PAL DVD w/ghosting - Nothing seems to help

[Sighting]

Hello, I was trying to deinterlace one of my movies and make into progressive content, but so far without any luck.

There seems to be a pattern of ghosting/field blending followed by progressive frame(s). While the interlacing can easily be removed, the ghosting never goes away. Likely because it is already present on the DVD.

It appears to be a bad NTSC to PAL conversion. I also tried Restore24 and some other functions/filters, but I couldn't get it right. I was hoping anyone could tell me how to restore the progressive frames. I put 2 sample vobs on my webspace.
If you check both samples, there appears to be 1 progressive frame in every 25 frames (1 sec).

edit: samples removed

Help is much appreciated.

:thanks:

[MOmonster]

Hi Sighting
Your samples show a fieldblending pattern. I don`t know how constant the pattern of the whole film is, but if you want more informations, search for:
restore24
easy manual constant pattern deblending
another restore function
My Crestore function seems to have no problem with your source.
I made a quick test with this settings:
ord = last.getparity() ? 1 : 0

a = last.leakkernelbob(order=ord, sharp=false, twoway=true, threshold=7)
b = a.BilinearResize(440,240)

Crestore_film(bob=a, bclip=b, bthresh=50, range=3)

Didee´s restore24 should also do its job, but will be a lot slower (depends on your system). Scharfis_Brain function could also be a solution for you (don´t know how constant your pattern is).
Just try it ;)

[Sighting]

Thanks for the reply :) I'm giving your filter a try.. The speeds aren't good though on my 3.2ghz HT: around 1 fps.. could you let me know if that's normal?

[MOmonster]

Definitive not. On my athlonxp 2000+ (256mb ram) it runs with 15fps. Do you use it with the newest leakkerneldeint.dll? There was a bug in an older version. Use range=2 or 3, higher could decrease speed very strong. Don´t use other filters before the restoring and only less after.
Please post, if something helped you. :)

[Sighting]

Hey.. ah yes I was using the version from the Avisynth Filter Collection page.. Now I downloaded the latest version of leakkerneldeint and speeds are very good :thanks:

[Leak]

Hey.. ah yes I was using the version from the Avisynth Filter Collection page.. Now I downloaded the latest version of leakkerneldeint and speeds are very good :thanks:
That's strange - I just checked, and the WarpEnterprises page lists the same version I have up for download... *wonders*

np: Marsen Jules - Couer Saignant (Red, Green, Blue And Other Summer Feelings (http://www.sutemos.net/en/release;gid,11))

[Sighting]

hmm yeah I mixed them up then.. I think I used the one that came with the Restore24 package..

speeds for Restore24 are around 2 fps, which means Crestore_film() is 5-10 times faster.. am I right about this, or is it the cause of a bad version again? :confused:

[Didée]

Could be possible. The latest "official" version of Restore24 is rather old and, in fact, outdated. Can't even remember what speeds I was getting with it ... ;)
Due to unknown desease I just don't manage to complete the last few missing pieces of the new version ... which will perform even better than the old one, in terms of quality and speed as well. (With simple settings, about realtime speed could be possible on current >3GHz machines. HiQ mode will stay in range 5~10 fps, though.)

[Sighting]

Could be possible. The latest "official" version of Restore24 is rather old and, in fact, outdated. Can't even remember what speeds I was getting with it ... ;)
Due to unknown desease I just don't manage to complete the last few missing pieces of the new version ... which will perform even better than the old one, in terms of quality and speed as well. (With simple settings, about realtime speed could be possible on current >3GHz machines. HiQ mode will stay in range 5~10 fps, though.)

Good to hear that you're still improving it..I'm using the recommended settings:

a2=r24kernelbob(0)
b2=TDeint(mode=1,tryweave=false)

restore24(a2, b2)

Does the 'HiQ' mode in the future version have significant improvements in quality over those settings? Maybe I would prefer waiting for the new version then instead of spending 35 hours of encoding on a result that by the sounds of it could still be better :(

[Didée]

No ... there won't be a "HiQ mode" as such. For deblending/decimating scripts like we're talking about here, the term "quality" falls apart in

a) quality of motion

b) picture quality

and I was referring to the motion quality (a) only, because that's what can still be improved.
The picture quality (b) depends almost exclusively on the deinterlacer being used. And in that area I have very little to add ... TDeint is one of the best deinterlacers deinterlacers we have available for AviSynth. So, regarding PQ, the current version of R24 is just fine.

