Anime ivtc/deinterlace mixed telecine+interlaced help

[6ari8]

Hello,

I have an anime DVD that DGindex reports to be 100% interlaced Video.

However, there are sections that I think are telecined like this sample (http://www.solidfiles.com/d/075f601089/)
and some which are pure interlaced like this one (http://www.solidfiles.com/d/47a116c6f2/).

I tried QTGMC with this but in some panning scenes (like the one in the first sample) the static objects shake/wobble as it pans.

I tried TFM but some frames are still combed and the interlaced scenes are very blurry.

Am I stuck with either of those solutions or is there some better way to process a source like this?

I can provide more samples if needed.


Thanks!

[6ari8]

After some tries, I settled on this script:
TFM(d2v ="VTS_01_1.d2v",slow=2,pp=5,display=false,mode=1,mi=80,cthresh=9, hint=true, clip2=QTGMC( Preset="slow", SourceMatch=3, lossless=2,FPSDivisor=2, TR2=2))
TDecimate(mode = 1)

crop(4,0,-4,0)
Spline36Resize(720,480)

SMDegrain(tr=3,thSAD=500,contrasharp=true,refinemotion=true,lsb=true,lsb_out=false)
daa3mod()

Edge=MT_Edge("prewitt",ThY1=20,ThY2=80).RemoveGrain(17)
Mask=MT_Logic(Edge.MT_Expand().MT_Expand().MT_Expand().MT_Expand().MT_Expand(),Edge.MT_Inflate().MT_Inflate().MT_Inpand(),"xor")
MT_Merge(EdgeCleaner(strength=25, rep=True, rmode=17, smode=0, hot=False),Mask,Luma=True)

Toon(strength=0.5)
LSFmod(strength=85,defaults="slow")


Is there anyway to improve on this?

[creaothceann]

Readability, imo... (http://pastebin.com/NZCiz6MQ)

[6ari8]

Readability, imo... (http://pastebin.com/NZCiz6MQ)

Thanks!


I have one more question.

The opening is mostly pure interlaced (I can see movement on every field after using separatefields() ) so I want them to be 60p seperately.
QTGMC produces ghosting with all my tweaking. NNEDI3 doesn't but the picture quality is not as good as the former.

Is there a good solution for animated content for this task?

[huhn]

the sample is 3:2 maybe one odd part but that's not interlaced. can you post a sample from an interlaced part?

i guess this 30/24p mix used a lot for toho animations.

the sample shouldn't be deinterlaced.

[6ari8]

the sample is 3:2 maybe one odd part but that's not interlaced. can you post a sample from an interlaced part?

i guess this 30/24p mix used a lot for toho animations.

the sample shouldn't be deinterlaced.

The two samples are provided in the OP but I'll put it again here.
This is one of the interlaced samples:
interlaced (http://www.solidfiles.com/d/47a116c6f2/)

Most of the episodes are telecined with small portions that are pure interlaced like the sample above. So I'll IVTC the whole episode and settle with the jerky motion in the interlaced portions because I hate VFR and blending. :p

But I'm getting a lot of combing so I have to use Anti-aliasing after the IVTC to get rid of them so I was hoping for a better way.

The opening is also interlaced. I want to encode it on its own in 60p so I'm looking for a way to do that without ghosting.

[huhn]

if you don't want VFR than there is nothing i can help with but this is really strange.

[~ VEGETA ~]

I would refer to using YATTA. Section out the parts you think they are different than pure hard telecine ones, then deal with them separately. For example, make a whole section for the opening if it is just interlaced (no telecine), then put "no decimate" to that section. After that, you need to de-interlace it... you can do that by making a "preset" which holds your favorite deinterlacer, then assign it to that section.

to sum up, you would end with this:

1- the whole video: IVTC (tfm+tdecimate), you can manually tune everything (YATTA purpose of life).
2- the opening: has its own section with no tdecimate on it, it has a preset which applies a deinterlacer for it.

you can do many stuff in YATTA, which is I think the best tool for such sources. Today, all sources are nearly clean (TV .ts) and don't need much tuning. Finally, you will end up with a VFR clip (it is a MUST) with a timecodes file (convert it to v2 if it is not). encode and mux that tc file = final video = success desu!

I hope this was useful.

[~ VEGETA ~]

I forgot to mention that you will have lots of ghosting between sections (sections are scene changes detected by scxvid). I solve them manually in YATTA (takes some time but not much), but there is Vinverse filter for them. Add it after all ivtc and deinterlace things.

