TDeint and TIVTC

[tritical]

@DarkNite
You wouldn't happen to have a clip that results in different mode 5 output (latest version vs. older versions) would you? I've been trying to duplicate the issue but haven't been successful. I pm'd Asrial a couple days ago about getting a clip, but I don't think he visits very often.

@foxyshadis
Thanks for the report. I'll try to duplicate it.

[foxyshadis]

Never mind, I just didn't understand the docs well enough. I didn't realize that it only uses clip2 for combed frames, so I was checking the wrong frames, but using a much lower cthresh for testing forced it to work. Sorry!

[DarkNite]

I don't recall exactly what portion of which source it was atm (I have 6 projects going on now), but for you I will attempt to hunt it down in my free time.

[tritical]

@DarkNite
Thanks, though if you don't remember where it was don't worry about it too much... I'm sure you have better things to do with your free time .

@All
I have managed to fix the mode 2 problems. The drifting on long sequences that Xesdeeni reported was caused by having a '>' instead of a '>='. The crash ChiDragon reported was caused by a bug in the sorting algorithm that was used for very large cycles (could only happen when using mode 2 with an input file and non-standard framerates). I also made some minor improvements.

I'm still trying to duplicate the mode 5 problem. I've been comparing TIVTC v1.0 RC1 vs v1.0.1 on a few hybrid sources and have yet to see any major differences. Although, I did find a bug in tfm's cubic pping. The handling of the very last line (for field=1) and very first line (for field=0) has been slightly messed up since ~ April, 06.

[Terranigma]

tritical, any new news on EEDI's successor? How is it progressing?

[Livesms]

Originally Posted by foxyshadis

I don't think the TFM clip2 parameter is working as it should. I've been using:

clipt=TDeint(field=1,emask=TMM(field=1),edeint=separatefields.selecteven.eedi2(field=1))
clipb=TDeint(field=0,emask=TMM(field=0),edeint=separatefields.selectodd.eedi2(field=0))
TFM(clip2=clipt,pp=4)

But the frames I get back are exactly the same as with TFM(), and very different from clipt. I haven't had any luck figuring out why.

Can you explain this script Please

[tritical]

@Livesms
clipt/clipb are just tdeint deinterlaced versions of the original source... clipt keeping top fields and clipb keeping bottom fields. In both cases TMM() is being used to build the motion mask instead of tdeint's built in motion masking, and eedi2 is being used to interpolate pixels instead of one of tdeint's built in interpolation methods. One of those two clips is then being used by tfm's clip2 parameter (in this example it is clipt). That means that when tfm detects a combed frame combing out of field matching it requests that frame from clipt instead of deinterlacing it itself. When pp is set < 5, the frame from clip 2 will be delivered as is. If pp is >= 5 then tfm will still apply its own motion masking.

@Terranigma
It's doing quite well. I haven't commented on it for a while because I'm planning on using the idea as part of my master's thesis, which means that a public binary is still a ways off.

[foxyshadis]

Maaaaaaaaan, that would be nice to see indeed. ;_;

The reason I used two different tdeints, if you're curious, was because I had a filter in between that would swap in frames from the other (a quick hack of fizick's badframes). It was an old VHS capture and had an annoying amount of partially smeared lines that could be mostly fixed by interpolating over the other field. I got sick of finding bad frames pretty quickly, though.

[Terranigma]

tritical, thanks for keeping me informed. I can only imagine the possibilities this new filter will be able to achieve (Specifically when it comes to aliasing/deinterlacing).

[audyovydeo]

@tritical

This is just a word of appreciation as I really like TDeint. I've read around the forums and seen some speed comparisons with other deinterlacers. I dont do comparative testing as it's precious time taken off real encoding.
I find TDeint is a real workhorse : I have two avs scripts each TDeinting two DV_PAL.avi files and layering them, both scripts fed into Adobe Premiere for yet another transparency, and it works like a beauty (on a T5200 cpu). That means very good workload scaling, which to me is more important than absolute speed.

