help to decimate this xvid

[spiritgumm]

I have a choppy xvid someone made from a video on demand - probably used a screen capture program set at 29.97 fps. I'd like to reencode it in, and while TDecimate (in Avisynth) improves it, there's still left-over dupes every couple seconds, causing a pause. I added TFM (which is probably unnecessary), and looked into the different TDecimate options but I havent hit on the right solution. Also tried Multidecimate with no luck.

[sanlyn]

The sample appears to have originally been film-based progressive with 3:2 pulldown but has been incorrectly deinterlaced. Try the Srestore() plugin to restore it to progressive 23.972 FPS.

[spiritgumm]

tried it already, but thanks

[jagabo]

TDecimate(mode=2, rate=23.976) comes close. But there are still a few duplicate frames and skipped frames. You can probably get better results with a little fine tuning (maxndl, dupThresh, etc.).

[spiritgumm]

it was the fine tuning I needed help with. I'd try AnimeIVTC but the plugin package is dead.
I thought there'd be an easier way to set TDecimate to delete the extra set of duplicates every couple seconds (at least that's what it looks like on the video which might not be apparent on the clip), but I tried some of the TDecimate settings without improvement.

[jagabo]

Well, the source clip is missing frames so you will never get perfect results. For example, between frames 119 and 121 there is a duplicate frame. The real frame 120 is missing. The same happens between frames 124 and 126. This is why there are still duplicate frames after TDecimate. To hit the right frame rate those duplicate frames must be left in to make up for the missing frames. You could decimate to a lower frame rate to remove the duplicates but you'll still have jerks at the missing frames. You could try using ReplaceFramesMC() (aka, RX()) to reconstruct the missing frames. But you'll have to locate the duplicate frames by hand.

Frames 96, 100, 156, 160 (after decimation) replaced with motion compensated frames.

import("C:\Program Files (x86)\AviSynth 2.5\plugins\ReplaceFramesMC.avs")

AviSource("Short.avi") 
TDecimate(mode=2, rate=23.976)
ReplaceFramesMC(96,1)
ReplaceFramesMC(100,1)
ReplaceFramesMC(156,1)
ReplaceFramesMC(160,1)

[spiritgumm]

thanks. As it's an xivd, I'll just live with the TDecimate version.
So this was suppose to be recorded from VOD (either from Amazon or Netflix), and usually people use screen capture programs. Can we say if the problem was caused by wrong capture setting?

[jagabo]

Note I added a sample script and clip.

So this was suppose to be recorded from VOD (either from Amazon or Netflix), and usually people use screen capture programs. Can we say if the problem was caused by wrong capture setting?

There's no way of telling. Unless you want to pay to stream the clip yourself.

[spiritgumm]

if I could find a reliable free screen capture program, I'd try it. In this case, the movie is no longer available on those sites. If we knew the problem was the wrong capture rate, I'd ask the guy with the xvid if he still had the VOD and would try a new setting.

Jagabo, the clip looks great, but even if it were dvd video and not xvid, it's not worth it to write a script itemizing missing frames every every couple seconds for a 90 minute movie.

[poisondeathray]

You can automatically fill in duplicates with motion interpolation using Mug Funky's "filldrops" or related filters

I would actually leave it 29.97 , it is the smoothest to my eyes. The other option is to use some sort of decimation then filldrops, or filldrops then some decimation resulting in 23.976. . Test it the options on some longer sequences

To tweak the results of detection, add show=true at the conditionalfilter line . 1.3 is usually a good value

AVISource()
Filldrops2()


function filldrops2 (clip c)
{
even = c
super_even=MSuper(even,pel=2)
vfe=manalyse(super_even,truemotion=true,isb=false, delta=1)
vbe=manalyse(super_even,truemotion=true,isb=true,d elta=1)
filldrops_e = mflowinter(even,super_even,vbe,vfe,time=50)

ConditionalFilter(even, filldrops_e, even, "YDifferenceFromPrevious()", "lessthan", "1.3")

}

[jagabo]

You can automatically fill in duplicates with motion interpolation using Mug Funky's "filldrops" or related filters

Interesting filter. Sample #1 has a speed up, slow down, jerkiness several times a second. It didn't really work in #3. It might be an acceptable automated compromise for the OP though.

I'm trying to figure out exactly what the filter does. I think it replaces duplicate frames with motion interpolated frames from the two surrounding frames?

