FairUse and AutoGK - about AutoIVTC methods

[bilu]

AutoGK AutoIVTC methods:
http://forum.doom9.org/showthread.php?s=&threadid=65985

FairUse AutoIVTC methods:
http://forum.doom9.org/showthread.php?s=&postid=395572#post395572

I know little about FairUse, never used it. I know more about AutoGK because I've been trying to help len0x with Decomb's hybrid thresholding, that hopefully will be available on the next version :)

I became curious about FairUse since it doesn't use Avisynth at all, but user feedback seems pretty good (although I've seen little feedback about this tool).

What I'm looking here is for some feedback about FairUse AutoIVTC methods, how accurate are they, if they can handle hard-telecined (Telecine without MPEG-2 RFF flags) and hybrid streams (streams with 24->30 fps Telecine + 30 fps Video) and how good is its deinterlacing process.

About AutoGK, it can already deal with both these kind of streams, but a fine-tuned hybrid treatment will be available on the next version. I'd like to know what is the current FairUse status about this, mainly because it isn't Avisynth-based, which is something new to me :)


Best regards,
Bilu

[fu2k]

Originally posted by bilu
if they can handle hard-telecined (Telecine without MPEG-2 RFF flags)
FairUse will handle soft telecine (RFF), hard telecine (no RFF) or any arbritary mixture of the two (some movies are mostly soft telecine except near chapter breaks where they use a second or two of hard telecine).
hybrid streams (streams with 24->30 fps Telecine + 30 fps Video)
It will process them, but getting a good result may be difficult. If you select native mode (no IVTC, 29.97fps) then it just obeys all the RFF flags so the telecined portions will have duplicate fields. If you select IVTC (23.976fps) then the telecined portions will be converted back to progressive fine, but in the true 29.97fps sections it will attempt to align fields so as to minimise interlacing artifacts, and it will have to discard 1 of every 5 frames (on average) to maintain the lower frame rate. What else can be done?
and how good is its deinterlacing process.
Deinterlacing is achieved as a byproduct of low-pass filtering during the vertical resize. I haven't compared it with other dedicated deinterlacing methods.

[bilu]

Could you describe the Auto-Detect process? How do you choose between FILM, Interlace, and Hybrid? If Hybrid, how do you choose between 30 fps and 24 fps?


Bilu

[fu2k]

The algorithm that FairUse uses to suggest IVTC or native is quite weak. Here is the description I posted previously (in another thread):

"The code that tries to predict whether IVTC is appropriate simply processes a subset of the non-RFF parts of the input video using the IVTC algorithm and looks at the interlacing levels of the output. If there is "too much" interlacing then IVTC is not recommended to the user."

So if the IVTC seems to work well then IVTC is suggested. Unfortunately, this method is ultra conservative. Often, even a small amount of interlacing left over after an otherwise successful IVTC can cause the auto-detect to "throw up it's hands" and recommend native mode. Of course, in these cases, the user can override this suggestion and use IVTC anyway.

[bilu]

This thread, although Avisynth-based, could provide inspiration for improvements in FairUse:

http://forum.doom9.org/showthread.php?s=&threadid=65985

It describes a bit of the algorythms used in AutoGK. The most important improvement that FairUse could get IMHO would be doing analisys only over moving parts of the movie. It could greatly improve its statistics.


Bilu

[fu2k]

I think a better solution is to use a video container that allows a variable frame rate. Thus the telecined portions can be converted and stored at 23.976fps and the native portions can be stored at 29.97fps.

[bilu]

Been there, done that. ;)

http://forum.doom9.org/showthread.php?s=&postid=395247#post395247

But not very helpful for standalones :rolleyes:


Bilu