progressive-interlaced frames?

Yes.

As for (3) and (4), could you clarify what those mean in terms of patterns?

I mentioned before that it makes no use at all of DGIndex for this. I could be wrong about that, though. I've seen some said to be pure FILM that perhaps are checking the FILM percentage at the bottom of every D2V. If it finds 100% FILM, then it can Force Film and dispense with the regular analysis. So I guess I'll have to amend my earlier answer. That also means your greater than 95% example from your earlier post would undergo the full analysis and not be Force Filmed, unless it's 100% FILM. And it would be IVTC'd. It does its own analysis of a percentage of the frames. I expect, but am not entirely sure, that 3 is also checking a 5 frame pattern, looking for a 30fps progressive source. It's rare, but it happens. Also, some videos will have a percentage of 30fps progressive frames among the film or video. I expect 4 inspects a similar 5 frame pattern for interlacing. I'm not really sure of the reason. To help determine better if it has to be deinterlaced, I guess. len0x (the developer) was always kind of secretive about exactly how this analyzing worked. His was the first to do it, and AutoGK is still far and away the best at it, and he didn't want to make it easy for people developing their own competing programs to figure it out.

I'm sorry if I'm a bit foggy on the details now. len0x consulted me regularly about these kinds of things when he was developing the program, but it's been several years now. In addition, he never did tell anyone the exact workings of the analysis. Some of what I've written is just educated guesses.

[kikker]

OK the loose ends are coming together! Thanks

Now, as far as DGIndex is concerned when analyzing a VOB:

When the frame type displays progressive, a series of intact frames are being previewed, and when the frame type displays interlaced, a series of pre-split frames (fields) are now being shown in the preview window, correct? And, the frame type is not influenced by the presence or absence of pull-down flags, right?

On the other hand, the Video Type is determined by both the flags (Yes=Film, No=NTSC) and the fps rate (24=Film, 30=NTSC), correct?


I assume the following chart holds some truths to guide me and others following this thread?

Trying to come up with examples of the 4 combinations then:

1. FILM + Progressive = Most modern Hollywood-type movies on DVD - can be IVTC'd.

2. FILM + Interlaced = Hard-encoded Interlacing - can be IVTC'd

3. NTSC + Progressive = Special effects in movies, some broadcast TV - could be IVTC'd as part of movie but shouldn't.

4. NTSC + Interlaced = Many TV series and TV movies (esp lower budget ones) - not to be IVTC'd.

[manono]

When the frame type displays progressive, a series of intact frames are being previewed, and when the frame type displays interlaced, a series of pre-split frames (fields) are now being shown in the preview window, correct?

It's always showing full frames. 720x480 usually, for NTSC. Have you ever seen any 720x240 fields being shown in DGIndex?

And, the frame type is not influenced by the presence or absence of pull-down flags, right?

It's influenced by the presence or absence of the progressive frame flag. If encoded as Progressive 23.976fps, it'll show as Progressive. Sometimes it can be Progressive 29.97fps. I just saw one recently. PAL can be Progressive, but usually isn't. Anything else, PAL (usually, but not always), true interlace shot using interlaced 30fps video cameras, or hard telecined film is, shown as Interlaced. And that flag is sometimes, but not often, thank goodness, set incorrectly.

On the other hand, the Video Type is determined by both the flags (Yes=Film, No=NTSC) and the fps rate (24=Film, 30=NTSC), correct?

Only something encoded as Progressive 23.976 with the pulldown flags set will show as Film. Everything else will be NTSC (or PAL). So, yes.

1. They don't have to be IVTC'd usually. They can be Forced Film. You can IVTC if you like, but it'll slow the encoding, and the result won't be as good as if you had Forced Film on something that was 100% Film to begin with.
2. No such thing. Hard telecine shows as NTSC Interlaced.
3. No IVTC. You keep it as 29.97fps progressive. IVTC will make it play jerky, as you're usually removing unique frames. UNLESS it was made 29.97fps progressive by repeating every 4th frame, as was true in the example I cited a minute ago. For that, you don't IVTC, but just remove the duplicate frames. That stuff like CG effects is usually, but not always, encoded as interlaced, even if it's from a progressive source.
4. Either true interlace (every frame is interlaced), or hard telecine (2 of 5 frames interlaced, the other 3 progressive). Hard telecine can be IVTC'd.

Read:

http://forum.doom9.org/showthread.php?t=87809
http://www.doom9.org/ivtc-tut.htm

The second link is outdated about the methods to use, but the core information is still good.