another restore function

[Dogway]

Other experienced users recommended me to use it. I investigated on the meGUI avs script creator, and the help web, and they didnt mention anything about distributor. So lets check if thats the reason now...

[Caroliano]

Quickly browsing through the long sample, it mostly looks like "progressive telecine with fieldblending interrupted by duplicates".

Impossible? No, not impossible. It's animation!
If I undertood it correctly, AnimeIVTC(mode=1, blend=true) seems to have an nice trick to counter this:

#Remove duplicates (like a safe selecteven/odd) for further processing before final decimation (hybrid
#decimation (except blend decimation to prevent having 2 calls of tdecimate) or blend telecine IVTC)
#Once bobbed, in mode=1 (tdecimate(1,3,5), in mode=1 + blend = back to 30fps for further deblending,
#in mode=2 srestore will do the job, in mode=3 tdecimate(1,3,5), in mode 4 VFR needs 30, CFR can do it (personal notes)
debprep = chrfix ? mode==1 && blend || mode==4 && omode==2 ? chrdeb.tdecimate(1,1,2) : chrdeb : i
and then it does: debprep.srestore(omode="pp3")

It seems to improve SRestore detection and error correction in my source


BUT I have an problem.

SRestore is artfacting on me in places that even tfm() works perfectly:

animeivtc(mode=1) on the left, animeivtc(mode=1, blend=true) on the right. Simple srestore(omode="pp3").Decimate() also produces the same artfacts.

http://img641.imageshack.us/img641/9505/animeivtcsimples.th.png (http://img641.imageshack.us/my.php?image=animeivtcsimples.png)http://img717.imageshack.us/img717/5143/srestorefudendo.th.png (http://img717.imageshack.us/my.php?image=srestorefudendo.png)

http://img695.imageshack.us/img695/5482/animeivtc.th.png (http://img695.imageshack.us/my.php?image=animeivtc.png)http://img203.imageshack.us/img203/9558/srestoredestroando.th.png (http://img203.imageshack.us/my.php?image=srestoredestroando.png)

But if I use Spline64resize(720, 240).SRestore(omode="pp3").Decimate().Spline64resize(720, 480) I have plain blends instead of those artfacts. Is SRestore suposed to produce such artfacts? IF yes, how can I detect them to fall back to standard IVTC?

I already managed to detect the combed frames that SRestore outputs with IsCombedTIVTC(), and if I dodge this type of artfact, then I will have an almost perfect output. I noticed that there is an increase in brightness or contrast on the affected areas. But I don't know how use masktools or something like that to generate an boolean value for conditionalfilter().

An sample specific of this problem: http://www.mediafire.com/?4jzzm5jjtod
Further samples not so adequated for this specific issue in this thread: http://forum.doom9.org/showthread.php?t=153095

[GMJCZP]

Hello,

I'm getting an old problem rearing it's ugly head again.
"I don't know what d32 means".
It is difficult to work out why this is written across the video sometimes and not others. If it appears I simply abort virtualdub and save the file again and it will be fine. This means that I can never use fast recompress mode as I need to see the video!

Any thoughts most welcome...


Hello everyone:

When by chance I tried the option with AnimeIVTC chrfix = true I got exactly the same error that Floatingshed. I'm using version 2.7e SRestore, compared it with the version 2.65 (yes I work, well, the error message I don't see) and I see this line in 2.7e:

d43 = jmp ? d32 : d_v
global d32 = jmp ? d21 : d_v

At 2.65 it looks:

global d43 = jmp ? d32 : d_v
global d32 = jmp ? d21 : d_v

Missing global statement before d43 in 2.7e.

I'm not familiar with this language, but perhaps there is achieved a solution. Greetings.

[Gavino]

Missing global statement before d43 in 2.7e.
d43 does not need to be global, as its value (unlike d32, etc) does not need to be retained between successive calls to 'srestore_inside'.

The logic of the function should ensure d32 is set the first time it is called (jmp will be false), so I can't see why the "don't know what d32 means" error should occur.

