detelecine nonstandard video

[JClement]

I have 2 videos that I wish to detelecine before spot removal, and dejitter, but both are non standard.

1. One has 2 original frames followed by 3 interlaced frames. All of the detelecine programs handle 3:2 but not the reverse. Are there any that can handle this type of pattern??

2. The other has the 4 original frames followed by a ghosted combination frame. I have tried some modifications to detelecine, with very mixed results. I can ID a fair fraction of the ghosted frames, but by not as many as is needed. The last time I tried it I could ID maybe 3/4 of the ghosts. Again is there a program that can handle this type of thing? First I changed the comparison from interlaced of nearby frames to added together next neighbors. Then I changed the algorithm to detect which frame was ghosted. The original one looked at the largest difference between combined pixels in a group, but I used an STD of the differences in groups to find the ghosted frame. This worked better. Of course the ghosted frame was never in the same location in each group of 5 frames or I could have fixed that easily.

I have given up on 3:2 where the 2 are ghosts of the nearby rather than interlaced.

[burfadel]

For the first one maybe you could apply nnedi3 over it, force it to use the top or bottom field? If you end up with 4 frames followed by one still frame you could simply decimate it by using decimate() filter.

The second video sounds like the fifth frame is a combination of the fourth frame and the first frame of the next sequence. If this is the case, again you can simply decimate it by decimate() or decimate(cycle=5), and that video would be solved as well.

[johnmeyer]

I haven't a clue unless you post a sample, and make sure that it has not been re-rendered from the original.

SRestore can sometimes handle non-standard problems created by doing things other than standard pulldown (i.e., repeated fields) to convert film to video.

[JClement]

nnedi3 throws away half of the information, thus reducing the detail. It is an NTSC standard DVD, with nonstandard telecines. There is some jitter between the interlaced fields so throwing away half of the resolution might be OK. The ghosted telecines can not be fixed by just decimating. The ghosted frame is not always the fifth frame, and it can not be identified by just looking for a duplicate. I am trying to achieve 480P24 so I can reduce annoying artifacts.

Take the original film to be A1,B1,C1,D1,A2,B2....
The 2:3 telecining has A1, B1, B1b:C1t, C1b:D1t, D1b:A2t, A2, B2... with "BFF". The telecining is not uniform in that the progressive frames are not always A,B but could be B,C... detelecine usually fixes 2 of the 3 interlaced frames, but then always misses the third. I have not figured out the logic well enough to make it hunt for a third interlaced frame. Perhaps there is already a routine for this, as I have encountered other examples of this kind of telecining.

The ghost telecined one has A1,B1,C1,D1, D1+A2/2, A1,A2... It looks like progressive, with no "mice teeth" Again it is not a uniform pattern in that the ghosted frame could be C+D/2. This one is much more difficult because it is an old movie where the camera work tends to be static and the background stays the same. As a result comparing frames such as comparing B'+D'/2 with C' by subtracting them often yield mainly a dark frame with just a few differences in some areas. I doubt there is a clever program that can detect this. As I say I have gotten the bad hits down, so I may try some more things. Comparing differences across a wider time window might work better along with looking at how differences change with time. At present I just throw out the frame which has the lowest difference in a group of 5.

This sort of encoding is difficult to figure how it was achieved. Regular telecining 3:2 or 2:3 nonstandard could be achieved in a simple mechanical projector which switches frames during the flyback between interlace fields for the telecined frames. Persistence in the camera can then add ghosting. But the ghosting will only appear in one half frame, and not in the other.

[manono]

Upload an untouched sample as johnmeyer has already requested. A short video is worth a thousand words. Cut an M2V using DGIndex. 10 well-chosen seconds of a part with steady movement will be plenty. The upload it to some filesharing site and provide the link. MEGA, Sendspace, Mediafire, etc..

[JClement]

I have put a 10s copy of a section from the 2:3 nonstandard telecines video into:

https://www.dropbox.com/s/3s948ylh010mogc/2-3%20test%20video.avi?dl=0

Actually I have a second video I just got that also has the same problem, and is in more need of fixing than the sample one. I can't locate the ghosted one at present.

[manolito]

ConVertToYV12(interlaced=true)
TDeint(mode=1)
srestore(23.976)

looks pretty good to me...

