Decomb 3.1 & PAL

[Ripe73]

HI!
I have a movie Dungeons & Dragons PAL Interlaced i tried swap field in DVD2AVI but it did not help so i want to try out Decomb and what script should i put in to the *.avs
decimate(2) gives me 12.5fps:D i dont want to change the framrate so i tried with Telecide() only and it looks okay when i scrool in V-dub but is that right?
Thanks for help

[Guest]

@Ripe73

Yes, Sir, you have it right. You need Decimate only when you want to reduce the framerate. Typically, PAL does not need that, although there are some cases where an extra frame is added for every 24 frames; in this case you'd want Decimate(cycle=25). If you see stuttering about once per second, you can try it. Or you can step through the processed video and see if you have one duplicate frame out of every 25 frames.

Your script can be simply:

Telecide()

and follow it with

Decimate(cycle=25)

...if required.

[Ripe73]

Thanks!!
I used only Telecide() but the result is little jerky and i checked the movie in V-dub and some frames are double sometimes so i started new firstpass with Decimate(cycle=25) and the fps in V-dub is very unstable 7-15fps is that normal?
Thanks

[Guest]

Yes, it is normal when Decimate(25) is applied. How is the final result?

[Ripe73]

HI!

The result is very good with XviD,Decomb and Ogg:D
Thanks

[wasbeer666]

Greetings fellow encoders,

As you see, this is my first post. I subscribed merely to give you (neuron) my respect for creating a filter which is in such a need.:)

I used the decomb package on both regio codes.
On NTSC it was a peace a cake with the Telecide (chroma=true) and decimate (5) it produced beautiful results> real sharp encodes. Also, it took me 5 minutes to apply this filter in the AVS script whereas I never IVTC'ed before in my life. Concluding that your manual is very easy to understand.

Mostly however I use PAL dvd's anime. I also used the decomb with the
Telecide (chroma=true) and decimate (25) settings. Probably this decimate wasn't needed but no jerkiness occurred. The filter produced again really sharp encodes (I read that with previous releases blurrings sometimes appeared). It's equal to Greedy or even surpasses it qua ghosting and artefacts. However still a significant amount of ghosting and ghosting artefacts occur.

The standard threshold was enough to get rid of all the interlace, besides the fades in/outs, I think this 's because the detector works on luma change which is very little in these motion sequences (screen goes black), correct me if I wrong:)

Hope you can do something usefull with these observations.
PS: are you planning on further development with pal deinterlace?

With regards,

[Guest]

@wasbeer66

Thank you for the feedback and kind words.

>However still a significant amount of ghosting and
>ghosting artefacts occur.

Can you be a little more precise here for me? Do you mean you are getting blended frames? If so, have you determined whether your source has blended fields, and if so, what their distribution is? There is a lot of PAL material that is converted from NTSC and often very simple conversions are done that result in blended fields. If these blended fields are limited to either the top or bottom fields, then Telecide 3.11 (not released yet!) can deal with them. If the blended fields appear in both top and bottom, then things are more difficult. The Telecide for VirtualDub can handle that and I plan to add this feature to Decomb's Telecide very soon, as many people really object to these blends (which I find tolerable when played at normal speed). This should help out with your PAL videos. Beyond that, I'm open to suggestions. What do you think is needed?

If by ghosting you are referring to the blend done by the deinterlacing, then you can try turning off the blend option.

[wasbeer666]

I have taken it into consideration and wil perform it in my next rip. Expect results within days:).

[trbarry]

If these blended fields are limited to either the top or bottom fields, then Telecide 3.11 (not released yet!) can deal with them. If the blended fields appear in both top and bottom, then things are more difficult. The Telecide for VirtualDub can handle that and I plan to add this feature to Decomb's Telecide very soon, as many people really object to these blends (which I find tolerable when played at normal speed).

Hey, nifty! How do you recognize blended fields?

I've never made the attempt but there is sort of a major problem. If you hit pause it will stop on a key frame, which tends to also stop on scene changes. And this is where the stupid blended fields show up so you have to handle all the questions about them. ;)

- Tom

[wasbeer666]

Thought you can do this, with the separate fields command in the avs script and then when loaded in nandub/virtualdub you can scroll through the file with the arrow keys. Or am I wrong:)?

[Guest]

@trbarry


Hey, nifty! How do you recognize blended fields?
I can't recognize them standing alone. But when matched with either of the opposite fields from the previous and following frames (that made up the blend), you'll get combing that can be detected. So, if I am matching the top field of the current frame to the bottom candidates (I use 3 candidates; you use only 2), if I cannot find a match that does not have combing, I will enter a second phase where I try to match the bottom field to the three top candidates. I say "will" because it is not yet implemented in Decomb, but works well in the VirtualDub version of Telecide. As long as both top and bottom are not blends, a good progressive can usually be found.