[Gavino]

When by chance I tried the option with AnimeIVTC chrfix = true I got exactly the same error that Floatingshed.
Looking at the source of AnimeIVTC (which uses SRestore), I see that when chrfix=true, it calls a function chroma_rep - this function writes to a global variable 'lfr' inside a ScriptClip call. Now, SRestore also uses a global variable lfr for its own purposes, controlling the correct initialisation of variables like d32.

I suspect the "I don't know what d32 means" error is a result of interference between the two functions using the lfr global. If I'm right, the fix is to use a different (and unique) variable name in AnimeIVTC's chroma_rep function. (Has anyone come across this error when not using AnimeIVTC?)

Note that (as I warned here) problems with SRestore can occur if you have a global variable in your script (or in a function called by your script) with any of the following names:
dm, om, bom, thr, frfac, numr, denm, mec, det, bclp, dclp, fin, lfr, offs, ldet, lpos.

[Boulder]

Are those global variables one reason why SRestore shouldn't be used when multithreading the script (that is, you should use SetMTMode(5) before the call) ?

[Dogway]

I havent had any issues using SetMTMode(2) with srestore so far.

[Floatingshed]

I get the d32 error with the very simplest of scripts. Something like:

avisource("...etc")
tdeint(mode=1, order=1)
Srestore(frate=25)

[Gavino]

@Floatingshed

Which version of SRestore are you using?
Are you using vanilla Avisynth (no MT)?

[Floatingshed]

@Floatingshed

Which version of SRestore are you using?
Are you using vanilla Avisynth (no MT)?


Yes, vanilla avisynth 257.

I have tried SRestore 2.7c & 2.7e.
2.7c gives me "I don't know what d32 means" (can't remember which line it points to at the moment.)

2.7e gives me "I don't know what bp2 means" ([scriptclip] line 46)

I hope we can resolve this, SRestore does a superb job, when I can get it to work! Thanks.

[Gavino]

2.7c gives me "I don't know what d32 means" (can't remember which line it points to at the moment.)
2.7e gives me "I don't know what bp2 means" ([scriptclip] line 46)
That's puzzling. In 2.7e, the scriptclip part only has one line.
However, bp2 does occur on line 46 of 2.7c's scriptclip section, so maybe you got these the wrong way round?

This is quite a strange problem, so every bit of information helps in tracking it down.
Can you say (for each SRestore version):
- does the error message appear on every frame, or only certain frames?
- have you tried with Avisynth 2.58?

BTW In case anyone is wondering, I am not the author of SRestore and I don't even use it myself, but I have a technical interest in its implementation after fixing the memory leak in version 2.7c.

[Boulder]

Is there any way to fix this kind of video automatically (i.e. without trimming the video into different parts treated differently)?

http://www.mediafire.com/download.php?xnmnwqonj1g

The beginning can be repaired using SRestore(speed=-1) but the outdoor shots leave some blending behind. Or is the source just FUBAR regarding the outdoor shots? I recall the second season was shot using different equipment so I'll need to check that out too. I'd very much like to get both seasons on my Popcorn Hour with better video quality than the DVD can offer.

[Floatingshed]

That's puzzling. In 2.7e, the scriptclip part only has one line.
However, bp2 does occur on line 46 of 2.7c's scriptclip section, so maybe you got these the wrong way round?

This is quite a strange problem, so every bit of information helps in tracking it down.
Can you say (for each SRestore version):
- does the error message appear on every frame, or only certain frames?
- have you tried with Avisynth 2.58?

BTW In case anyone is wondering, I am not the author of SRestore and I don't even use it myself, but I have a technical interest in its implementation after fixing the memory leak in version 2.7c.

Hello again, It is possible that I got it the wrong way around. I am away from home and working mostly from memory. I get home this weekend so will let you have concrete info then.
I haven't used avisynth 2.58, I've always had a strict policy with software, if it works for me don't risk upgrading! I'll try it though.
As regards the error message it appears on every frame.
Thanks for your continuing help.

[18fps]

Can you say (for each SRestore version):
- does the error message appear on every frame, or only certain frames?
- have you tried with Avisynth 2.58?