[JClement]

Does that sequence bring back the original frames and does it smear defects across adjacent frames? Bobbing of course throws away every other line, but you still have them on the subsequent frame. If srestore just picks which frames to keep, then 1/2 of the detail is lost. It it combines frames it may be possible that dust & dirt will be smeared over 2 frames which makes automatic detection impossible. There should be a way to detect the interlaced frames and restore the originals almost exactly. Severe dirt or dust should appear on just 1 frame, which makes subsequent rotoscoping much easier. I will consider it.

[manono]

Does that sequence bring back the original frames...
No, as the original frames are gone. It bobs the video and then picks out the unblended frames. This isn't even a field blended standards conversion. It's field blended from film to NTSC. Only the worst fly-by-night companies do that. They could easily encoded it progressive or, at worst, hard telecined it so a standard IVTC would have returned the progressive frames. But no, they had to go and screw it up. Blame your source. I'm not a big fan of bobbing using TDeint, but it gets the job done.

Did you capture this? If so, next time don't cap as RGB.

Bobbing of course throws away every other line

Not exactly. Good bobbers keep pieces of the original frame that aren't moving. You lose half the resolution when there's movement. Again, blame your crappy source.

There should be a way to detect the interlaced frames and restore the originals almost exactly.
Right. And when you figure that out, please come back and share it with us. It's already been ruined beyond complete redemption.

Severe dirt or dust should appear on just 1 frame

There's ways around that (having the noise or dirt spread over 2 frames) involving splitting the video into even and odd frames and then applying the temporal filter to every other frame before then reassembling all the frames. I do it all the time. Ask if you want a script.

If you don't understand field blending, bob it any way you like and then go through it frame by frame, noting all the blended/ghosted/double imaged frames.

It is an NTSC standard DVD, with nonstandard telecines.

There was never any telecine at all; therefore it can't be inverse telecined, only unblended. It's pure garbage.

[Katie Boundary]

I'd like to see that clip out of morbid curiosity, but it's an avi file in a format that neither VLC nor Vdub recognizes. Can we get an actual M2V file?

[johnmeyer]

I'd like to see that clip out of morbid curiosity, but it's an avi file in a format that neither VLC nor Vdub recognizes. Can we get an actual M2V file?You need to check your computer configuration. It is just Lagarith and it plays in both VirtualDub and VLC on my computer. Pretty standard stuff.looks pretty good to me...Manolito, the output of that SRestore script looks perfect to me. Nice job!

[Groucho2004]

I'd like to see that clip out of morbid curiosity, but it's an avi file in a format that neither VLC nor Vdub recognizes. Can we get an actual M2V file?
You have to install the Lagarith codec (https://lags.leetcode.net/codec.html) in order to load the file in VDub.

[Sharc]

I'd like to see that clip out of morbid curiosity, but it's an avi file in a format that neither VLC nor Vdub recognizes. Can we get an actual M2V file?
Open it in VirtualDub FilterMod (https://sourceforge.net/projects/vdfiltermod/).
Or in avisynth using the DirectShowSource filter, rather than AVISource.
Or in MPC-HC.

[Groucho2004]

Or in avisynth using the DirectShowSource filter, rather than AVISource.
Or in MPC-HC.Both of these need either LAV or the LAGS codec.

Open it in VirtualDub FilterMod (https://sourceforge.net/projects/vdfiltermod/).That actually works without LAV/LAGS. Nice mod, I think I'll use this from now on.

[StainlessS]

That actually works without LAV/LAGS. Nice mod, I think I'll use this from now on.

Yeh, its great, also got a nifty lil avs script editor.

[Katie Boundary]

I had uninstalled lagarith after being told of its horrible bugginess and speed-vs-compression tradeoff compared to Huffy. However, reinstalling it did not make the video openable. I shall try vdub filtermod next...

EDIT: AGAIN, "unknown or unsupported file type". Maybe something is going wrong with the download process. Is the file supposed to be only 221 KB?

[Groucho2004]

Is the file supposed to be only 221 KB?It's about 100MB.

[wonkey_monkey]

You need to visit the page to find the download link, rather than right-click and Save As (which saves the page, but confusingly under the right file name).