How to deal with combing and hybrid source with TFM+TDecimate?

[kenshin1991]

Hey guys,

I've spent hours trying to properly IVTC this hybrid footage however I'm still getting combing on the video sections (30fps). The film sections however are fine.

Any ideas? I've uploaded a segment of the original source material (3MB) to mediafire:

http://www.mediafire.com/download/xnisskexsv8dgic/Monster.m2v

MPEG2Source("Monster_CleanOP.d2v")

tfm(d2v="Monster_CleanOP.d2v",mode=5,pp=7,MI=64)
tdecimate(mode=1,hybrid=1)

[Sparktank]

I used DGdecNV and ran this:

tfm(order=1, mode=5, slow=2, pp=7, field=1).TDecimate(mode=1, hybrid=1)

The very first frame looks like this with the above:
http://i.imgur.com/sEQK6qL.png

A really odd thing:
DGdec (free version), it lists it as Progressive
DGdecNV, lists it as Interlaced (TFF).

[kenshin1991]

Thanks for the reply.

Sadly, I'm still getting artifacts despite using your script.

[Sparktank]

Thanks for the reply.

Sadly, I'm still getting artifacts despite using your script.

Seeing that DGdecNV is out of your budget or reason, I would suggest heading over to the new forums of the Neuron2.
Since, he's not a member here anymore, this issue might provoke an update over the outdated freeware DGdec.
Or provoke you to purchase a license for DGdecNV, provided you have an NVidia card.

Since the freeware version shows it as progressive while the licensed version lists it as interlaced, I'd say the free version is in dire need of updating.
It's quite old, while the NVidia version is always being updated.

[manolito]

Since the freeware version shows it as progressive while the licensed version lists it as interlaced, I'd say the free version is in dire need of updating.
It's quite old, while the NVidia version is always being updated.

Sorry I cannot reproduce your findings that the freeware version of DGIndex/DGDecode sees this M2V as progressive. I tried it with the current version 1.58 as well as with the very old version 1.45, and both report that this M2V file is Interlaced TFF.


Cheers
manolito

[LemMotlow]

I tried opening your sample with every utility I had, from DGIndex to LSMASH, and got the same results every time. Something wrong with your sample, which appears to have broken GOPs (if you yadif the first two original frames you'll get reverse movement on the first 4 deinterlaced frames regardless of how you specify the field order). There is telecine and frame duping going on, and the original creation appears to be done at 15 fps. I got rid of the combing, achieved 23.976fps with no combing, and some odd 3-frame dupes in the last scene, using the following (and using anything other than DGIndex produced the same results):

yadif(mode=1,order=1)
srestore()

I get similar results with srestore(frate=25), the main difference being the first few frames (looks as if cut on a B frame with no smart rendering). You'll get rid of more dupes at the end using frate(15).

The first frame in your sample might look the posted image on a deinterlacing player, but that's not the way it looks in editing. The combing is apparently from a previous frame.
http://forum.doom9.org/attachment.php?attachmentid=14525&stc=1&d=1418743914

Could probably play with it forever, but there's something wrong with your sample. Just playing it gets odd results. Maybe another sample cut on key frames, a little longer than this one?

[Sparktank]

Sorry I cannot reproduce your findings that the freeware version of DGIndex/DGDecode sees this M2V as progressive.

Hmmm. I use the one provided by Hank on HCenc website.
Since it's one of the most updated versions.

last scene

I've noticed, no matter what, the end scene comes in very choppy and abnormal, even for an anime.

[LemMotlow]

MediaInfo says the sample is interlaced. I've seen PAL->NTSC imports having 3 combed frame for every 5 that break down to 23.976. It could be telecine+interlace. Looks like a screwed-up import to me.

For 15fps originals, it's not unusual to see duplicate frames later.

[Sparktank]

Attachments Pending Approval

I'd suggest creating an account at something like Imgur or some other image hosting site to upload pictures.
It's faster than waiting for admin/mod approval.

Also, using a lossless format like PNG or BMP.
JPG will suffer some compression and introduce unwanted artifacts for what you are trying to achieve.