I've seen this error mainly when in AvsP I'm trying filters and I uncomment srestore and press F5 to refresh. Then the message appears for every frame. But usually I don't care, because when processing the script with virtualdub the error doesn't appears anymore. So I had assumed it had to do with unsequencial access in the timeline.

[Floatingshed]

OK. bp2 error occurs with version 2.7c, d32 error is with version 2.7e, so yes I got confused there!
The error message is always burned-in to every frame.
I tried Avisynth 2.58 and got cavisynth stream errors! I now recall that I tried 2.58 once before and found it very unstable.
I should point out that these problems occur on more than one machine.
Any further help much appreciated...
Thanks.

[Boulder]

What are the contents of your Avisynth plugins folder?

[Floatingshed]

What are the contents of your Avisynth plugins folder?


It would be tedious to list. Do you have a suspect in mind?

[Boulder]

You can output the contents by using MS-DOS prompt and typing dir "c:\program files\avisynth 2.5\plugins" > list.txt . Replace the path inside the quotes with your Avisynth plugins folder if needed. The output of the dir command can then be found in the list.txt file which is created in the directory where you ran the command.

I don't have any specific suspect in mind, but the problem you described is often caused by a single plugin. One thing to try is to empty your plugins folder and put in there only the files that you need to run the script.

EDIT: Also, what is your OS and what is the name of your Masktools v2.0 dll?

[Sub-zero]

Hi, I have a problem with a blended anime source.
It is 24fps telecined to 30fps. However after a applied IVTC on the source (simple tfm()+tdecimate()) i noticed too much residual combing. I invistigated more with a script like this:

DGDecode_Mpeg2source("kaichou01.d2v", cpu=0)
tfm()
tdecimate()
Separatefields()
Levels(85, 0.092, 219, 0, 255)

I used Levels instead of histogram(mode="luma") because the blends were easier to investigate this way, after all I still a newbie.
I'm not very sure about these blends, they are strange, they are blends, yet not very visible, I noticed them by chance.

I took a sequence of frames and tried to find a pattern for blends, I succeeded in a pattern like this: (each char stands for a field, c=clear, b=blend, note that this pattern defining was post-ivtc, and the first field is top field)
c c b b c c b c
this pattern repeated for more than 5 cycles (ccbbccbc ccbbccbc ccbbccbc ccbbccbc ... etc )

I tried Srestore(omode=3,speed=-26) but the result wasn't very satisfying for me. so i thought of applying the filter on top fields and bottom fields separately with something like this:
a=Separatefields().SelectEven().Srestore(omode=3,speed=-26)
b=Separatefields().Selectodd().Srestore(omode=3,speed=-26)
interleave(a,b)
forgetting that Srestore's vars are global, which caused the script to fail.
here is a link to 4 seconds from the source: http://www.mediafire.com/?mumdiy3tujz
I'm a complete newb at IVTC, so guide me if i'm doing it wrong. Thanks.

[Episodio1]

Hi!

I'm new to AviSynth. I have to use it now cause I got this movie DVD: ntsc 29.97fps interlaced with ghosting. DVD cover even shows "remastered from HD" xD

I dont know if I have to use only "Srestore" or more filters. What filters should I need to use?

Sample: http://www.sendspace.com/file/gexrro (12 Mb)

Or here: http://www.megaupload.com/?d=B2U23NPS


Thank you.

[manono]

Hi, I have a problem with a blended anime source.
It's not blended, at least the sample you provided isn't blended.
However after a applied IVTC on the source (simple tfm()+tdecimate()) i noticed too much residual combing.
I noticed no residual combing in the sample provided. If something's wrong, you'll have to provide a sample that demonstrates it.

[manono]

I have to use it now cause I got this movie DVD: ntsc 29.97fps interlaced with ghosting.
You should be using AviSynth anyway, with easy or difficult sources. The blending in the beginning is intentional. Your DVD isn't blended and just needs a standard IVTC (Decomb, TIVTC, whatever).

