doubleblending solutions

[jmac698]

I have the worst case of blending, every frame is blended. Here is how it is: tb,tb,tb,bb,bb,bb where tb=top field blended, bb=bottom field blended.
Since there are no clear frames around blends, normal tools can't work. I've tried removeblend, cdeblend, unblend, ivtcblend, but none would work.
So I've written a script that can bob the remaining fields. What's worse is that you can't use motion adaptive smart bobbers either.
My next script is going to manually subtract frames to try to recover them, but with macroblocking noise it looks pretty bad as well and has to be smoothed.
Any other ideas?
Btw, the blends are , first two are blended with previous, last is blended with next. Like this:
t1=t1*.66+t0*.33, b1, t2=t2*.5+t1*.5, b2, t3=t3*.5+t4*.5,b3,
t4,b4=b4*.66+.33*b3, t5, b5=.5*b5+.5*b4, t6, b6=.5*b6+.5*b5.

That isn't written exactly right, the blends are with the original frame, there is no more than two images forming one frame.
I haven't worked out the exact formula yet.
It may be pointless to fix :(

Update: I found newest restore24 and restore24fps, see if I can get them to work... brilliant stuff. I get very lost reading these threads, they need a stable homepage!

[jmac698]

Update: I went through all the deblending threads and tried just about every snippet posted: repal, restore24, restorefps, cdeblend, unblend, tdeint(tryweave=true), none of them removed blends. I have blends on top and bottom fields, and no clear frames.
Now I really need help!
http://rapidshare.com/files/3553731/deint_test.demuxed.zip.html

[Didée]

and tried just about every snippet posted: repal, restore24, restorefps, cdeblend, unblend, tdeint(tryweave=true), none of them removed blends.
Well ... if you tried all of them in the way you tried restore24 (just randomly throwing some stuff out of the script (http://forum.doom9.org/showthread.php?p=900939#post900939) when there're probs loading a plugin, and then wondering that it doesn't do what you want), then it's no surprise. ;)

I have blends on top and bottom fields, and no clear frames.
Didn't look at your sample ... but if *everything* in the source is blended, and there are no clear fields at all during motion, then there is nothing you can do.

Some things are just FUBAR. End of story.

[jmac698]

No, only one field is blended, the other field is fine. I made a script to bob the clean fields and that works, until there's a pattern break, but also it doesn't look very sharp.
I'd like to figure out how to make restore24 work, and why restorefps simply crashes in 2.56. Does it work for you?

[Pookie]

Re: Restore24


You might want to try the recently modded SmartDecimate and Avisynth_c plugins that Tritical compiled.

http://forum.doom9.org/showthread.php?p=897598#post897598


Also, create a separate plugins folder for the Restore24 dlls and load them manually in your .AVS


As far as your troubles with Repair.dll, there are two versions of the RemoveGrain package floating around - v0.9 and v1.0

http://home.arcor.de/kassandro/RemoveGrain/RemoveGrain.rar

and

http://home.pages.at/kassandro/RemoveGrain/RemoveGrain.zip

Of course, don't mix the two RemoveGrain packages up, then you'll get blended dlls :D http://www.laughterlab.com/wav/lach8a.wav

[jmac698]

That was an excellent answer, thanks. I'll try it out.

[MOmonster]

It´s a simple pal to ntsc normconversion with blends. I don´t know if repal works with such big resolutions, but the conversion itself isn´t the big problem. Maybe you just don´t choose the rigth fieldorder. Don´t trust always the given order from dgmpegdec.
On my old pc the script was extrem slow with your hd source, but this simple solution worked for me:
leakkernelbob(order=0)
Mrestore() # v1.6

[jmac698]

Mo: I really appreciate you taking the time to try that for me. I will check it out.

[jmac698]

MO: I tried it, it didn't work at first, but I found that there was a screwed up field order. It only works if it's processed with dgindex first. I'm pretty happy now :) Just need to do some deblocking.

[MOmonster]

Yes, I always use dgindex, so I didn´t have this fieldorder problem.

[bkman]

You can try undeftmatcher if the blended fields alternate. I've had some success with it.

Also, unblend + fdecimate works, but overblended sections end up jerky.

[jmac698]

MO: another question, briefly how does it work? Are the blended fields subtracted (blend=a*w1+b*w2,a=blend-b*w2), OR interpolated, (and with blending or motion compensating)? It looks pretty good, better than the script I wrote which interpolated the bad fields only.
I also found bob() instead of tdeint was much faster and not much difference.

In the big picture, I have two more quests now, fix horrible blocking, and reverse engineering proper aspect (oh, and also decoing some flashing white lines, at line 21).

[MOmonster]

Mrestore does the same as repal or restore24, it just outputs the clear frames and delete the blended ones. If you have a closer look to your bobbed source you can see that all clear fields are there.
Indeed the conditional code, used in Mrestore is better for NTSC restoring from PAL sources. The preview2 will hoply have a better code for these small cycles, but I don´t think that this will make a big difference with your source.
Instead of the simple bob I prefer leakkernelbob(0,4,true,true) in your case. Is also really fast and better than bob().
Heavy blocking is a problem of some hd encoders. Can I see it also in your short sample or only at scenechanges and so on?

[jmac698]

Ok, I tried leakbob, but it messes up the video. I now get duplicate frames and some ghosting. Is it possible it's messing up detection?
As for blocking, it's in some high action scenes, but not that sample.
It's a real pain that the site is down, I can't download deblockQED.
Check this out, it may affect your next version:
http://rapidshare.com/files/4579323/leaktest.zip.html