(BTW, Crestore becomes notíceably slower, too, when using TDeint instead of LeakKernelBob and TDecimate instead of Decimate :) )

[MOmonster]

@sighting
Also the new leakkerneldeint won´t speed up the last official restore24 version significantly. On my pc 0.3 fps faster or so. Yes the results of the last version of restore24 I have are really nice , you should wait for it. My function will be also updated soon. I implement the idea of another function into Crestore, the range factor is good to catch offset changes, but bad for speed. My next version will do it just a little bit faster and should find the pattern more accurate, what should give a better motion. :)

@Didee
With tdecimate I still don´t get it run on my pc, my new version don´t needs any longer a decimater, so a problem less. :p

[Sighting]

I tried the old Restore24 function on the source and unfortunately there seems to be a problem with decimating... Sometimes there is a duplicate frame left (2 exactly the same frames). Later on a frameskip can be noticed, meaning the wrong frame was decimated.

Is there any way I could fix that problem? :( I would also be very interested in testing the latest Restore24 version (if it's faster it would also save me a lot of encoding time). :)

[MOmonster]

@Sighting
If my function works for you, why not using it. You can of course also use tdeint as input. Use range > 3 to get better results (to high will slow down the process a lot).
ord = last.getparity() ? 1 : 0

a = last.tdeint(mode=1)
b = last.leakkernelbob(order=ord, sharp=false, twoway=true, threshold=5).BilinearResize(440,240)

Crestore_film(bob=a, bclip=b, bthresh=50, range=5)
For restore24 I think you should wait till Didee think it´s ready to release. This is more save, but of course it´s Didee´s decission. :)

[Sighting]

Hey MOmonster.. yes I tried your function, but unfortunately it's giving me duplicate frames as well.

[MOmonster]

Yes, I know this problem very well. Maybe lowering the bthresh just a little bit could help. If the dups are really, really seldom, it´s because of to many edits (offset changes). My next version will be released soon and will try to create the dups in areas, where they are less visible (because I have no solution against this dups at this moment, don´t checked how the new restore24 do it).
I think in the next week, I will offer the next version. Hope it will be done better.
Theoretical, if my function works right the number of the dups should be nearly the same as with restore24, right?

[Sighting]

Hey MoMonster,

I noticed you released a new version of your functions :)

I tried the default settings, but I'm still getting random dupe frames.. Could you please post me parameters that should eliminate the problem? Or is there any way I could deblend and then use a different decimate filter and then change the fps to 23.976?

TIA

[MOmonster]

I don´t know what are the best settings for you. This evening I will test a little bit. Maybe I find a good way for you. Of course you can use Cdeblend and then decimate it to 23,976. Play a little bit with the function, till you get good results. :)

Are there many dups, or are they less then with my last version?

[Sighting]

These parameters seem to give me pretty good results :)

ord = last.getparity() ? 1 : 0

a = last.tdeint(mode=1)
b = last.leakkernelbob(order=ord, sharp=false, twoway=true, threshold=6).BilinearResize(480,288)

Cdeblend(a, tclip=b, thresh=50)

TDecimate(mode=2,cycle=300)

[MOmonster]

Good, if this works for you. try also a higher thresh value (50 is sometimes to less. To avoid wrong blend detection on low motion scenes, you can higher the mthresh value a little bit. :cool:

[krieger2005]

I saw in the develop. Forum a thread where a filter was presented, which restore constant blended frames. Maybe this works for. Try it... it's name: i guess RestoreFPS.

krieger2005

[MOmonster]

I think it won´t give you the results you hope at this point of development, but in the near future this filter will be important for such conversion. The potencial of this filter is great. :cool:
If you have the source on a dvd, you should maybe wait for the next restore24 release that will be sonn I think. The motion and very strong the speed will be much better as Didee said it. ;)

[Sighting]

Hey,

overall the blend removal is good enough, but it's the dupe frames that are the real problem. The parameters that I posted in my previous post looked good when moving through the frames in virtualdub. It detects the right dupe frames then, but somehow tdecimate fails to detect some of the right frames when encoding.

I could always try if there's a real pattern of dupe frames. If there isn't, I think I'll just wait for that new Restore24 version or try some other decimate filters.

[MOmonster]

It detects the right dupe frames then, but somehow tdecimate fails to detect some of the right frames when encoding.
I can´t help you with this problem. Get sure that the encoder does all right! Try also smartdecimate, maybe a higher maxndl or a lower framerate (23,5 and then speed up) can help you. Don´t know what to do else. :(