[Episodio1]

MANONO, thanks. I meant with "I have to use it now" that "I have to tweak it manually for 1st time" :) :)

[Sub-zero]

It's not blended, at least the sample you provided isn't blended.

I noticed no residual combing in the sample provided. If something's wrong, you'll have to provide a sample that demonstrates it.

Levels(85, 0.092, 219, 0, 255)

http://img441.imageshack.us/img441/5683/cleanfield02.th.png (http://img441.imageshack.us/i/cleanfield02.png/)
http://img3.imageshack.us/img3/3655/blendedfield02.th.png (http://img3.imageshack.us/i/blendedfield02.png/)

http://img63.imageshack.us/img63/2439/cleanfield.th.png (http://img63.imageshack.us/i/cleanfield.png/)
http://img90.imageshack.us/img90/3429/blendedfield.th.png (http://img90.imageshack.us/i/blendedfield.png/)

http://img171.imageshack.us/img171/2489/blendedfieldframe.th.png (http://img171.imageshack.us/i/blendedfieldframe.png/)
http://img140.imageshack.us/img140/5179/blend01.th.png (http://img140.imageshack.us/i/blend01.png/)

or i shouldn't care about such minor (not visible until you look for them) artifacts?

[manono]

And what's the point of wrecking the luma levels to see something that's unnoticeable otherwise?
or i shouldn't care about such minor (not visible until you look for them) artifacts?
Yes, I think what you're doing is a complete waste of time. If it were I, I'd focus my energies on something else, like removing that annoying logo. Maybe others think differently about what you 'discovered'.

[MOmonster]

@Sub-zero
I can also see no blending on my display. There is no function that would detect this. Imho it makes no sense try to remove this, because it's not visible. It's caused by a temporal filter that let gleam the last frame a little bit. Srestore is no option here.
The residual combing is also hardly visible. But when it's really that annoying for you, you can use for example yadif(mode=1).blendbob()
instead of tfm.

[lapinot]

Hi.

Some will say I didn't probably try hard enough, and you can throw me the rotten tomatoes if the sample I propose is a piece of cake for you guys, but I spent some time trying in vain to get a proper output of this :
http://demo.ovh.net/fr/ca197199e611e028467cfe7aa1beee13/


this is probably the same story as Shaw Bros dvd, ntsc stuff transcoded from Pal masters (4 clean frames + 5th/6th blended), except it seems much more tough to make it work

I started with the usual
tdeint(mode=1)
srestore()
which usually was enough for sd sources

doesn't work at all (choppy/ghosting)
then read in doc (at last :cool: ) srestore required negative speeds with hd sources

so I went for srestore(speed=-x) #seems less catastrophic between -2 & -5
although far from any kind of decency. I played with threshold without valuable results either

then I tried yadif(mode=1) -worse- , then tfm+changefps(60000,1001), which gave me something a bit less crappy than deinterlacers but yet very far from being proper

I played a bit with blockx/blocky too within tfm in case hd required different values, didn't find any worthwhile value.

in the end, nothing that looks like a proper ghosting-free pal output,... I'm open to all suggestions that don't require math degrees :)
thanks

[Didée]

Just tried this:
AVCSource("F:\Downloads\sample_(lapinot).dga")
crop(0,128,-0,-128,true)
bob(.0,.5) # yadif(mode=1) # ... or whichever bobber that fits your needs ...

x2 = reduceby2().reduceby2() # downsize for analyzing (probably needed)
x1 = spline36resize(1024,440) # downsize for actual output (not needed, it's just to get a result QUICKLY

x1.SRestore(frate=24.975,dclip=x2)
return(last)

and got this: KLICK (http://www.mediafire.com/file/jk2xokmmqvy/SRestoreTest.avi) (MediaFire, 5 MB)


Perhaps it's because I'm more asleep than awake, but ... I don't see a problem?

