Good Inverse Telecine Program?

[LanceSteel]

I've tried using Virtualdub's Inv telecine (Reconstruct from Fields-Adaptive) but still get a large amount of interlacing throughout a captured stream. Are there any other programs that are good at inverse telecining or are there any guides (here or at other sites) that talk about how to de-telecine a vid cap?

[MITSUI_1]

Use avisynth with decomb.dll

There is no substitute!

Easy way to get it all installed: Go to doom9.org and download the DVD2SVCD "bundle" and install. You'll get a ton of the tools you need anyway, and it'll install the best version of AVISYNTH for you.

Then all ya need to do is get a script written.

[stanwebber]

aviutil's auto-ivtc algorithm is my favorite, although it's one of the slowest. i use tmpgenc to manually ivtc simple sources. decomb is good for hybrid sources where portions need to be frame-decimated to 23.976fps. virtualdub's algorithm is one of the fastest & it's not all that bad. you need to follow up with a specialized area-based deinterlace filter(gunner thalin, donald graft) after any ivtc operation to clean up misjudged frames while leaving good frames relatively untouched

[Nelson37]

I use InverseTelecine.dll for AVIsynth, have tried Decomb.dll, others. I am getting varying degrees of an occassional stutter, which I used to think were framedrops. Capture software says I am dropping 0 frames.

The visual effect is a short jerk in playback, very much like a framedrop though now I am assuming a frame is being decimated out of place. On movies where the effect is worst, I notice the Telecining pattern occassionally varies from P,P,P,I,I to P,P,I,P,I - usually only for one or two frames, several times in 5 minutes or less.

Any suggestions, or is this pattern change (assuming that is the cause of my problem) just something that I gotta live with?

[LanceSteel]

Nelson37, I get the same jumpiness if I decimate after doing Telecide with the decomb filter in avisynth. If I only Telecide with no decimate, I get a repeating frame every 5th frame but I don't get the jumpiness any more. The stream seems fairly fluid too (I don't notice the extra frame.) Does anyone know how to do a good, smart decimate?

[Nelson37]

Lance - I think the problem is in the decimate function, sure leaving them in looks OK but then you don't get the benefit of the 20% frame reduction.
I am in the process of running some tests, just got the latest TMPGenc. Been using CCE w/Avisynth and the IVTC dll. Most movies get 1 jerk every 5 minutes or so on average, sometimes only 2 or 3 in 90 min.
As I recall, using TMPGenc IVTC and auto setting (letting it run before encode) gave the best results, am gonna test Chicken Run, just captured off Digital Cable, jerks horribly after CCE-AVIsynth-IVTC.Dll-pulldown.

What DVD player do you have - I am noticing the jerk effect is less pronounced on PC - I have APEX 500?

[LanceSteel]

Nelson37, I tried Tmpeg's IVTC once before (not recently) but I remember that a few bits of interlacing made it through. If I remember correctly when using avisynth and decomb.dll, I tried the Inverse Telecine with Pattern Guidance and decimated every 5 frames. Telecide(guide=1)
Decimate(cycle=5)
If there is a shift in the telecine pattern then decimate with start dropping motion frames and will no longer be dropping the repeating frame (I'm realizing this as I am typing.) I'll try this when I get home but if you do a Telecide(guide=1) and then Decimate(mode=1, threshold=50). You still don't gain the 30 -> 24 fps conversion but it looks like you will no longer have a repeating frame every 5th frame (it gets blended with the frame after it.) I still have a lot of other settings to try. Let me know how Chicken Run turns out, you can email me too.
I have an Apex ad-500w and notice the motion jerking on both the computer and the dvd player so I don't think its your player.

[Nelson37]

Lance - looks like you tried the same decomb settings I did. What I don't understand, is if Decimate is cutting 1 frame in 5, or even pretty close to 1 in 5, why won't the framerate change? I want that goddamn frame reduction! AARGH!
BTW, I F'd up chicken run and I gotta do it over, after that and some more test I may start a new post on this IVTC thing, I'll let ya know.

[LanceSteel]

When I did the decimate=5 it did reduce the frame rate down to 23.947 (or something like that.) When I looked under the framerate setting in virtualdub (after loading the *.avs file) the framerate should read 23.9-something (if your source was 29.947.) I'm encoding to dvd format so I used the 3:2 pulldown in Tmpeg to bring it back to 29.947. Then I figured why decimate? Anyway the frame rate should change.

[MITSUI_1]

Lance - I think the problem is in the decimate function, sure leaving them in looks OK but then you don't get the benefit of the 20% frame reduction.

Actually, if you're converting to MPEG you'll get some fairly significant benefit due to the fact that the odds are that the "dupe" frames will wind up as B or P frames in the encode.

