Remove every Nth frame with Hybrid?

[blinkHQ]

Hey guys, I have a video which is saved as 60p but is in reality 50p. Every 6th frame is torn and is half a duplicate of the previous and half a duplicate of the next frame. How could I remove these frames in Hybrid? (and if it's not possible with it, what other software would you recommend?) Thank you

[manono]

SelectEvery(6,0,1,3,4,5)

You'll probably have to adjust the missing number for the dupe frame in the sequence of 6. And if the pattern changes as you go along, it won't work at all.

[blinkHQ]

Okay, thank you, I'll try this. It doesn't change afaik so this might work

[johnmeyer]

If the pattern "hiccups" and doesn't repeat perfectly all the way through (which, unfortunately is probably the case) you may be able to tackle the problem with TFM/TDecimate. I've used these tools for all manner of weird non-repeating repeats (if that phrase makes any sense). You can usually nail tough cases by increasing the Cycle parameter in TDecimate, so it looks at a really large number of fields/frames before deciding which ones to delete. This slows down the operation a lot, but if you only have to do it once, that doesn't matter too much.

[blinkHQ]

Thank you! Does this work also if it's not a full duplicate but a very slightly shifted torn duplicate of two frames?

[poisondeathray]

A 50/50 mixed frame will usually be detected as different frame, if you use default settings

Since every 6th frame is a (top half frame before/ bottom half frame after) mix , another approach is to use TDecimate's clip2 parameter .

Specify clip2 as the original, but the input as a cropped version (top or bottom half) . The metrics will be run on the cropped version, but the returned frames will be from the full sized clip

And there are many other settings you can tweak , the point is to do this adaptively in case the cadence breaks (eg. maybe it was edited after it was put into that format)

[manono]

Specify clip2 as the original, but the input as a cropped version (top or bottom half) . The metrics will be run on the cropped version, but the returned frames will be from the full sized clip

Nice! I hadn't thought of that. I'd still try the SelectEvery method first, hoping the pattern doesn't change, but if it does yours is definitely the way to go.

[blinkHQ]

The problem is bigger than I thought. Not only is it not regular (not exactly every 6 frames) the tearing is irregular as well. Meaning that every tear is in a different position. It starts at the top, then every torn frame (every 5-6 frames) has the tear about 5% lower. This goes on until it hits the bottom then it loops back up. Is there any way to fix this?

[johnmeyer]

At this point you have to provide a sample. This moving "tear" is sounding like something more than a standards conversion screw up.

[blinkHQ]

I attached a 30 second clip. So this is a 30i video which was deinterlaced to 60p with QTGMC. The show was broadcast in Australia, thus should be 25i, but it isn't; the tearing is present on the source as well. (it's upscaled to 4k because the final product will be for youtube and that helps with bitrate)

[poisondeathray]

You should upload a sample of the original video instead of a processed, upscaled version

QTGMC can introduce errors, upscaling can introduce errors. e.g. a duplicate split frame can cause QTGMC to introduce motion, when there was none. This will impair your chances of detection and "fixing" it. You need to examine the original fields

[blinkHQ]

Okay, here's the source (well, part of it)

[blinkHQ]

bump

[davexnet]

bump

you can use Avisynth/Srestore and reduce the frames per second to 25,
seems to be the underlying frame rate

mpeg2source("C:\Users\davex\Desktop\avs_test\vob4.d2v")
yadifmod2(mode=1, edeint=nnedi3(field=-2))
srestore()

EDIT - Actually this is better, preserves the fluidity

mpeg2source("C:\Users\davex\Desktop\avs_test\vob4.d2v")
qtgmc(preset="slow",edithreads=3)
tdecimate(mode=0,cycle=12,cycleR=2)
crop(12,8,-24,-12).depansafe()
spline36resize(640,480)