[lapinot]

Just tried this:
AVCSource("F:\Downloads\sample_(lapinot).dga")
crop(0,128,-0,-128,true)
bob(.0,.5) # yadif(mode=1) # ... or whichever bobber that fits your needs ...

x2 = reduceby2().reduceby2() # downsize for analyzing (probably needed)
x1 = spline36resize(1024,440) # downsize for actual output (not needed, it's just to get a result QUICKLY

x1.SRestore(frate=24.975,dclip=x2)
return(last)

and got this: KLICK (http://www.mediafire.com/file/jk2xokmmqvy/SRestoreTest.avi) (MediaFire, 5 MB)


Perhaps it's because I'm more asleep than awake, but ... I don't see a problem?

I don't either :thanks:
didn't think of using the donwsizing for analyzing, seems it helps a lot :cool:

edit : tried same part on the movie with same script, doesn't work at all at home, I guess there's a problem with my install somewhere (upgraded to lucid yesterday, maybe the wine install + avisynth 258 goes berzerk for some reason. )
At least now I know it's a local problem, nothing to do with srestore or even the source...

...oh, and just switched from dgavcdec/avcsource to ffms2/videosource and it works like a treat now. maybe I'll have to reinstall dgavcdec (or just stick with ffmpegsource2)

[levi]

Any of you guys able to help recommend some settings for this one?

http://www.mediafire.com/download.php?ej40e0zzjtn

[Great Dragon]

@live, I believe this:
s=DGDecode_mpeg2source("my2.d2v")
s.yadifmod(mode=1, edeint=s.nnedi2(-2))
srestore(23.976)
will do the job pretty well.

x1.SRestore(frate=24.975,dclip=x2)

I wonder how you determine that such a strange frame rate is right?
Maybe for this (33 MB) (http://a.imagehost.org/download/0963/xmen_sample2_pal_demuxed) kind of video I need custom frame rate too? I've used srestore() default for it and frames rate become 11.988 and still there are some blended frames left. Then I need to blend resulting video to 23.976 to remove some jerkiness on horizontal/vertical panning.
Can you or anyone take a look at it and give some advices?

[MOmonster]

@Great Dragon
It is not the exact framerate but 11.25fps is close to it. Use this:srestore(frate=11.25,speed=1)
or for example this:
last.srestore(speed=-1,omode=3,mode=-1).blendbob().tdecimate(mode=1,cycle=10)

You can't remove the left blendings. Some parts are just overblended.

[Great Dragon]

@MOmonster, it kills to much good frames for me.
I've done some more test and found this
srestore(frate=21.976, speed=-25, thresh=12, mode=3)
preserve good frames better especially in intro. There are still blended frames left but not so much.

Here are other samples
xmen_sample3_pal_demuxed.m2v (20 MB) (http://j.imagehost.org/download/0110/xmen_sample3_pal_demuxed) xmen_sample4_pal_demuxed.m2v (9.6 MB) (http://a.imagehost.org/download/0389/xmen_sample4_pal_demuxed)

[txporter]

There has been some discussion (starting here (http://forum.doom9.org/showthread.php?p=1413769#post1413769)) in the TGMC thread about the proper way to rebuild PAL source from PAL=>NTSC DVDs while still using TGMC for the actual deinterlace. Initially, I had been simply using a TGMC bob'd stream to feed srestore. der_groschi brought it to my attention that this was not proper, so I did a bit of searching and found some threads on exactly this subject. It was suggested to use a detection clip that was not processed through TGMC to pick out the blends.

I am trying to figure out how to determine which bobber to use for the dclip. I have attempted a few difference approaches but still see different frames chosen depending on what I use for the dclip. I have a copy of my source clip + 2 different stacked videos here (http://txporter.tonidoid.com:10001/app/websharepro/share/dclip/) to compare the various methods.

key for dclip:
bob1: bob(0,0.5)
bob3 (suggestion from Didee in another thread, I couldn't get reduceflicker=false to work though): bob1 = o.bob(0,0.5)
bob2 = bob1.clense(reduceflicker=false).merge(bob1,0.5)
bob3 = bob2.mt_adddiff(mt_makediff(bob2,bob2.clense(reduceflicker=false),U=3,V=3),U=3,V=3)
yadif: yadif(mode=1, order=1)
tgmc: no dclip, fed with tgmc bob'd video
bob1.crop: same as bob1 + crop(8,16,-8,-24) [based on Srestore wiki (http://avisynth.org/mediawiki/Srestore#Examples)]
bob3.crop: same as bob3 + crop(8,16,-8,-24)
yadif.crop: same as yadif + crop(8,16,-8,-24)

stack.mp4 = stack of bob1 vs bob3 vs yadif vs tgmc + diffs
stack2.mp4 = stack of various bobs vs bobs+crop + diffs

