Need help determing deinterlace solution.

[Golem]

Sample clip - https://mega.nz/file/EfMG0KKY#kpUVgn4E3-3v-Js0kHkhVguI_ztuaU-zly6cIaxS3ss

I've been using FFMPEG to attempt a progressive output of what appears to be a hard telecine source. Three utilities flag this as a BFF interlace but ffmpeg "idet" indicates only about 1% as BFF/TFF (26/4). After about 30 attempts of wildly different combinations of attempted deinterlacing, detelecining (hard & soft), yadif, fieldmatching, including 20-switch scripts the best I can do is get it reduced to 2+2 with minor ghosting.

I'm hoping someone can show me where I'm overlooking something simple.

[Sharc]

There is nothing to deinterlace. It's a pretty hopeless mess IMO. Ditch it.

[ajk]

This doesn't seem to be an interlaced source, it's progressive 29.97 with blended frames. Assuming it was shot on film, it needs to be brought back down to 23.976. I don't know if you can do that with FFMPEG by itself, but an AviSynth script using Srestore can get rid of at least the majority of the blends. Quick test attached.

[Golem]

Appreciate the replies!

ajk - You actually did a fine job on that clip. As you imply, not perfect, but a very acceptable output. Seems to blend artifact for a single frame only at select scene transitions. Unfortunately it doesn't seem to hold true to a sequence. I'm tempted to ask to borrow the script you used but I've not tinkered with Avisynth since last using VirtualDubMod more than 15 years ago.

[ajk]

Sure, I can go through the process in a bit more detail. So what we start with is a sequence such as this:



We've got a clean frame "A", then another "B", but then what follows are a pair of blended frames. I've marked these as "B+C" and "C+D". The last frame is again clean, "D". So we need to extract frame "C", which currently only exists as a "ghost" in the blended frames. Srestore() has a post-processing mode for this purpose and that is actually all we need, since the video is not interlaced.

Srestore(omode="pp2")

You can try the other modes also and see what looks best. After running that bit of script we now have the following sequence:



Frame "C" has been extracted from the blended frames with some clever substraction using the clean "B" and "D" frames for reference. Observe for example the man's left hand as it approaches the bottom of the frame.

All that remains is to remove one of the now duplicated "D" frames to achieve smooth motion. TDecimate() is good for this purpose:

TDecimate(cycleR=1, cycle=5)

The parameters instruct the filter to remove one frame out of every five and it will automatically pick one that most appears to be a duplicate. This brings the frame rate down to 4/5 of the original, i.e. from 29.97 to 23.976.

Some blends will inevitably remain after this processing. I don't know for sure but I suspect the editing for the video has been done after it was transferred from film. This means that the blend pattern shifts at most scene changes, and there are a lot of scene changes in music videos. You could manually go in and replace the remaining blended frames with other means, e.g. using motion interpolation to recreate the missing frames but this is a topic in itself.


Now as for getting started with AviSynth itself, it can indeed seem intimidating. Fortunately though the filters needed here are not particularly bad in their dependencies, probably about 5-6 things need to be downloaded. The filters are described here:

http://avisynth.nl/index.php/Srestore
http://avisynth.nl/index.php/TIVTC/TDecimate

Additionally you will need a source filter for importing the video into AviSynth, in this case I used FFMpegSource() since I had it installed, but there are others that may work even better with .vob/mpeg2 sources.

http://avisynth.nl/index.php/FFmpegSource

If you install AviSynth+ and those filters it shouldn't be too hard to get this video sorted out.

I hope this helps!

[jagabo]

Try FixBlendIVTC() followed by TDecimate() in AviSynth:

LWLibavVideoSource("Sherrie Austin - Lucky In Love.vob") 
FixBlendIVTC()
TDecimate()

It does have problems with the changing pattern near cuts though. And compression artifacts lead to artifacts in some of the unblending.

[davexnet]

It must be my turn now! I tried exblend:
1st pass

d2vsource("C:\Users\davex\Desktop\avs_test\lucky1.d2v")
exblend(mode=1,show=1,override="c:\users\davex\desktop\ov.txt")

2nd pass

d2vsource("C:\Users\davex\Desktop\avs_test\lucky1.d2v")
exblend(mode=2)

ov.txt contains

205
260

[Golem]

Much appreciate the replies!

ajk - The explanations of insight and process are very welcomed. Yes, it definitely helps.

jagabo & davexnet - Those are quite good outputs. Surprising how well some of those filters work with what look to be default settings.

All of you have definitely supplied me with a bit more knowledge, insight and tools to accomplish my task(s). Many thanks!