[foxyshadis]

This is pretty much an impossible sample; aside from framerate issues it was edited after telecine, destroying the pattern and leaving some fields completely missing. I'd just throw Vinverse at the residual combing; either dumb-bobbing or field-blending is about the best you can achieve without a miracle.

Are any of the other episode OPs any better? Maybe later discs are different.

[kenshin1991]

Looks like a screwed-up import to me.

The sample is from my retail R2J NTSC DVDs.


Are any of the other episode OPs any better? Maybe later discs are different.

Sadly they are all the same.

I've uploaded longer versions of the clips here:

7zipped: (~100MB)
http://www.mediafire.com/download/3vh6lcy72d5r6wc/Doom9.7z

or individually:

Clean OP (50sec | 50MB):
http://www.mediafire.com/watch/gg9u8zfrfir5gk1/MonsterOP.m2v (http://www.mediafire.com/download/gg9u8zfrfir5gk1/MonsterOP.m2v)

Opening from episode (50sec | 50MB):
http://www.mediafire.com/watch/12utbczkwj0x5em/MonsterEP1.m2v (http://www.mediafire.com/download/12utbczkwj0x5em/MonsterEP1.m2v)

Credits from episode (10sec | 13MB):
http://www.mediafire.com/watch/gfh2cs3v5fxkk6s/MonsterCredits.m2v (http://www.mediafire.com/download/gfh2cs3v5fxkk6s/MonsterCredits.m2v)

[colours]

I skimmed through the source (MonsterOP.m2v) and it seems that it's telecine with just one scene (frames 329-366) that was done at 48 fps. (I only looked at the first few scenes carefully so this assessment is probably not entirely accurate.)

The nicest way of dealing with this is probably to deinterlace that one scene to 60 fps, then hit it with TDecimate and splice it into the rest. The script below leaves the result as 48 fps, though it shouldn't be too hard to make it VFR if you so desire.

MPEG2Source("O:\MonsterOP.d2v", cpu=0)
ivtc = tfm(clip2=nnedi3()).tdecimate(mode=1)
deint = qtgmc().tdecimate(mode=1)
rfs(ivtc.selectevery(1,0,0),deint,"[526 586]")

[kenshin1991]

I skimmed through the source (MonsterOP.m2v) and it seems that it's telecine with just one scene (frames 329-366) that was done at 48 fps. (I only looked at the first few scenes carefully so this assessment is probably not entirely accurate.)

The nicest way of dealing with this is probably to deinterlace that one scene to 60 fps, then hit it with TDecimate and splice it into the rest. The script below leaves the result as 48 fps, though it shouldn't be too hard to make it VFR if you so desire.

MPEG2Source("O:\MonsterOP.d2v", cpu=0)
ivtc = tfm(clip2=nnedi3()).tdecimate(mode=1)
deint = qtgmc().tdecimate(mode=1)
rfs(ivtc.selectevery(1,0,0),deint,"[526 586]")

Thankyou.

Your script definitely fixed the issue with that particular scene. Do you know of a way to convert the 48fps down to 23.976 CFR via blending or am I stuck with creating a VFR file?

Also, I'm getting strange text artifacts using the same script on ''Opening from episode 1'' clip:

No Artifact:
http://i.imgur.com/TScnU4D.png

Script w/ Artifact:
http://i.imgur.com/qtgReCQ.png

[colours]

Also, I'm getting strange text artifacts using the same script on ''Opening from episode 1'' clip

Whelp, that seems to be TFM's combing detection going haywire (the credits seem to be 30 fps progressive, which interferes with field matching). You could try using tfm(pp=0) instead, or if you want to go the :effort: route, ivtc_txt30mc. (This is probably also going to be useful when dealing with the ending credits.)

As for wanting a 24 fps result, you could use something like the following.

ivtc = tfm(pp=0).tdecimate(mode=1)
deint = qtgmc().selectevery(5,0,2) # this drops about half of the frames
rfs(ivtc,deint,"[263 293]")