[real.finder]

The two samples are provided in the OP but I'll put it again here.
This is one of the interlaced samples:
interlaced (http://www.solidfiles.com/d/47a116c6f2/)

Most of the episodes are telecined with small portions that are pure interlaced like the sample above. So I'll IVTC the whole episode and settle with the jerky motion in the interlaced portions because I hate VFR and blending. :p

But I'm getting a lot of combing so I have to use Anti-aliasing after the IVTC to get rid of them so I was hoping for a better way.

The opening is also interlaced. I want to encode it on its own in 60p so I'm looking for a way to do that without ghosting.


AnimeIVTC(1,0,5, ediandnn=3, repwithtdeint=true, tfmm=0, micmatching=0, o3025cfr=true).TDecimate(1)
mcdaa3()


and for 60i op to 60p


yadifmod(mode=3, edeint=eedi3(-2,sclip=nnedi3(-2))).Repair(TDeint(1,emask=tmm(1)))
mcdaa3()

[6ari8]

I would refer to using YATTA. Section out the parts you think they are different than pure hard telecine ones, then deal with them separately. For example, make a whole section for the opening if it is just interlaced (no telecine), then put "no decimate" to that section. After that, you need to de-interlace it... you can do that by making a "preset" which holds your favorite deinterlacer, then assign it to that section.

to sum up, you would end with this:

1- the whole video: IVTC (tfm+tdecimate), you can manually tune everything (YATTA purpose of life).
2- the opening: has its own section with no tdecimate on it, it has a preset which applies a deinterlacer for it.

you can do many stuff in YATTA, which is I think the best tool for such sources. Today, all sources are nearly clean (TV .ts) and don't need much tuning. Finally, you will end up with a VFR clip (it is a MUST) with a timecodes file (convert it to v2 if it is not). encode and mux that tc file = final video = success desu!

I hope this was useful.

Thanks.
I did take a look at YATTA but it looked very complex and time consuming. I will try it again if all else fails.


AnimeIVTC(1,0,5, ediandnn=3, repwithtdeint=true, tfmm=0, micmatching=0, o3025cfr=true).TDecimate(1)
mcdaa3()


and for 60i op to 60p


yadifmod(mode=3, edeint=eedi3(-2,sclip=nnedi3(-2))).Repair(TDeint(1,emask=tmm(1)))
mcdaa3()


Can you please tell me what those parameters do? specifially, "repwithtdeint" and "o3025cfr".

Also, why did you choose these parameters "tfmm=0, micmatching=0"
?

Thank you.

[real.finder]

Can you please tell me what those parameters do? specifially, "repwithtdeint" and "o3025cfr".

Also, why did you choose these parameters "tfmm=0, micmatching=0"
?

Thank you.

see the "AnimeIVTC 2 mod" link in my signature

and regard "tfmm=0, micmatching=0" the reason here http://forum.doom9.org/showpost.php?p=1725899&postcount=48

anything else see the replies in AnimeIVTC 2 mod thread first

[6ari8]

see the "AnimeIVTC 2 mod" link in my signature

and regard "tfmm=0, micmatching=0" the reason here http://forum.doom9.org/showpost.php?p=1725899&postcount=48

anything else see the replies in AnimeIVTC 2 mod thread first

I read the thread but couldn't find anything about "o3025cfr" and "repwithtdeint" in it or here (http://avisynth.nl/index.php/AnimeIVTC)

But I tried it and it doesn't produce results better than this:

TFM(d2v ="VTS_01_1.d2v",slow=2,pp=5,display=false,mode=1,mi=80,cthresh=9, hint=true, clip2=QTGMC( Preset="slow", SourceMatch=3, lossless=2,FPSDivisor=2, TR2=2))
TDecimate(mode = 1)
daa3mod()

I'm still having issues with mouths being combed, and it becomes very apparent in face close-up scenes.

Here's a screenshot of it:

IVTC only:
http://s1.postimg.org/x9nem3bvf/Aivtc_mouth.jpg (http://postimg.org/image/x9nem3bvf/)

IVTC+mcdaa3():
http://s1.postimg.org/xaxcfidp7/Aivtc_mcdaa_mouth.jpg (http://postimg.org/image/xaxcfidp7/)

IVTC+daa3()
http://s1.postimg.org/k47wfzhzv/Aivtc_daa3_mouth.jpg (http://postimg.org/image/k47wfzhzv/)

Only when I do IVTC+daa3()+daa3()+daa3() does it become slightly better:
http://s1.postimg.org/4sy3fdknv/Aivtc_daa3_daa3_daa3_mouth.jpg (http://postimg.org/image/4sy3fdknv/)

but this causes issues (curves become deformed) in many other scenes like this:

ivtc:
http://s13.postimg.org/bx2v9l83n/ivtc.png (http://postimg.org/image/bx2v9l83n/)

ivtc+daa3+daa3+daa3:
http://s13.postimg.org/b6a53t5qb/ivtc_daa3_daa3_daa3.png (http://postimg.org/image/b6a53t5qb/)


Is there something that can be done with TFM's parameters or an AA better suited for this type of combs?