Edit: It just occurred to me that in the second phase I do not need to compare again for the candidate in the current frame, as that match would have already been tested in the first phase. Just didn't want anyone supposing I couldn't see that. :)

[trbarry]

Ok, prior to your previous comments I hadn't realized that usually only one field was blended. I wonder why this is. ?? What sort of telecine, rendering, or mangling process would decide to do it this way? There is probably a good reason for it.

I usually try to figure out what the other guy was up to before trying to put things back together.

I notice this even with US HDTV captures that presumably never had to change frame rate, so maybe it is a part of some normal upconversion process.

- Tom

[wasbeer666]

@neuron2

Currently I'm working on a new anime pal dvd rip.
I would like to show you what I exactly meant by the "artefacts"

Unfortunately, most of the terms you experts use on the deinterlacing process I don't understand.

So please bear with me, I will try to take you through the encode process with screenshots (I these pics ain't sufficient, please request what you exactly want instead).


screenshot 1 (http://home01.wxs.nl/~wassi129/example1.jpg)
This is the original vob loaded in dvd2avi, the interlace as it appears naturally.

screenshot2 (http://home01.wxs.nl/~wassi129/example2.jpg)
Another picture of the same vobfile, still orginal interlace.

screenshot3 (http://home01.wxs.nl/~wassi129/example3.jpg)
The d2v file opened in gknot, still orginal interlace> these are some of the artefacts I was mentioning.

screenshot4 (http://home01.wxs.nl/~wassi129/example5.jpg)
This is the ghosting I was talking about(opened in nandub), applied the following script:
#
# Created with Gordian Knot
#
# http://thewef.nav.to
#
# PLUGINS
# get them from http://users.win.be/dividee
LoadPlugin("C:\PROGRA~1\AVISYNTH\mpeg2dec.dll")
LoadPlugin("C:\PROGRA~1\AVISYNTH\decomb.dll")
#LoadPlugin("C:\PROGRA~1\AVISYNTH\InverseTelecine.dll")
#LoadPlugin("C:\PROGRA~1\AVISYNTH\Avisynth_Spatial.dll")
#LoadPlugin("C:\PROGRA~1\AVISYNTH\GreedyHMA.dll")
#LoadPlugin("C:\WINDOWS\SYSTEM\vobsub.dll")
#
# SOURCE
mpeg2source("C:\DVD Ripping\imperfect cell - discovery\His name is cell.d2v")
#
# TRIM
#trim(startframe,endframe)
#
# IVTC
#InverseTelecine(40,10,15)
# or use
#GreedyHMA(1,0,4,0,0,0,0,0)
#
Telecide(chroma=true,blend=false)
# CROPPING
crop(15,6,691,566)
#
# DEINTERLACING
#SmartDeinterlace(2,15,true,true,true)
# or use
#VerticalReduceBy2
# or maybe
#GreedyHMA(1,0,0,0,0,0,0,0)
#
# DENOISING: choose one combination (or none)
# 1) little noise (fast)
#TemporalSmoother(2,1)
#
# 2) medium noise (slow)
#SpatialSoftenMMX(1,4,6,false,false,4,4,6,8)
#TemporalSmoother(2)
#
# 3) heavy noise (very slow, you have been warned)
#SpatialSoftenMMX(2,4,6,false,false,4,4,6,8)
#TemporalSmoother(3)
#SpatialSoftenMMX(1,4,6,false,false,4,4,6,8)
#
# SUBTITLES
#VobSub("FileName")
#
# RESIZING
BicubicResize(512,384,0,0.75)
#
# BORDERS
#AddBorders(left,top,right,bottom)
#
# COMPRESSIBILITY CHECK
# !Snip Size has to be 13 for use in GKnot!
#SelectRangeEvery(260,13)
#
# FOOL CCEnc
#ResampleAudio(44100)
------------------------------------------------------------
screenshot5 (http://home01.wxs.nl/~wassi129/example6.jpg)
Other deinterlace artefacts after applying the previous script (the lines are remains from a previous frame).

screenshot6 (http://home01.wxs.nl/~wassi129/example7.jpg)
This is the same image but with no deinterlace filters applied at all.
This is where I stand corrected:), because if no deinterlace is applied then wouldn't you theoretically end up with the same file as the d2v file? , In the above image (and the rest of the file) deinterlace IS used somewhere but I haven't figured out yet where

:sly:

PS: When encoding a NTSC dvd from the same anime, all that is needed is IVTC (to framerate 23.976, don't pin me down on this one as it is from memory:)), thereafter no ghosting and no other artefacts were present. So concluding this is major caused by transverting a NTSC dvd to PAL? (correct me if I'm wrong)

With regards,
Guido

[Guest]

I see what are probably blended fields causing your ghosting. To comment on the rest, I'd have to have a VOB file. Can you make one available?

[meleth]

Out of curiosity. Wouldn't it be possible to get better results if you used DoubleWeave and then used Decimate(2) Decimate(5) ?

