Inverse Telecine: Double Checking My Work

[pulsar8]

So I'm new to this whole inverse telecine thing, but I read and got to the point where my output is 23.976024fps.

This is from a DVD cartoon. The pattern is that every 5th frame had the distortion from the telecine conversion to TV:


[Attachment 71840 - Click to enlarge]

Here's the m2v:

https://forum.videohelp.com/attachment.php?attachmentid=71841&stc=1&d=1687206470

Here's a sample of my result:

https://forum.videohelp.com/attachment.php?attachmentid=71842&stc=1&d=1687208224

Script information was:

tfm()
TDecimate(Mode=0, Cycle=5, CycleR=1, Denoise=true)

I think that means the 5th frame is handled? Which which fits the pattern? I think? I noticed not all patterns are the same for this which confuses me but I think I did it right.

Thanks

[jagabo]

I suggest you upload a long medium speed panning shot. It's much easier to tell what the frame rate should be.

[johnmeyer]

IVTC of cartoons is difficult because they have such a strange and unpredictable repeat pattern. I suggest you search to find settings which are specific to classic cell animated cartoons.

[pulsar8]

I suggest you upload a long medium speed panning shot. It's much easier to tell what the frame rate should be.

Does this work? Long enough and slow enough or?

https://forum.videohelp.com/attachment.php?attachmentid=71843&stc=1&d=1687217069

Thanks!

[jagabo]

Yes, that shot is fine. It's standard 3:2 pulldown so your

tfm()
TDecimate(Mode=0, Cycle=5, CycleR=1, Denoise=true)

should work fine.

[pulsar8]

Yes, that shot is fine. It's standard 3:2 pulldown so your

tfm()
TDecimate(Mode=0, Cycle=5, CycleR=1, Denoise=true)

should work fine.

Ok great. The results look normal so I guess I did it right.

Thanks

[jagabo]

The reason I asked for a panning shot was because those are almost always 24 different frames per second. That's because it's easy to move the background, take picture, move the background, take a picture... Character animation is much more difficult. It's usually done at 12 or even 8 frames per second to reduce the number of cells that have to be painted/drawn. With so little motion it's hard to tell what the correct frame rate is -- which duplicates are from the pulldown (or frame rate conversions) vs. which frames are duplicate film frames.

[JadHC]

Yes, that shot is fine. It's standard 3:2 pulldown so your

tfm()
TDecimate(Mode=0, Cycle=5, CycleR=1, Denoise=true)

should work fine.

Ok great. The results look normal so I guess I did it right.

Thanks

Random question out of curiosity. Does this bring ur FPS from 24 fps down to 19 fps? Will u interpolate missing frames or keep as is?

[pulsar8]

Yes, that shot is fine. It's standard 3:2 pulldown so your

tfm()
TDecimate(Mode=0, Cycle=5, CycleR=1, Denoise=true)

should work fine.

Ok great. The results look normal so I guess I did it right.

Thanks

Random question out of curiosity. Does this bring ur FPS from 24 fps down to 19 fps? Will u interpolate missing frames or keep as is?

Well mine was about ~30 fps when I started so I'm not sure, but the result was spot on 23.x frames per second.

[JadHC]

Yes, that shot is fine. It's standard 3:2 pulldown so your

tfm()
TDecimate(Mode=0, Cycle=5, CycleR=1, Denoise=true)

should work fine.

Ok great. The results look normal so I guess I did it right.

Thanks

Random question out of curiosity. Does this bring ur FPS from 24 fps down to 19 fps? Will u interpolate missing frames or keep as is?

Well mine was about ~30 fps when I started so I'm not sure, but the result was spot on 23.x frames per second.

gotcha. u started at 29.97 and got down to 23.976. That's the right way to go. I sometimes have PAL videos burned on NTSC DVD format with 29.97 fps but instead of inverse telecine, i just have every fifth frame a duplicate so i run regular decimate and that fixes the issue so we get the same results for different issues. thanks for clarifying

[SaurusX]

The way you did things is "correct", but if you read the documentation for TDecimate it says to just use mode=1 as the only parameter when using animation. You're probably getting to the same endpoint anyway.

The way TDecimate works here is by removing the "most duplicated" frame in every sequence of 5 frames. It makes sense, but the problem is that animation doesn't follow strict rules. There are always instances where the wrong frame is decimated thereby changing the original frame sequence. The harder, though more accurate, way is to manually examine the telecine pattern and then remove the extra frame with a selectevery() command.

If the show was edited on tape with a lot of orphan fields then the above just causes more problems, so a more automated approach would be better. And then barring that the telecine pattern can change at scene or reel changes, which you have to look out for.

This is probably more info than necessary, but IVTC is a bit of a rabbit hole like a lot of video processing.

[pulsar8]

The way you did things is "correct", but if you read the documentation for TDecimate it says to just use mode=1 as the only parameter when using animation. You're probably getting to the same endpoint anyway.

The way TDecimate works here is by removing the "most duplicated" frame in every sequence of 5 frames. It makes sense, but the problem is that animation doesn't follow strict rules. There are always instances where the wrong frame is decimated thereby changing the original frame sequence. The harder, though more accurate, way is to manually examine the telecine pattern and then remove the extra frame with a selectevery() command.

If the show was edited on tape with a lot of orphan fields then the above just causes more problems, so a more automated approach would be better. And then barring that the telecine pattern can change at scene or reel changes, which you have to look out for.

This is probably more info than necessary, but IVTC is a bit of a rabbit hole like a lot of video processing.

Thanks for the information. It sounds really hard TBH... I mean if its always changing then I guess you have to just decimate every 5 and cross your fingers? It doesn't seem practical to analyze it every minute etc to see if the pattern changed. And apparently the pattern can shift multiple times in one episode of a show? Yikes.

[SaurusX]

Here's another problem with TFM. With slightly shaky or wobbly video it will sometimes weave incorrect fields together because it passes the combing metric it uses to determine the presence of interlacing. The combining can even output seemingly identical fields stacked on top of each other resulting in a very chunky looking frame that looks like it was bob deinterlaced. Se below for an example of the smaller kind of error.

https://imgsli.com/NzA0MjU

[pulsar8]

Here's another problem with TFM. With slightly shaky or wobbly video it will sometimes weave incorrect fields together because it passes the combing metric it uses to determine the presence of interlacing. The combining can even output seemingly identical fields stacked on top of each other resulting in a very chunky looking frame that looks like it was bob deinterlaced. Se below for an example of the smaller kind of error.

https://imgsli.com/NzA0MjU

Yeesh... well good to know. I'm not sure where this leaves me. It's not practical for me to re-check the patterns in a show over and over again, and then micro adjust the settings. That would take forever.

I guess I either just live with occasional errors with what I have known or not bother to decimate at all.

[SaurusX]

I think in the end, the TFM and TDecimate combo is really the best anyone can reasonably expect. Especially for a long series. For someone (like me) who requires near autistic-level of perfection I'll closely watch the video and then create override files to correct the errors for a second pass. Really, though, I wouldn't expect that from anyone.

[pulsar8]

I think in the end, the TFM and TDecimate combo is really the best anyone can reasonably expect. Especially for a long series. For someone (like me) who requires near autistic-level of perfection I'll closely watch the video and then create override files to correct the errors for a second pass. Really, though, I wouldn't expect that from anyone.

Well when I level up in my video editing career maybe I'll demand the same level of perfection, but for now thanks for the tips.