[foxyshadis]

Have you tried playing with Vinverse? It exists specifically to get rid of a little residual combing where the match was otherwise good. YATTA was created because older anime was often butchered during mastering (and some anime encoders are very OCD), you have to decide how badly you want it to be perfect, because it won't be possible to do it all automatically. Throwing Vinverse at it as a final step is good enough for me.

[Desbreko]

daa actually works better for that than Vinverse. I recommend upscaling the height by 1.5x before applying it and then downscaling again afterward, though, or else it'll cause more blurring than necessary. Or just use xaa(ssh=1.5, mode="daa") to have it do that automatically.

[real.finder]

I read the thread but couldn't find anything about "o3025cfr" and "repwithtdeint" in it or here (http://avisynth.nl/index.php/AnimeIVTC)

But I tried it and it doesn't produce results better than this:

TFM(d2v ="VTS_01_1.d2v",slow=2,pp=5,display=false,mode=1,mi=80,cthresh=9, hint=true, clip2=QTGMC( Preset="slow", SourceMatch=3, lossless=2,FPSDivisor=2, TR2=2))
TDecimate(mode = 1)
daa3mod()

I'm still having issues with mouths being combed, and it becomes very apparent in face close-up scenes.

Here's a screenshot of it:

IVTC only:
http://s1.postimg.org/x9nem3bvf/Aivtc_mouth.jpg (http://postimg.org/image/x9nem3bvf/)

IVTC+mcdaa3():
http://s1.postimg.org/xaxcfidp7/Aivtc_mcdaa_mouth.jpg (http://postimg.org/image/xaxcfidp7/)

IVTC+daa3()
http://s1.postimg.org/k47wfzhzv/Aivtc_daa3_mouth.jpg (http://postimg.org/image/k47wfzhzv/)

Only when I do IVTC+daa3()+daa3()+daa3() does it become slightly better:
http://s1.postimg.org/4sy3fdknv/Aivtc_daa3_daa3_daa3_mouth.jpg (http://postimg.org/image/4sy3fdknv/)

but this causes issues (curves become deformed) in many other scenes like this:

ivtc:
http://s13.postimg.org/bx2v9l83n/ivtc.png (http://postimg.org/image/bx2v9l83n/)

ivtc+daa3+daa3+daa3:
http://s13.postimg.org/b6a53t5qb/ivtc_daa3_daa3_daa3.png (http://postimg.org/image/b6a53t5qb/)


Is there something that can be done with TFM's parameters or an AA better suited for this type of combs?

from http://forum.doom9.org/showthread.php?p=1673955#post1673955

add o3025cfr to out 25 in PAL and 30 in NTSC, and it useful to avoid blended with omode=1


it simply mean no Decimate in http://forum.doom9.org/showpost.php?p=1746339&postcount=10 because original AnimeIVTC don't use mode 1 in TDecimate internally

from http://forum.doom9.org/showthread.php?p=1673955#post1673955 again

add yadifrepwithtdeint to use yadifmod with less Artefacts


yadifrepwithtdeint = yadif (yadifmod) repair with tdeint

then


# yadifrepwithtdeint now repwithtdeint and work with bbob 3 and 4 and precision 5
# add crepwithtdeint for cbob 3 and 4


http://avisynth.nl/index.php/AnimeIVTC now outdated

so, you didn't look well

good luck

[LigH]

Sorry 4 Necro ;) ... just wanted to share another cruel sample of a failed Blu-ray production. I don't know the title, just saw a sample (http://www.mediafire.com/download/tac6994nbeoawqx/00006.mkv) and discovered an advanced level of stupid in their video studio.

This one has been processed with a motion-compensating norm converter to blended interlacing... here are just 5 bobbed fields to facepalm over: Notice a) the shift in the window behind the guy, and b) the blend-in of the car in the left corner. On top, this scene has layered animation, the car has rather fluid motion, in contrast to the stop-motion of the guy.

http://www.ligh.de/pics/00006.gif

If you have such material, think well if you really want to even try to process it. You would possibly have to start with removing all blended or potentially motion-compensated fields...