[poisondeathray]

You can automatically fill in duplicates with motion interpolation using Mug Funky's "filldrops" or related filters

Interesting filter. Sample #1 has a speed up, slow down, jerkiness several times a second. It didn't really work in #3. It might be an acceptable automated compromise for the OP though.

I'm trying to figure out exactly what the filter does. I think it replaces duplicate frames with motion interpolated frames from the two surrounding frames?

It detects duplicate pairs, motion interpolates the 2nd frame with an "inbetween" frame

His problem , however , is there are legitimate duplicates from telecine, and non legit duplicates (from dropped frames due to capture ) . This filter won't distinguish between which are legit which are not

[jagabo]

It detects duplicate pairs, motion interpolates the 2nd frame with an "inbetween" frame

Which explains why it doesn't work with some of the duplicates after decimating the OP's video -- on those the first of the duplicates is the one that needs to be interpolated, not the second.

[spiritgumm]

posion, if you saw the whole video, you'd see its unacceptable.
Seems like regular decimation should come first, followed by filldrops - assuming it's substituting a new field for a duplicate field.
I thought the first clip was the best of the three, but I still see a pause near the start.

[jagabo]

Yes, FillDrops2() is designed for use when you have captures with dropped frames -- where the capture device substituted a copy of the previous frame to fill in for the missing frame. In cases like that you have a sequence like:

1233567...

A second copy of frame 3 was substituted for the missing frame 4. When FillDrops2() sees the two copies of frame 3 it generates an in between frame between the first 3 and the following 5. Which comes close to what frame 4 would have looked like.

123(4)567

But after decimating your sample video you have some sequences like:

1235567...

In this case FillDrops2() replaces the second 5 with motions interpolated between 5 and 6. So you still have a jerk between frames 3 and 5, followed by "5.5" then 67...:

1235(5.5)67...

[spiritgumm]

huh.
After nominal decimation, the duplicate frame occurs (from what I see) once every couple seconds. Seems like the solution should be at our fingertips.

[spiritgumm]

poison, this filldrops concept should work with this, but it seems to stop working and go bonkers. I tried another clip, decimated it first, then ran that clip thru your script. It worked with the first pair if dupe frames, didnt quite catch the 2nd group, then started outputting very slow video.

[poisondeathray]

poison, this filldrops concept should work with this, but it seems to stop working and go bonkers. I tried another clip, decimated it first, then ran that clip thru your script. It worked with the first pair if dupe frames, didnt quite catch the 2nd group, then started outputting very slow video.

Are they duplicate pairs , or triplicates ? This will only detect duplicates

Theoretically you shouldn't decimate it first because you are missing frames (unless the decimation was done manually). Script based decimation isn't always perfect here - If you throw out the wrong frame with the decimation before you fill in the missing frames, you will never get smooth motion

None of these scripts will work perfectly, because you are missing frames, and have telecine (legitimate duplicates) . Since it can't distinguish between between the 2 types of duplicates , you either have to do it manually as jagabo suggested, or accept this "half measure"

Filldrops and related filters are designed for video ie. 25 or 29.97 captures, where you have dropped frames that manifest as duplicates (and they are not supposed to have duplicates), not telecine

Without a better description or sample, I'm guessing you might need to tweak the conditional filter value (you might get false positives - it might replace "good frames", thinking that 2 different frames are the same, or you might get missed duplicates - if the detection value isn't sensitive enough)

[jagabo]

poison, this filldrops concept should work with this, but it seems to stop working and go bonkers. I tried another clip, decimated it first, then ran that clip thru your script. It worked with the first pair if dupe frames, didnt quite catch the 2nd group, then started outputting very slow video.

I already explained why that happens. In short, filldrops assumes the second of the duplicate frames is the one that needs replacing. But in your video it's sometimes the first that needs replacing.

[spiritgumm]

poison, this filldrops concept should work with this, but it seems to stop working and go bonkers. I tried another clip, decimated it first, then ran that clip thru your script. It worked with the first pair if dupe frames, didnt quite catch the 2nd group, then started outputting very slow video.

I already explained why that happens. In short, filldrops assumes the second of the duplicate frames is the one that needs replacing. But in your video it's sometimes the first that needs replacing.

I know, but I was just prompting the old boy to have a look at the Filldrops script. If it works half the time, that would be an improvement...but what about the crazy added frames later - a glitch in the script?