[wasbeer666]

@neuron2

I uploaded a chapter from the vob (would be kinda overkill to up the entire vob:)).
You can find it at ftp>> 137.120.25.194, inlog: neuron2/neuron2.

[Guest]

Looking at the clip...
If you save the AVI out of DVD2AVI with no field processing, and then use SeparateFields() to view the original clip fields, you can easily see that there are blended fields; these will result in blended frames after processing. The ugly remnants you mentioned are also already there in the source. There is currently nothing you can do to clean this up, although an imminent release of Decomb will handle blended fields better in some circumstances.

[trbarry]

This blended fields crap is starting to become more irksome. It seems there must be some better more general solution. But maybe you've already got it. Dunno.

- Tom

[meleth]

Well as i figured it anime contains so much blended field since it's usually originally shot at 20 or sometimes 15fps and then they blend the fields in order to get a smooth transition from frame to frame.

I was thinking that perhaps the best way to solve this would be to remove all blended frames and then develop a new directshow filter that would do the blending instead.

[wasbeer666]

@ripe73

That maybe the case,

However have you taken my note into consideration from my previous post? If I encode the same material from NTSC then there are very few ,perhaps none, blended fields. Only operation I have to perform on it is IVTC. But the PAL material is hell, see my previous post.

[Guest]

@wasbeer666

The reason only the PAL version is messed up is because it was converted from the NTSC version. And whenever the frame rate is changed, as it is when going from NTSC to PAL, blending is used.

[wasbeer666]

Thanx, that was some need-to-know information:).

I read all posts about the IVTC'thingies that are around now such as decomb and IVTC4. But I must admit: the more I read, the more confused I get:). However, I'm getting quite intriged by these filters and their mechanism, currently I'am self-educating myself about the deinterlace subject so please don't be bothered when I post a nOObie question, it's just the eagerness to learn more:)

at your service,
Guido

By the way, If I can help with providing some PAL anime material for research, don't hesitate to ask.

[frankio]

I was encoding Pink Floyd's The Wall PAL and found it was interlaced.
Mainly I rip PAL dvd and so this was my first time I really fell in trouble with interlace artifacts.
I've read everything about interlaced and/or telecined material (thanks to Doom9, Nicky's Guide and Luke's Video) and I've found so far Decomb working best.
Following neuron2 guidelines I can see blended fields in both top and bottom (I thought it was "perverse" telecined, right?).

Neither Telecide,GreedyHMA nor Decomb WITHOUT post-processing can get rid of interlacing artifacts.

On the other hand using Deinterlace filters removed a lot of them but the real problem are scene changes (no cross-fading) which takes about 3 frames per each (it's a bit annoying and encoder can't properly put I-frames).
The only one that can handle correctly this situation was Decomb (quite perfect SC) but I can see duplicated frames (1 in a second). So I decimate:

Telecide()
Decimate(cycle=25)

but stuttering is always there once in a second (I've checked: number of frames was decimated properly).
On the other hand using only Deinterlace filter (Smart Deinterlace or FieldDeinterlace()) I have no stuttering.

So the question is: I've missed something or under some particular circumstances Decomb gives this weird result?

I've read in the above posts that Telecide does a good job in this cases, but I can't confirm that, Decomb 3.3 worked better. What else can I try? Maybe TMPGEnc IVTC?

Hoping in a deinterlaced future, greets and many thanks to neuron2 for his great piece of software and to everybody contributed in developing deinterlacing and IVTCing method.

[manono]

Hi frankio-

So the question is: I've missed something or under some particular circumstances Decomb gives this weird result?

Yes you missed something, but it's not something all that obvious. I didn't realize it myself until neuron2 set me straight. Your Pink Floyd movie is pure interlace. It has no progressive frames to reconstruct, which is what Telecide() is trying to do. And so, it screws up and makes the movie look a bit jerky-stuttery. Your only choice in such cases is to use FieldDeinterlace() and then the Decimate(cycle=25) when you see the duplicate frames.

It's to your credit that you figured out the answer yourself. I had to get help from the Boss ;)

[frankio]

Hi manono, glad to "hear" a senior opinion ;)

Thus my supposition was wrong (there weren't any blended fields!) and I've to stuck with blended frame in scene changes (only Decomb avoided this).

Thanks for clarifying me.

[wasbeer666]

@neuron2

I was wondering,

I finished the rip from which i posted the example last week.
The result is excellent, almost no significant ghosting.
However, the image seems to stutter once in a second or so.
Is it possible to implement something in decomb that a interpolated frame is inserted or so which would prevent stuttering?

Greetz,
Guido

[Guest]

Before contemplating any new features, I'd want to analyse the problem and determine the root cause. Have you done that?

For example, are you getting a duplicate frame every 25 frames? If so, then you can apply Decimate(cycle=25).

[daxab]

@wasbeer666

By the way, please don't use IVTC4 on PAL material. It's strictly for NTSC telecined video. That is, for 24fps stuff that has been field-juggled to generate 30fps. I use it for U.S. TV captures and Region 1 anime.