AnimeIVTC "Clip2"

[thecoreyburton]

I'm using AnimeIVTC's mode 3 on a problem-source I have and am very pleased with the results. The source itself is meant to be 24fps, but at various moments a border (which I believe was added digitally in post, as there are some infrequent other elements added in the same manner throughout the source at 30fps) comes in around the image and throws off the frame rate restoration because each frame suddenly varies by a lot more.

In situations like this I often use functions like TDecimate's "Clip2" argument for applying the detection to a cropped source (cropping out the border, to be clear) but the decimation itself to the full, uncropped source. I've read through several long pieces of documentation and can't find any similar argument for AnimeIVTC. Is there one? Or is there perhaps a work-around I can use to achieve the same result?

Thanks for checking this out!

[real.finder]

o3025cfr=true in my edit of animeivtc (2.20)

this will just do field matching

then you can use TDecimate after animeivtc as you want

edit: but as you use mode 3 there are no field matching, so maybe this will not help you

[thecoreyburton]

Being stuck with mode=3 is unfortunate, but because it's a double hard telecined source it's the only thing that results in smooth motion. That being said, at least mode=3 does handle it correctly and fix it up, it's just those pesky borders giving me trouble.

I'll have a look anyway, it'll either help or perhaps give me ideas to make modifications of my own to achieve similar results. Thanks for the reply!

[real.finder]

can you upload sample from your source? I want to see how double hard telecined look like

[thecoreyburton]

Sure thing, it could be a couple of days though - I'm not at that particular PC right now but I'll edit this post with a link when I can!

[thecoreyburton]

Here you go. (http://www.mediafire.com/file/pb8tx8i5gayv23v/clip_example.zip) A few notes on the source / clips that may be of some use:

These are taken from two consecutive disks, released as volumes of the DiC version of the show (produced in the 90s). It looks like they added the logo in post, at 30i over the top of the telecined 24fps film, along with several other post-production elements. These don't matter to me as they're insignificant and in very few scenes, so I'd rather focus on the film than go the VFR route in this case.

Clip01 is a good example of panning with the border included. For these moments I originally was using a clip2 argument with around 120 pixels cropped from all sides, so it only used the relevant part of the film for detection (also, just a note that clip2 is handled differently in SRestore and TDecimate, and I used it appropriately in both cases). For the most part, this clip appears to be telecined regularly.

Clip02 is telecined regularly too, but also has the logo at 30i.

Clip03 is included for reference, see below.

Clip04 is the first "problem" clip. I'm honestly not sure if it's double hard telecined or not - but AnimeIVTC's handling of it with this mode is perfect with AnimeIVTC(Mode=2,KillComb=3,AA=3).

Clip05 is also included for reference, see below again.

The biggest issue is that there's no rhyme or reason as to when the videos switch between what you see in the third clip to what you see in the fourth clip. There's no consistency between episodes, scenes, etc - so there's no real way to predict whether the scene is going to require a simple field match or something more complicated.

My first few attempts with this consisted with various QTGMC settings, accompanied with various SRestore settings (various thresholds, speeds, etc). The moon panning to the title (as seen in clip02 and clip05) was one of the least consistent scenes in terms of results, which is why I included those clips. I found QTGMC+SRestore to work fine on footage like clip05, but clip02 produced jumpy results under certain circumstances. On settings that worked for both scenes, clip03's motion was completely destroyed.

I wouldn't have assumed it was a double hard telecine (and still have my doubts) but AnimeIVTC with the settings mentioned above does a fantastic job, I just need a way to make it only use the center of the image for detection but apply the filter to the whole image. I'm also open to any number of alternatives, provided the motion in all 5 clips is consistent and there's as little residual blending as possible (the logo is a different story and doesn't matter).

Hopefully the clips provide more insight to the problem, sorry they took so long to get up and apologies if any of the terms I used were incorrect - this is my first time dealing with a source of this nature.

[real.finder]

before doing another testing


AnimeIVTC(1,0,5, ediandnn=3, repwithtdeint=true, tfmm=0, micmatching=0, o3025cfr=true).TDecimate(1)
smam()


try this out

[thecoreyburton]

That's definitely the best result so far. The first four clips are fantastic, and the fifth turned out the best I think I've seen it. The moon moving into the corner is somewhat inconsistent, but given the source it could be the best it'll get, especially when I'm trying to get other sections smooth as well with the same script.

Was the footage actually double hard telecined or was I incorrect? And are there any further adjustments that can be made?

Again, thank you for taking the time to help. You've already improved this more than I could have!

[real.finder]

That's definitely the best result so far. The first four clips are fantastic, and the fifth turned out the best I think I've seen it. The moon moving into the corner is somewhat inconsistent, but given the source it could be the best it'll get, especially when I'm trying to get other sections smooth as well with the same script.

Was the footage actually double hard telecined or was I incorrect? And are there any further adjustments that can be made?

Again, thank you for taking the time to help. You've already improved this more than I could have!

I suggested this earlier https://forum.doom9.org/showthread.php?p=1778491#post1778491

don't know if your source called double hard telecined or not

and what is the definition of double hard telecined? in original AnimeIVTC Thread there are only dead links images

but Double hard telecine - your source has at least one of the "intra-frame variable framerate" situations and the telecining was applied separately on each part of the frame, so the pattern is not constant and many frames are interlaced.

is like Clip01

anyway, there are a lot of cases (Especially in old dvd) that not pure telecined, there are even 60i panning, 60i fading credits on 24 that can destroy IVTC in some frames, Scene change with unmatchable fields (cuz there are some fields trimmed in scene change as they did the show for TV only), and other cases

Except the cases of field matching failure with vertical pan in 480i, see here https://forum.doom9.org/showpost.php?p=1725899&postcount=48


And are there any further adjustments that can be made?


maybe a little, I always look for the most lazy solution and work as a general solution too, you can always use something manual like YATTA and kill your life time with it :devil:

but anyway you can always development others ideas and scripts, try and see, that is the open source benefit

[thecoreyburton]

Yeah, you're spot on. I'll play around and see what I can get, if not though, that's definitely the best outcome I've had so far so either way I'm happy.

Thanks for the script and explanation, all of this has been really helpful.