Finding a good B-Frame settings

[ChronoCross]

Ok I have determined that a max of 2 consecutive bframes should be used for the most part but what is a good way to figure out the other settings. the default is 1.50 for that one. is there any way to figure out what would be the best to get perfect quality b-frames?

[Blue_MiSfit]

:search: !!!

There is no one setting that will give you perfect quality. If you want to maximize the quality of your individual b-frames, then set the quantizer ratio ( the "middle" b-frame setting) to 1 - this makes b-frames start at the same quantizer as p-frames, which sort of reduces their usefulness.

In all honesty, 1.5 seems to work pretty well for most people.

When in doubt, just make some comp tests and see how it looks :)

[Teegedeck]

I'm sorry, while I can only support most of what Blue_MiSfit wrote, I have to contradict him in that b-frames at ratio=1 look no better than p-frames but are bigger than p-frames. Don't use ratio=1. Or only with at least offset=1. B-frames were meant to be compressed higher than p-frames, so please leave them at their default-setting.

@ChronoCross: Welcome to our community. It is a common misconception that default settings are there to be tweaked. XviD's default settings are only to be changed in order to enable advanced features that some people may not want to use because their hardware-player doesn't play them back or because they take too much time (GMC...) - but apart from that they are fine! There is nothing like an 'XviD Freaks secret society' that uses very advanced and secret settings which enable them to reach a quality superior to that outsiders can reach. (/giggling with my compadres of the XviD Freaks Secret Society while we have another beer and ridicule outsiders). Believe me, changing b-frame settings will only make the outcome worse; leave it be.

I should think that in 90% of all cases fiddling with b-frame ratios will only make a negative difference, if one at all. If you want better quality, you can reach it by enabling the more time-consuming features of XviD (VHQ, GMC, QP...) and by using high bitrates with the appropriate matrices.

If you want to have better-quality b-frames, you'll have to wait for XviD 1.1 with vhq for b-frames (or risk it and use the unstable alphas of 1.1).

[SeeMoreDigital]

Originally posted by Teegedeck
If you want to have better-quality b-frames, you'll have to wait for XviD 1.1 with vhq for b-frames (or risk it and use the unstable alphas of 1.1). Koepi has been kind enough to let me test one of the proposed new XviD (v1.1) builds.

In the main my interest has been with the new DSdec filter (which incorporates auto AR switching and de-interlacing options).

Unfortunately for me, the short 1B-VOP and 2B-VOP encodes I've generated still don't work (in hardware) with my Xcard.

But for you guys who watch your encodes (in software) on your PC's... your going to love it :)


Cheers

[Neo Neko]

If you want the best encodes you can; learn to use the zones effectively. It can make a big difference. Of course there is no perfect settings for such. It is all a matter of taste. That and I have had good results with reduced quantizer bframes. I tend to use more than 2 bframes often with good results. But on longer sequences multiple bframes can degrade and cause mosquito noise etc. I have found that dropping the bframe quantizer ratio to 1.25 at lowest keeps the mosquito noise from getting to bad with only a slight increase in file size. Then again it may all be pointless coincidence.

[Teegedeck]

I've also found zones very valuable when I aimed ad compressing anime as much as possible.

Ringing: This depends also very much on how strongly you compress and what kind of quantization you use. At p-frame quants 2-5 with Didée's matrices I don't have any problems with mosquito-noise.

[nightrhyme]

I'm too fiddling a little with the B-FRAME settings since Manono pointed me to the tool DRF-ANALYZER which seems to interpret your encodes as better quality(Lower Quant) when for instance using B-Frame settings = (1 - 1 - 1) instead of Default (2 - 1,5 - 1).

Here's an example of DRF ANALYZER Report Using Default B-Frame settings = (2 - 1,5 - 1) B-Frame-Default Link (http://hjem.tele2adsl.dk/nightrhyme/B-FRAME-DEFAULT.txt)

Here's an example of DRF ANALYZER Report Using B-Frame settings = (1 - 1 - 1) B-Frame-1-1-1 Link (http://hjem.tele2adsl.dk/nightrhyme/B-FRAME-1-1-1.txt)

Other XviD settings used in both encodes: VHQ=1, Motion Search=6,Quantization: 2-31 on all, Trellis, Closed GOV,NO packed bitstream, Custom Matrix= Didees SixOfNine Max=20

Are the DRF ANALYZER Reports not to be trusted ?

Hope you can help me on this since I'm very new to this and really would like to know.

//Thanx in advance

[sysKin]

Originally posted by nightrhyme
Are the DRF ANALYZER Reports not to be trusted ?
Oh you can trust them. The average quantizer is just a statistical fact, you really can't distrust such simple facts.

The interpretation of this result is a completely different story, and to make it short - lower average quant does not have to mean higher quality. In this particular case it doesn't. You realy *can* drown in a stream with an average depth of 15 centimeters...

Radek

[nightrhyme]

Ok
So which of the 2 Encodes is the better ?

[Teegedeck]

Isn't that obvious? The one that looks better!

All those artificial benchmarks and analyzers are only there for two reasons: 1) marketing for a codec (similar stuff is done with CPU-benchmarks) 2) help for programmers to optimize something without actually having to encode (which takes a lot of time that a programmer hasn't got)

Keep in mind that such programs merely try to guess what an actual human will find visually pleasing. You better use the real thing; i.e. eyes.