Periodic interlace or wrong IVTC?

[Mystery Keeper]

I deal a lot with Japanese NTSC DVD. So, my scripts usually start with something like this:
string d2vfile = <path>
mpeg2source(d2v=d2vfile, info=3)
ColorMatrix(d2v=d2vfile, interlaced=true)

crop( 0, 64, -0, -64)

tfm(d2v=d2vfile, order=-1, pp=0, chroma=true, slow=2).tdecimate(mode=2)
I often get result with some sequences having every 4th frame interlaced. Not the entire clip, but several sequences here and there. Am I doing IVTC wrong or is it mastering artifact?

[Guest]

There's no way for us to guess. Please post a link to an unprocessed source sample that we can use to reproduce the issue. Possibly you have field-blended source.

[Mystery Keeper]

You know I can not legally give you any part of copyrighted DVD. Guess I'll have to learn more about telecine to distinguish different source types. Still, if anyone had similar experience, your input is welcome.

[Guest]

You know I can not legally give you any part of copyrighted DVD. That's wrong. You can give a small sample under the fair use clause of copyright. It's not rule 6 material is it?

http://en.wikipedia.org/wiki/Fair_use

[Mystery Keeper]

*sigh* Was about to make a sample, but as I was looking for better part, I realized the source is much more heavily and randomly interlaced than just those several sequences. And so are most of sources I've got. This thread is meaningless as I'm still better off doing complete deinterlace than finding individual interlaced frames and doing conditional filtering.

[Guest]

Have it your way. But using terms like "randomly interlaced" reveals that you are basically a noob, so you could serve yourself better by showing a sample and letting people that know about this stuff help you.

For example, if you have blended fields that can be restored with a deblender then you will needlessly degrade the video by performing "complete deinterlace".

[Mystery Keeper]

Ok. Here's the part I was going to upload.
http://www.mediafire.com/?n462884slxa8v18
I'll hold back any comments of my own for a while and let you pros analyze it.

[LemMotlow]

http://neuron2.net/faq.html
Look to me like film-based progressive + 3:2 pulldown. Using separatefields(), the frame (field) sequence is aa bb bc cd dd ee ff fg gh etc. etc.. If you prefer a repeating-image sequence, aa bbb cc ddd ee fff gg hhh etc.
That is, that's what it looks like when something moves.

[johnmeyer]

Looks like standard 3:2 pulldown to me.

TFM()
TDecimate()

is all you need. Use the defaults (i.e., don't specify any overrides).

[Guest]

Looks like a mix of hard telecine and 29.97 progressive. I would just field match it and not decimate it. Or you can go VFR.

[Mystery Keeper]

johnmeyer, I don't know what good would it make to not use any overrides in TIVTC. But I tried it anyway. Awful result with a lot of combing.

neuron2, suggestion taken, but that still leaves me with a lot of combing (which was what I meant by "randomly interlaced"). It mostly occurs on obviously edited frames - crossfades and overlays. But that's not always the case. See the very first leaves sequence.

[Guest]

I'm not seeing any combing in the leaves with this script:

dgsource("iberia-sample.dgi")
telecide(post=0)

Please tell me where you see combing (frame numbers) and give your script.

[Mystery Keeper]

Ughhhh. I tried using HistogramAdjust to make combing more pronounced, and now I see the *whole* leaves sequence combed *and* blocky. Script is:
string d2vfile = "F:\Video to Process\Sound Horizon - 石畳の緋き悪魔\iberia-sample.d2v"
MPEG2Source(d2v=d2vfile, info=3)
ColorMatrix(d2v=d2vfile, interlaced=true)

tfm(d2v=d2vfile, order=-1, pp=0, chroma=true, slow=2)

HistogramAdjust(limit=50)

[Guest]

Try my script please and report. It looks like you are shooting yourself in the foot. Forget HistogramAdjust(). You don't need it to see combing, and it may mislead you.

[Mystery Keeper]

Could only try it with MPEG2Source(d2v=d2vfile). Haven't got DGDecNV license.

string d2vfile = "F:\Video to Process\Sound Horizon - 石畳の緋き悪魔\iberia-sample.d2v"
DGDecode_MPEG2Source(d2v=d2vfile)
telecide(post=0)

Still the same combing.

[Guest]

Make sure you have field operation set to Honor Pulldown in DGIndex! Forced film will mess up the field matching.

No combing here with that same script and Honor Pulldown. Please post a screenshot showing the combing and tell me the exact frame number.

[Mystery Keeper]

string d2vfile = "F:\Video to Process\Sound Horizon - 石畳の緋き悪魔\iberia-sample.d2v"
DGDecode_MPEG2Source(d2v=d2vfile)
telecide(post=0)
a = last
b = a.nnedi3(field=-1, nsize=3, nns=4, qual=2)

stack = StackHorizontal(a, b)
diff = StackHorizontal(mt_makediff(a, b, u=3, v=3), BlankClip(a))
StackVertical(stack, diff)

http://www.mediafire.com/?mkx68i205yp5j

Those are only selected frames. There are much more.

0-6: Leaves. Combing more or less visible in every frame. The most pronounced in the bottom.
96, 145, 221: Overlay and fade-in.
520: Layla's right side and face.
833: Twirl SFX sequence.
1057: Rider and horse on the right. Hint: heavily edited composite scene. In original there's a grey stone wall with no paintings and live woman.
1074: Blending.
1131: This particular sequence is weird. It is moving with *almost* constant combing. Field order swaps in few places.
1242: Shaking SFX. Note the mirrored shaitan's face.

[Guest]

OK, I can see issues with fades and cuts. Best you can do in my humble opinion is to enable post-processing. Then you deinterlace only where you need it.

telecide(vthresh=15)