Tweaking FieldDeinterlace?

[WildCelt]

Greetings all,

I just read the updated IVTC/Deinterlacing FAQ (http://www.doom9.org/ivtc-tut.htm) and it stated FieldDeinterlace "Requires tweaking, or quality can be a disappointment." What do I do to "tweak" FieldDeinterlace? I did a search but did not find any useful infomation about this.

Any info is appreciated.

[sh0dan]

There is documentation in the decomb package.

You should mainly play with the "dthreshold" parameter.

[WildCelt]

Thanks, sh0dan, for the info. I did not notice in the readme any specific tweaking tips, but I will study it further.

Reading the info in the package got me thinking...I have a source that is at times interlaced and at times telecined. Could I just run FieldDeinterlace (full=true) on the whole thing rather than breaking it up into interlaced/telecined sections? If I am reading the info correctly, it will catch everything, but will it leave the progressive frames in the telecined sections alone? I realize it will leave a spare frame in the telecined portions, but I was going to leave them anyway (I don't know of any other way to avoid the fps differences).

[sh0dan]

You should try telecide() instead - perhaps with guide=1, or fielddeinterlace(full=false) - default is true for fd.

[WildCelt]

Originally posted by sh0dan
You should try telecide() instead - perhaps with guide=1, or fielddeinterlace(full=false) - default is true for fd.

Will using telecide()catch the interlaced sections as well?

[Guest]

Originally posted by WildCelt
I just read the updated IVTC/Deinterlacing FAQ (http://www.doom9.org/ivtc-tut.htm) and it stated FieldDeinterlace "Requires tweaking, or quality can be a disappointment." I don't know why they wrote that and I don't agree with it. I suppose some people feel that the default value of 'dthreshold' is a little high, and they claim it leaves residual combing. But often people look for residual combing by zooming on the video. This is not a good idea for reasons I've described elsewhere. You can reduce 'dthreshold' if you like but as you lower it, you get less and less benefit from the adaptive deinterlacing algorithm.

I am changing the default for 'dthreshold' to 7 in the next release (I'm also changing the default for 'threshold' to 20 and 'gthresh' to 25). Remember that you can easily set your own default values if you don't like the ones I've chosen.

[Guest]

Originally posted by WildCelt
Will using telecide()catch the interlaced sections as well? Try the help file. It's very well written. :)

As long as you do not disable postprocessing, yes, the interlaced sections will be deinterlaced. Note that Decomb provides some special modes designed for hybrid material. I refer you to the help file.

[WildCelt]

Help file, eh? Who'da thunk. :)

Sorry, I guess I should have studied it a bit more carefully before posting. I read what you termed "hybrid" to mean something else. I will hit the readme. Thanks to both of you for your expertise and assistance. This will save me a ton of time!

[Didée]

The topic title reminded me on something I was wondering about some time ago:

I feel that some "scenechange detection" could be useful for fieldeinterlace: mpeg-4 codecs tend to often not set keyframes when the scenechanges got blended ... so: a possibility to use fieldeinterlace with (blend=true) in general, but (blend=false) on scenechanges is what's on my mind.

Would that be difficult to integrate?

Didée

[Guest]

Originally posted by Didée
Would that be difficult to integrate?
Perhaps not difficult, but certainly non-trivial. I don't see the point of your request, however, so I can't judge if it would be worth the effort. Why would it be desirable to treat the scene-change frames differently?

[Guest]

In regard to tweaking 'dthreshold', the latest version of Decomb allows you to display the motion map and thereby visualize the effects of different values of 'dthreshold'. Please see the "Decomb beta thread".