I have included the frame number on each frame so I can see which frame was chosen by Srestore. You can see that various approaches cause different frames to be chosen. It seems that very dark frames are the most likely to be different from dclip to dclip choice. I believe that the only case where duplicate frames were chosen were in the opening credits where the yadif.cropped dclip chosen frame 73 twice. I don't see any case where a blended frame is left over.

Is this something that should be adjusted using the speed parameter in srestore or does it really not matter over much which frame is chosen as long as blends are gone? One of the arguments against using a tgmc stream for srestore was that it was felt that srestore would not be able to pick out the blends. Based on testing with this clip, it seems like srestore does a good job even with a tgmc stream (but it does take longer).

[Didée]

[clense]
I couldn't get reduceflicker=false to work though)
This implies that

a) most probably you're using the latest v1.0 release of the RemoveGrain package (Amongst others, Clense works differently in 1.0final than in previous versions. I'm always recomending the pre-1.0 build - the official 1.0 comes with additional modes that no-one uses and that aren't even documented, and in exchange, it breaks functionality of previously-working scripts. As you now have experienced.)

and more importantly

b) if you didn't get that part working, then "my proposed" script is not doing what it's supposed to do (simply removing the reduceflicker parameter does not give you the processing method that I intended to be used), and therefore there is no point in evaluating the end result. You broke the chain.

[txporter]

This implies that

a) most probably you're using the latest v1.0 release of the RemoveGrain package (Amongst others, Clense works differently in 1.0final than in previous versions. I'm always recomending the pre-1.0 build - the official 1.0 comes with additional modes that no-one uses and that aren't even documented, and in exchange, it breaks functionality of previously-working scripts. As you now have experienced.)

and more importantly

b) if you didn't get that part working, then "my proposed" script is not doing what it's supposed to do (simply removing the reduceflicker parameter does not give you the processing method that I intended to be used), and therefore there is no point in evaluating the end result. You broke the chain.

Yeah, I understand that I broke your chain. ;) Was still attempting what I could get to work to see if it made a difference.

Here is what I did to test out your script with clense(reduceflicker=false). I went to the RemoveGrain wiki (http://avisynth.org/mediawiki/Removegrain) and downloaded the v1.0 pre-release Removegrain plugin (this is the one that I was already using). It comes with 12 plugins: RemoveGrain[S/SSE2/SSE3] + Repair & RSharpen of the same flavors. Using these plugins gives me an error that no function Clense is found. The only thing that I found that didn't give me an error about Clense was a RemoveGrainT package that kassandro posted on another forum (can't find link atm).

I just found an old RemoveGrain.dll from 2005 (pre-release were 2007 builds) that seems to have the Clense function and allows Reduceflicker=false (I think that is the v0.9 build). Will see how that goes.

[levi]

Looking for recommendations on this one. It seems to be 11-12 fps?

http://www.zshare.net/download/787671864e463c69/

:thanks:

[EuropeanMan]

HI guys, I'm back...with another problem

yadif - mode 3
srestore - frate 25

applied on my dvd9...worked fine EXCEPT for one scene
here is a demuxed m2v file...the problem starts on frame 200 (to 339) for that SCENE only...the guy running up the stairs...the rest is fine...and ONLY that very small section

the problem i have is that after my encode in megui to avi (XviD), i checked the whole avi...and THIS PARTICULAR scene, i still have ghosting here with him
running up the stairs and am bewildered as to the WHY.

i looked at the untouched VOB in vdub...it seems as if the interlacement pattern is different. how do i tackle JUST this portion to output @ 25fps.

thanks in advance

sample (11MB, no audio): http://www.sendspace.com/file/m78abq

[manono]

how do i tackle JUST this portion to output @ 25fps.
Same as the rest and put up with the blends. As you said, the guy on the steps was done differently and it has way too many blended fields for anything to be able to unblend it properly.

