a few questions about AGK XviD settings

[drcl]

I got the b-frames working....

Just wondering why trellis and adaptive quants arent used and possibly Q-pel also(taking in to account player compatibility), also only 1 B-frame not 2? I only ask this because almost all encoding guides reccommend these settings.

Also why are the quanizers restricted to MIN 2, MAX 3 on I,P, and B? That makes no sense does it, especially for the B-frames?

Or is it only my setup that is manifesting these wierd xvid settings?

[manono]

Hi-

At the risk of provoking howls of outrage, I'll give you my take on it.

I don't know why Trellis isn't used. I use it myself. len0x will have to answer that one. But whether it's used or not, I don't think it's any big deal either way. My own opinion is that Adaptive Quants degrade the video in low bitrate or dark areas. Dark areas, especially, are places where the bitrate should be increased, and not decreased, in order to keep the so-called "dark blocks" from being worse than they already are. Q-Pel, as you suggested, isn't used because many standalones can't handle it. The same is true for more than one B-Frame.

Quant restriction is used in order to keep the quality of a 2-pass encode as even as possible. The idea is to approximate 1-pass quailty based encodes, but with precise file sizes. If you open one of your encodes in DRF Analyzer (http://www.geocities.com/analyzerDRF/), you'll see that even though the B-Frames might be set for 2/3, that the actual B-Frame quants are usually higher than that. Certainly the average B-Frame quant will be significantly higher than the average P-Frame quant, based on the 1/1.5/1 setting that AutoGK uses. And finally, those quant restrictions are in effect for only a certain range of compress test percentage results. If you encode with a 50% result, the quant restriction will be different. Evidently you either let AutoGK choose everything, or set it up yourself for high quality results.

Personally, I don't think those settings are "weird" at all, but make perfect sense, in spite of what other guides or "gurus" might say.

[drcl]

Thanks for the comprehensive explaination.

I'm completely happy with results I get from AGK, i was just a bit puzzled, because of the guides I have read.

I'm sure you have a good reason for not using trellis Len0x. Care to share?

[len0x]

I can see how trellis might help with low compressible sources, but considering that AutoGK always aims quite high (> 70%) removing of details doesn't sound that good to me. It makes sense to use this option with standard quantanization matrixes, but custom ones are supposed to compensate for this anyway (low bitrate Jawor's is certainly doing that).

P.S. thanks to manono for taking heat off me :)