deinterlace 1080i Pal source

[Mounir]

I'm in trouble with this source it seems take a look below
I've used qtgmc but it don't look good, source is 1080i (1088) Pal, live footage

zoom x400%
http://forum.doom9.org/attachment.php?attachmentid=13137&stc=1&d=1354373965

[Didée]

I see nothing special there. The interlacing structure seems okay.

Judging the image sample, it doesn't look like good quality to start with. Of course, QTGMC will not turn a crappy source into a brilliant result.

[Guest]

I think he's showing the deinterlaced result.

[Mounir]

the problem is there shouldn't be visible interlaced lines (see pic) as i have deinterlaced it with qtgmc...
i'll upload a video sample later

[Mounir]

sample :
http://www.mediafire.com/?gbrynwt6s3692yc

[Didée]

The content of that video basically is progressive. Try "TheSource.bob()", you'll see that always two frames show exactly the same motion state. The problem is that there is a vertical field shift, probably from wrong usage of an interlaced scaler during upscaling from SD to HD.

Since the video is not of "video-interlacing" type (where each field shows a different motion state), QTGMC is not really suited for this.

quickly shifting the even field down by 1 pixel:

separatefields()
ev=selecteven().spline36resize(width(),height(),0,-1,width(),height())
od=selectodd()
interleave(ev,od).weave()

Ooops, look at that, no more interlacing in the video content.
But hey, look at that, now there's combing in the text overlay.

(I'm not sure if 1 pixel is the fully-correct value to use. You try.)


When will people learn to look at the result of bob() first, before
pulling the QTGMC bazooka and complain if no magic does happen?

[Guest]

When will people learn to look at the result of bob() first, before pulling the QTGMC bazooka and complain if no magic does happen? Amen, bro', although I suggest to use SeparateFields() and inspect the actual fields themselves. Not too much between those, however, as long as you don't use some fancy adaptive bobber. Also, it forces you to understand what a field really is, and why it appears to bounce up and down after SeparateFields(). Just my eccentricity, really. :)

[Mounir]

thanks for the tip Didée
It actually look better to me if 1 is positive: spline36resize(width(),height(),0,1,width(),height())

The fields blend better and the output is slightly sharper but maybe that's just me
edit:
Nevermind, Didée was right

[Didée]

It actually look better to me if 1 is positive:
The source file is TFF. The -1 offset for topfield does fit when the source filter works correctly, by reporting the correct fieldorder.
If an offset of 1 loks better, then the source filter doesn't report the fieldorder, in that case Avisynth is (falsely) assuming BFF.