[Chainmax]

I apologize for this newbie question, but being the first time I use SRestore I want to make sure. Would this:

d = last.bob(-0.2,0.6).reduceflicker(strength=1)
Interp = nnedi3()
TDeint(type=1,edeint=Interp,emask=TMM())
SRestore(frate=23.976,omode="pp3",speed=-25,dclip=d)

be a correct call?

[EuropeanMan]

Interp=nnedi3()
TDeint(type=1edeint=Interp,emask=TMM())


=


Tdeint(type=1,edeint=nnedi3(),emaskTMM())


If you're not calling Interp again, no use setting a useless variable.

[Chainmax]

Good point, blame it on a brain fart. :o

[Great Dragon]

Hi guys.
I can't manage to deal with some strange NTSC pattern.
I've tried many different IVTC methods I know and none of them can't give me a good result.
Basically it's because of chroma fields has a different pattern than luma.
I've tried to use this updated Chroma_rep() (http://forum.doom9.org/showpost.php?p=1268074&postcount=367) but it just does nothing for me and I'm getting picture with combing artifacts or just blended frames.
Here is a sample. (http://www.mediafire.com/?fo65hl21mv87d15)
Thanks in advance.

[nautilus7]

Hi, i'm trying to restore original 25p fps framerate of a BBC documentary. Sample (http://www.mediafire.com/?0mau14anuakoirt) is from blu-ray which is conveterd to 29,97i fps with frame blending. I have tried RePAL and ExBlend as well as Srestore, but all of them leave some combing leftovers in the restored frames. Any kind of help will be welcome.

[Didée]

An example of the combing leftovers, and also the script you used?

RePAL and SRestore both rely on the bobbed stream you feed them with. (Not sure about ExBlend, haven't tried it yet.) It is impossible that any combing is created by these plugins. If at all, the combing is a result of the bob filter that has been used before.

[nautilus7]

Maybe i didn't said it with the right term but this is what i get:

DGSource("e01.dgi")
TDeint(mode=1)
Srestore(speed=-25, mode=-2)]

http://thumbnails8.imagebam.com/10234/d50292102333190.jpg (http://www.imagebam.com/image/d50292102333190)

[Didée]

Ah well, okay. However, that is not combing. It's the source that is already broken.

(No idea if that can be fixed.)

[nautilus7]

Didee, thanks for your answer, although, it seems that those artifacts aren't present in the source. ExBlend gives me artifacts in other places, different than those using srestore or repal.

DGSource("e01.dgi")
TDeint(mode=0, order=-1)
Exblend(mode=0, pal=1, ithresh=100, show=2)
TDecimate(mode=0, cycleR=1, cycle=6, rate=24.975, hybrid=0)
AssumeFPS(25000, 1000)

http://thumbnails29.imagebam.com/10242/5b8463102418433.jpg (http://www.imagebam.com/image/5b8463102418433)

[Didée]

Looking at:

DGSource("e01.dgi")
bob()
This simple script does not introduce any artifacts.

(bob-) Frame #10: blended
(bob-) Frame #11: blended
(bob-) Frame #12: blended

3 consecutive blended fields is what I consider "broken". Such should not happen with a "clean" blend-normconversion.

None of the available blend-restaurers is able to clean this up sucessfully.


edit: actually, it's the chroma channels that exhibit the additional blending. Seems someone didn't care for the particularities of interlaced YV12 during (or prior to) the conversion process.

See for yourself:

bob()
reduceby2()
stackvertical(last,last.greyscale())

Frames 11+12 of this script show the same frame (a dup from the conversion), but both frames show blended chroma. In effect, there is no clean chroma channel available for this frame.

It is likely that this problem is present throughout the complete source.

Welcome in the world of High Definition, where you can watch decade-old artifacts in unseen clarity! :D :(

[nautilus7]

I see. So, the source is messed up. Thanks for looking into it.