Lookisee:

• ***##***##
  1234567890
  IBBPBBPBBP

If the "#" characters are frames that are duplicates of the adjacent "#" character, you can see that they will take almost no space in the MPEG stream.

If you want to do a comparison, take a copy that your Telecide but leave as 29.97 fps, and encode it with TMPGEnc using CQ mode at say 85 Q level.

Then take that same clip and do not Telecide it, using instead one of the field blending deinterlacers or adaptive deinterlacers. Encode the same file usint TMPGenc using CQ mode at the same 85 Q number.

Notice the significantly larger file size in the second instance.

Now, just for shits and giggles, apply both a telecide and a decimate to get rid of the duplicate frames (thereby going down to 23.96 fps) and again encode with TMPGenc (but at 23.96 fps) using CQ mode and an 85 Q number.

You should find that the last file is not much smaller than the first.

Conclusion: Even leaving the "dupe" frames in there results in considerable file size savings, athough at the expense of smooth movement.

[/list][/quote]

[Nelson37]

Lance = I lost the framerate change when I used Decimate(mode=1), and (cycle=5) seemed to delete the wrong frame to varying degrees.

Mitusi - B,P frame benefit minimized cause I encode IBPB as I think it looks better, actually more fluid in motion which NOT running IVTC aggravates.

What I am really seeking is "Can IVTC be perfect?", particularly in cases where the 3,2 pattern changes to 2,1,1,1 and back again?

[vhelp]

late night all.

what I'd like to know is...
* Is there any real benefit or reason whey they do this, as in the above
post, ie, 3,2 patter andn then it's changed to 2,1,1,1 or whatever
else, as they always do... screwing us in the end, he, he...

If anyone has the answer, ...would love ta hear it!!
But, for the most part, all my encodes (23fps) usually come out w/out
any jitters/studders/hicups, etc, though I get some from time, to time.
...either I 23fps them or I leave them alone or I find a way to make
the clip smooth. About the only time it's not real smooth, is when my
miniDV tape has hicups or bad areas on the tape, or just poor firewire
transfering (usually due to too many apps open at once, or lost resources
as I do use up a lot in my endeavers, etc)

But, there are many ways ot ivtc'ing or 23fps'ing a clip, pending on
your source (and it's quality, should it be worth it?) some ie's,
* vdub
* avysynth, and it's many various forms for telecine, and 32pulldown,etc
* other standalone apps
* and a combination of all the above

I have around 9 profiles (23fps) methods so far. They all may not be
all that accurate 23fps w/ 32pulldown's, but they do work great and I
have (IMO) I feal they are the best quality encodes i'm doing to date.
And, yes, I've use just about every method out there (above) but still,
there are tough ones that just wont budge (again, above)

-vhelp

[LanceSteel]

Just an update, MITSUI_1 is right in saying that the size of the file doesn't change when decimating a repeated frame vs. leaving it in there. I compared the decomb.dll settings in one of my earlier posts
1) Telecide() only, with no decimation. I see a slight amount of stutter in the playback but only if I really look.
2) Then I tried the Telecide() w/ Decimate(mode=1, threshold=50) and I found that the the motion was much more fluid and the file size didn't change too much (960MB vs 1020MB). I noticed that avisynth makes a blurred "half image" between the repeated 5th frame and the frame after it. This is pretty much unnoticable during playback but it makes the overall playback smoother.

I was quite pleased with the results and my entire 42min capture was comb free (nothing snuck through) with no stutter.

[MITSUI_1]

Hi all,

Big thing to remember:

There are two things that can cause stuttering in the context we're discussing.

One is the "dupe" frame issue that resultes from doing a telecide without a decimate. The stutter you see (especially in camera sweeps, etc) is because one instant of time (a frame) takes twice as long to transaition as the othes. For some this is horribly distracting. For others, it is not. In non-computer animation (think Disney movies) it is hardly noticable.

The second one is MUCH worse, and it is the issue I think a lot of you are having. That is the problem that happens when you IVTC footage that is NOT telecined in the first place, or where your IVTC hits a "splice" that causes the wrong fields to be disgarded. This results in both combing and information for one frame out of four being displayed for TOO little time. Its a more noticable "jerk" than the other one.

I'll explain in a subsequent post if I have time.

[Nelson37]

Mitsui_1 - I'm sure its the second case, its not a freeze or stutter, its a jerk like a lost frame. In testing, I believe Decomb, using the guide settings, correctly performs IVTC, its the Decimate where I have the problem. With custom settings, I got it pretty close to deleting the correct frame but it left the framerate at 29 so playback was jerky overall, like a 3:2 pulldown was missing. No decimate at all after the Decomb probably looked best. Any ideas would certainly be appreciated.