IVTC of Super8 footage? (Daicon animations)

[Doc Sane]

I've recently gotten back to an old project of cleaning up an LD of the old Daicon 3 and 4 animations. The problem? They were shot on Super8 film (most likely at a frame rate of 18 fps), then telecined all to hell. The result is that no standard IVTC tools seem to work, as they all assume an original framerate of 24 fps. Now, how should I approach this? Try for the best 24 fps result, then figure out how to convert that down to 18 fps? How would I do that?

As you might guess, I'm at that awkward stage in learning AVISynth where I can sort of guess what to do, but I have nowhere near the scripting ability of the old hands here. Can anyone out there give me a hint of what to do? The plan is that, once I distill the animation down to its untelecined core, I can start doing things like despotting and descratching the footage.

Thanks in advance.

[manono]

Hi Doc, and welcome to the forum.

There are 3 ways I know of to get back to 18fps (assuming that's what it is). One is MultiDecimate, but it shouldn't be necessary. Another is SmartDecimate(18,60). And the third and most widely used IVTC is Decomb. By using 2 Decimate commands you can easily achieve 18fps:

Telecide(order=1)
Decimate(5).Decimate(4).

First you have to make sure it's really 18fps. To do that, apply Telecide with no Decimate:

Telecide(Order=1,Post=0).

And you should then see 2 duplicate frames in each cycle of 5 frames. And I'm assuming you have Top Field First (the Order=1 setting). The Decomb doc explains how to find out if it's TFF or BFF.

If none of that helps, perhaps you can make a short sample of the source available for us to have a look.

[Doc Sane]

I tried your suggestions, and now I'm not entirely sure if this was shot at 18 fps or 24. Just using Telecide and Decimate has given me an IVTC that seems to work in most cases. However, I'm still getting some blended frames in some scenes.

I should point out that I got frustrated with Decomb a while back and simply used the separate fields, doubleweave, pulldown commands in AVISynth, since I seemed to get more controllable IVTC results with that. This new version of Decomb works a lot better, though.

Anyway, how shall I go about determining the original frame rate of this? Shall I crunch the capture down and post a bit of it on my web space?

[manono]

Hi-

To determine the true framerate, you first have to figure out the cycle, and then determine how many duplicate frames are in that cycle. Using:

Telecide(Order=1,Post=0),

if there is one dupe in a cycle of 5 frames, then that's the result of a standard telecine, and the framerate will become 24fps (23.976fps) after Decimate(5) is added on. One dupe in six frames gives you 25fps after decimation. If there's one dupe in every three frames, then after Decimate(3), you'll get 20fps. Two dupes in every five frames is 18fps, and so on. Those are perhaps the most common and easiest. But they can get much more complex than that.

There are other ways to find out the framerate. One is to put on a SeparateFields() and then count how many original, unduplicated, unblended frames there are in a cycle of 60 (thanks scharfis_brain). And bilu has been working on another way to attack the problem:

http://forum.doom9.org/showthread.php?s=&threadid=69199

Shall I crunch the capture down and post a bit of it on my web space?

I'm not sure what you mean by "crunch down". We'll need the pure unadulterated, unprocessed source. But yes, that might be a good idea. There are lots of people around here that like to look at unusual material.

[Doc Sane]

Well, anyone who is interested in helping me figure this IVTC out (and who has a fast connection) can get a 12 second DV clip from http://neil.anime-translator.com/Daicon3_clip.avi I appreciate any help you guys might offer.

[scharfis_brain]

forget about all automatic/adaptive methods on this clip.

all you can do is something like this:

avisource("video.avi")
bob #or any other full-framerate deinterlacer you like

#duplicateframe(x).duplicateframe(x)
#duplicateframe(x).duplicateframe(x)
#deleteframe(x).deleteframe(x)
#deleteframe(x).deleteframe(x)

unblendpattern(off,23.976,false,true)
[code]

the function unblendpattern and its description can be found in this thread: http://forum.doom9.org/showthread.php?s=&threadid=66093

a LOT of work is in front of you, if you want to have a clean blend-free video!

at first, tweake the offset-parameter, until the beginning of the video (fisrt minute or so) is blend-free.

then step through you video and look for frames that are nearly 50:50 blends. the framenumber is always shown in the image.
now go back to your script and place

duplicateframe(blendedframenumber-1).duplicateframe(blendedframenumber-1)

between bob and unblendpattern

then reload the avs to vdub and examine this position a second time, if the blended frames had gone step to the next frames.