So, for me, every minute you spend on multithread optimization is wecome.

cheers
audyovydeo

[zdark]

I am having certain lag in slow scenes. If I to use...telecide() for But its filter it is better, I changed setting.

source original: DVD NTSC 720x480
my script for desintrelaced

interp = separatefields().eedi2(field=-2)
deinted = tdeint(mode=2,edeint=interp,type=3)
TFM(clip2=deinted,order=-1,mode=5,pp=7,field=-1,slow=2)

now its fine?

is for anime Vision of escaflowne

You can say me because she is giving lag?

http://rapidshare.com/files/33066521/teste2.mp4.html

[Leak]

Originally Posted by zdark

source original: DVD NTSC 720x480
[...]
interp = separatefields().eedi2(field=-2)
deinted = tdeint(mode=2,edeint=interp,type=3)
TFM(clip2=deinted,order=-1,mode=5,pp=7,field=-1,slow=2)

Since I very much doubt Escaflowne was animated at 30 FPS - where's your call to TDecimate?

Your current script will result in eveyr fifth frame being a duplicate.

Also, what's the deal with "fansubbing" something that's been released all over the world (at least Japan, America, Canada, UK, Germany and France) for years?

np: Legiac - Dide Skin (Mings Feaner)

[zdark]

I taste of you liven up old. I go to test adding the TDecimate!

[zdark]

leak my test.

no filter: http://rapidshare.com/files/33178778...ilter.mp4.html
filtered: http://rapidshare.com/files/33175940...ilter.mp4.html

now its fixed.

I used

edeintted = last.AssumeTFF().SeparateFields().SelectEven().EEDI2(field=-1)
deinted = TDeint(order=1,edeint=edeintted)
TFM(clip2=deinted,order=-1,mode=5,pp=7,field=-1,slow=2).tdecimate(hybrid=3)

thanks for help!

[DarkNite]

Also, what's the deal with "fansubbing" something that's been released all over the world (at least Japan, America, Canada, UK, Germany and France) for years?

Well, it may be that the translation was weak and erroneous (if it's anything like the English translation) or that it may not be available in his native language.

Or, it may just be that karaoke scripts are indeed that important to some people.

[tritical]

Haven't been able to reproduce the mode 5 problem, so I'm just gonna go ahead and release a new version with the fixes so far. [link removed], changes:

tdecimate:
   - Lots of fixes for mode 2 and non-standard framerates
   - Fixed problem in mode 0 decimation when using tfm hints and looking for 
        singleton match duplicates

tfm:
   - Fixed incorrect handling of the top/bottom lines in cubic post-processing

[Terranigma]

You know the drill

[zdark]

tritical thanks, good job!

[plugh]

Originally Posted by tritical

An exclusion band in tdecimate could be accomplished by using the clip2 parameter:

mpeg2source()
tfm()
saved = last
stackvertical(crop(),crop()) # set crops to remove lines containing subtitles
tdecimate(clip2 = saved)

In this case, tdecimate will calculate everything based on the input clip, but will output frames from clip2. The clip2 clip only has to have the same number of frames as the input. All other properties: height, width, and colorspace can be different. You could replace cropping/stackvertical with another method of removing the subtitles (set the subtitle area to black using overlay, etc...)..

If I may inject a question:

In a thread in the avisynth usage forum, I'm trying to use tdecimate on an animated film that was sped up slightly for broadcast. One of the problems I'm having is (I believe) related to vertical film jitter; the animation has its own dups, but frame to frame film jitter means they don't always align perfectly leading to higher difference metrics.

How can I get the dup matching code to search up/down a few lines to find the min difference metric? That is, given current frame "C" and previous frame "P" calc metrics between

lines C 1 thru N vs lines P 1 thru N as usual,
but also
lines C 1 thru N-1 vs lines P 2 thru N
lines C 1 thru N-2 vs lines P 3 thru N
and
lines C 2 thru N vs lines P 1 thru N-1
lines C 3 thru N vs lines P 1 thru N-2

and use the min metric for the dup check.

BTW, I also encountered a mode 2 rate dependant accvio with this movie. I see your latest version addresses this - I'll give it a whirl.

Thanks!

[ChiDragon]

plugh, I don't get why you'd want it to detect these animation duplicates as dups and decimate them. If anything the higher metrics caused by the jitter should help you only remove the telecine dups.