Why can't I get the hang of deinterlacing?

[mitsubishi]

SD stuff, the meGUI analyser seems to work fine, but just cannot handle HD and mbaff.

Forgetting my other problems with a partly interlaced source, I now have one which is (I think) progressive, until the credits.

http://www.mediafire.com/?fiikxrddygu (28MB)

I've tried all the options in meGUI and had a play around with bob or seperatefields, but everything comes out savage.

Most leave horrific artifacts.

This seems to work: "SeparateFields().selecteven().Spline36resize(1440,1080)"

The resize is just for testing, it'll be 720 at the end. It's still not smooth, is that just because of the speed it's moving at 25fps?

Is there any better option?

[Tack]

I'm afraid I don't know avisynth, but I suspect your problems could be related to field dominance. I have no problems deinterlacing this (with yadif) specifying top field first.

[mitsubishi]

I'll try Yadif, I don't have it and the download link was down, probbably on warpenterprises also.

Well I tried removing the selecteven and letting it run at 50fps and it was indeed smooth*, so I guess that's the reason, I can try to figure out how to re-interlace it or have a 50fps section. Oddly though the 50fps file is smaller on crf than the 25fps one.


*For some reason, it is smooth then stops, then jerks forward. It stops doing this if I go full screen and then mouse down to bring up the seek bar.. wierd.

[TheRyuu]

Try this (really simple):
LoadPlugin("C:\Program Files\Avisynth 2.5\Plugins\TIVTC.dll")
Source() #replace with appropriate source
tfm()

All you really need is field matching IIRC.
Read here:
http://avisynth.org.ru/docs/english/externalfilters/tivtc.htm

Scroll down to the "Pal" section.

[neuron2]

It's TFF interlaced scrolling credits in B&W. That is very challenging for any single-rate deinterlacer. Any decent bobber can make it look fine.

Playing the AVS script will probably be jerky due to processing delays but the encoded result is fine. I used this:

loadplugin("dgavcdecode.dll")
AVCSource("E:\tmp\cut.dga")
leakkernelbob(order=1,threshold=0)

You can replace leakkernelbob() with leakkerneldeint() and it looks OK but not as smooth.

[neuron2]

All you really need is field matching IIRC. It's interlaced video, not progressive!

[mitsubishi]

Thanks

Download link for Yadif was down before, which is why I didn't try it, but now I have it, it works great and doesn't leave all the artifacts the other methods did.

tfm() also works, now I have 3 methods, which all work and have to decide :(

Sep Fields: 10.4MB
Yadif: 13.4MB
TFM: 11.1MB

There is one thing I don't get in the Yadif docs, it says:

Field parity of source clip must be properly set (use AssumeTFF, AssumeBFF).

Why does it say this, is there a difference between:

AssumeTFF.Yadif()

and

Yadif(order=1)

They both appear to produce exactly the same result...

[TheRyuu]

It's interlaced video, not progressive!

It worked for me and I have no idea why...

[neuron2]

It worked for me and I have no idea why... TFM() does postprocessing by default. There is no point to doing field matching on interlaced video. Use Tdeint() if you want tritical-style deinterlacing.

[mitsubishi]

Thanks Neuron, those work too, LKD leaves artifacts when threshold is left, but setting them to 0 gets rid of them.