[manono]

scharfis_brain-

How do I get it to open? VDubMod says I don't have anything to decompress Sony Digital Video.

[scharfis_brain]

download a DV-Codec (eg. panasonic)

[Doc Sane]

Originally posted by scharfis_brain
forget about all automatic/adaptive methods on this clip.

all you can do is something like this:

...snip...

You know, I think I know why it's so hard to figure this clip out. I think the transfer was done extremely cheaply, using one of those long persistance phospher telesync set-ups. What this boils down to is that the video transfer was made with the intention of having the afterimages of the frames blending together while the video camera recorded them.

Which means that there's no way in hell I can ever properly IVTC these clips. The blending is built in, not caused by interlacing.

Any comments from the masters here?

[scharfis_brain]

pleas use my manual method. I've managed to get your sample-clip completely blend-free!

this is ,because you have the following pattern:


FilmFrame: A B C D E F G H I J K
VideoFrame -------------------------------------------
blended 0 2 5 7 10 12 15 17 20 22 25
normal 1 3 6 8 11 13 16 18 21 23 26
normal 4 9 14 19 24

As you can see, there is at least one field left, that is not blended.
Now it's up to you to select the right ones!

[Chainmax]

Doc Sane: Is there any chance that you might make Xvid versions of this and upload them somewhere if you manage to clean them up?

[Doc Sane]

Originally posted by scharfis_brain
pleas use my manual method. I've managed to get your sample-clip completely blend-free!

this is ,because you have the following pattern:


FilmFrame: A B C D E F G H I J K
VideoFrame -------------------------------------------
blended 0 2 5 7 10 12 15 17 20 22 25
normal 1 3 6 8 11 13 16 18 21 23 26
normal 4 9 14 19 24

As you can see, there is at least one field left, that is not blended.
Now it's up to you to select the right ones!

Um, I feel like an idiot, because no matter what value I plug into unblendpattern as an offset, I constantly get blended frames. Can you explain what the range of the offset is? And am I going to have to manually remove EVERY blended frame from this? That may take me months!

[Doc Sane]

Originally posted by Chainmax
Doc Sane: Is there any chance that you might make Xvid versions of this and upload them somewhere if you manage to clean them up?

With the amount of work this is turning into, I may just give copies of the raw DV capture to whomever sends me a DVD-R. The Daicon 3 and 4 animations (and the pencil tests for 4) all easily fit on one disc. The problem is that the LD itself was badly mastered, with lots of rot and the occasional analog glitch. I want to IVTC because it would allow me to get rid of a lot of the bad video without having to do outright painting of frames in Photoshop. There's splice tape all over the place, black spots, vertical scratches, blobs of dust on the frame... It's a real challenge, if you know what I mean.

[morsa]

use despot and descratch to get rid of those artifacts.

[scharfis_brain]

@morsa: it is very hard to get rid of 1-frame-defects, if the defective film has been fieldblended...

this means, that you have to remove that blending before removing dots & scratches.

@Doc Sane:
try to vary the off between 0 and 10.
the offset parameter only fits until a frame is dropped in the videostream (occurs when capturing from LD)

when you've found the correct offset for the beginning of the movie, you have to apply sone duplicateframe() oder deleteframe() commands if blendings occur in the stream again at later position...
deleteframe or duplicateframe are doing a offset to the blending pattern, thus helping unblendpattern selecting the right frames only.

but ensure weighting the count of deleteframes/duplicateframe - commands, else you'll get a audio async.