Blend frames when doing decimate

[Efenstor]

Is there a decimation plugin/script that can blend duplicate frames instead of simply killing them? This would give some additional quality for those sources in which the remains of interlacing are still visible after telecining and decimation (like those of my AVCHD camcorder's 24p mode).

[neuron2]

Both Decimate() and TDecimate() can do it. Refer to the respective users manuals.

IMHO, you'd be better off fixing your residual interlacing issue. It shouldn't be happening. Can you post an unprocessed source sample that we can use to duplicate the problem?

[Efenstor]

Here it is:
http://img171.imageshack.us/img171/3764/0000y.th.png (http://img171.imageshack.us/i/0000y.png/)

The 60i to 24p script is the following:
AVISource("ZZZ.avi")
AssumeTFF()
ConvertToYV12(interlaced=true)

Telecide(guide=1)
Decimate()
Sharpen(1)

I need some sharpening, since the source is too soft. ZZZ.avi is the directly converted source H264 MTS. I used Vegas Movie Studio Platinum 9.0 for the MTS to AVI conversion, but it is not the source of the problem, the other converters give the same results.

[vampiredom]

Is this a 24p clip with pulldown added or a "true" 60i clip?

[neuron2]

Here it is I was referring to posting an unprocessed source sample video clip.

[vampiredom]

I used Vegas Movie Studio Platinum 9.0 for the MTS to AVI conversion

Why not try neuron2's DGAVCDec (http://neuron2.net/dgavcdec/dgavcdec.html) instead -- and choose "Force Film"? That should work, no?

[neuron2]

Why not try neuron2's DGAVCDec (http://neuron2.net/dgavcdec/dgavcdec.html) instead -- and choose "Force Film"? That should work, no? Force Film is not a current feature of DGAVCDec.

[shoopdabloop]

what about

TDeint(1,-1,edeint=separatefields().EEDI2(field=-2))

for deinterlacing to 60p while leaving non-interlaced frames untouched, then getting rid of the extra frames for 24p?

[vampiredom]

Perhaps I'm confused: Doesn't DGAVCIndex have a Video / Field Operation / Forced Film option? It appears to, and the feature is referenced in the manual. No?

[neuron2]

Did you test it through DGAVCDecode.dll?

How would I know, I only wrote the program. :)

I better add that to the release notes.

[vampiredom]

Honestly, I cannot recall if I have ever used the Forced Film option in DVAVCIndex before. I don't really have any pulled-down 24p h.264 to test at the moment. I am hoping that Efenstor can avail the original AVCHD clip so it can be checked out.

[Efenstor]

The clip was shot in the true 24p mode, but the stream produced by the camera is the telecined 60i, that's why I need IVTC. I can't use DGAVCDecode because it gives these (http://neuron2.net/board/viewtopic.php?t=1311&sid=9a342df218018da22f8d622d59f4231c) artifacts with all of my source files (yes, my camcorder is Canon HF100, and as far as I know, the artifacts are caused by the bi-directional coding which is still unsupported by ffmpeg). The clip itself coming soon...

[Efenstor]

By the way, just donated for DGAVCDecNV, will try it.

[Efenstor]

Okay, now: http://www.efenstor.net/external/00066_trimmed.mts (72.4 MB). By the way, I give my permission to copy and use this file for any other AVC-related researching.

[neuron2]

It's hard telecined, so Force Film would not help.

The material is difficult for an adaptive field matcher. Since it is a constant pattern throughout, you can do it the old fashioned way:

loadplugin("dgdecodenv.dll")
DGSource("E:\tmp\00066_trimmed.dga")
assumetff()
separatefields()
selectevery(10,0,1,2,3,6,5,8,9)
weave()
trim(2,0)

I have the trim at the end because the first few frames are apparently not in pattern for some reason.

You'll have to determine the correct selectevery() invocation (telecine phase) for each clip.

[Efenstor]

Tried different methods and it seems that the best possible quality for Canon HF100 24p mode is achievable with DGAVCDecNV (without NVidia deinterlacing) + nnedi2 + Decimate.
neuron2, thanks for the help and DGAVCDecNV! :) It decodes my videos with excellent quality and no artifacts (hence I conclude no ffmpeg code is used there).

[Efenstor]

No, TFM(mode=4) + TDecimate() is the best.

[neuron2]

Best is relative. Don't forget rule 12.

Actually, any heuristic matcher is going to make mistakes that you won't get with the script I gave you.

[Efenstor]

Considering the amount of video I shoot and edit, it's impossible to use manual IVTC as you've suggested. I know that method is the best but unfortunately each scene has its own pulldown sequence offset, because the camcorder starts recording immediately after pressing the button and the internal frame format is always 60i. Hence I have nowhere to run from the automated methods. :)