Need comments on Xvid recommend settings

[Jerry_Sm@rt]

I get some recommend settings from a dvdrip forum.It's like this:
Common settings:
Motion search 6,VHQ 4
Quarter Pixel(do not use for low bitrate),Chroma motion,
Adaptive Quantization,GMC,hvs-best or hvs-better
2-PASS/1-PASS>70%:
no BF£¬I¡¢PF only,
IF¡¢PF¡¢BF Quantization£º1¡¢2¡¢1¡¢3¡¢1¡¢31
2-PASS/1-PASS between 60%-70%:
BVOPs : 2 /1.2/0.6
IF¡¢PF¡¢BF Quantization£º1¡¢2¡¢1¡¢3¡¢1¡¢31
2-PASS/1-PASS between 50%-60%£º
add some smooth filters,
BVOPs : 2 /1.2/0.6
IF¡¢PF¡¢BF Quantization£º1¡¢3¡¢1¡¢4¡¢1¡¢31


I tried these out,sometimes i got oversized files,but alwayls better
results in DivXDRFAnalyzer than default settings.

Comments,please.

[Sharktooth]

:readfaq: :search: :readguid: :thanks:

[communist]

Originally posted by Jerry_Sm@rt
I get some recommend settings from a dvdrip forum.It's like this:
Common settings:
Motion search 6,VHQ 4
Quarter Pixel(do not use for low bitrate),Chroma motion,
Adaptive Quantization,GMC,hvs-best or hvs-better
2-PASS/1-PASS>70%:
no BF£¬I¡¢PF only,
IF¡¢PF¡¢BF Quantization£º1¡¢2¡¢1¡¢3¡¢1¡¢31
2-PASS/1-PASS between 60%-70%:
BVOPs : 2 /1.2/0.6
IF¡¢PF¡¢BF Quantization£º1¡¢2¡¢1¡¢3¡¢1¡¢31
2-PASS/1-PASS between 50%-60%£º
add some smooth filters,
BVOPs : 2 /1.2/0.6
IF¡¢PF¡¢BF Quantization£º1¡¢3¡¢1¡¢4¡¢1¡¢31


I tried these out,sometimes i got oversized files,but alwayls better
results in DivXDRFAnalyzer than default settings.

Comments,please.
Apart from that your post is not easy to decrypt - MSP@6 /VHQ4 / CM / AQ and use of custom matrix is what a lot (ya I know some dont like AQ...) of people advise. I personally like the defaults pretty much. I just activate all the extra CPU requiring stuff and once in a while select a custom matrix. Its not that there are some magical settings that compresses 50% better than the default ones ;)

[unmei]

Especially the specific CQM recommendaton is a bit suspicious imo. This has a lot to do with your own preferences. And most likely even if you know a few CQMs you like, one might be better suited for higher bitrates while it looses badly on lower bit rates and vice versa.

I recently made a little PSNR test (yeah, no comments please :cool: ) for 25 CQMs i thought that could be worth testing for anime. The case that one matrix had higher PSNR on Q2, lower on 2000kbit and higher again on 1000kbit (compared to another) was not even all too seldom.
IMO it just shows that some matrices have sweet spots while others tend to give more uniform behaviour over a wide range of bitrates. The probably only thing you can say in general is that a matrix that maxes out at a low bitrate cannot give you too many details while most of those that have exremly high bitrate at Q2 do not show too high PSNR at lower bitrates.. surprise surprise, you can actually come to this conclusions without such a test. But it was not useless to me as it showed me a bit which matrices are rather hi-bitrate, which ones rather low and finally which ones seem not suitable for anime. (Once again, it was a pure PSNR test, and not aimed at pointing out an absolute ranking to be applied)

[manono]

Hi-

Apart from that your post is not easy to decrypt

Yeah, it took me a couple of minutes to figure out also. "no BF£¬I¡¢PF only" means with greater than 70% not to use B-Frames. "IF¡¢PF¡¢BF Quantization£º1¡¢2¡¢1¡¢3¡¢1¡¢31" means that the I-Frame/P-Frame/B-Frame quants are restricted to 1/2 1/3 1/31.

The DRFAnalyzer results are "better" with those settings because the B-Frame settings are a bit more conservative than the defaults, and the quants are capped. So, by "better", I assume that you mean the average quant is lower with those settings than when using the defaults, and that you get either a lot more green, or all green with those settings, and that the Standard Deviation is lower. Are they really "better"? Matter of opinion, I'd say. I'm one of those that refuses to use AQ, and there's no way I'd ever allow a quant 1 into my encodes. But I can see why you get oversized sometimes by blindly following those settings. Some obvious thought went into creating those settings, and I agree with some of the theory behind some of them, but, personally, I wouldn't use them. Not all of them, anyway. And, of course, the conventional wisdom around here is that restricting the quants isn't necessary and that the B-Frame defaults are the best.

But I thought that the settings were interesting, if only to provide a basis for discussion.

[Didée]

According to DRFanalyzer, most of my encodings are of either medium or low quality, and often may contain some very poor frames :D

DRFanalyzer is okay for analyzing a stream if you know what your're doing and looking for, and how to interpret its results.

Blindly believing its ranking is a big LOL, especially when one uses CQMs, since the proggy has not the slightest clue that something like CQMs does exist. All its calculations are based on H.263 quantization.

[manono]

Good point, Didée. I forgot to mention that. If I do higher bitrate encodes with Selur's matrix (no offense intended :)), my DRFAnalyzer results aren't so good anymore. But the movie sure looks good. And when using high quants for end credits, it scolds me with that "may contain some very poor frames" message.

[Didée]

Originally posted by manono
If I do higher bitrate encodes with Selur's matrix (no offense intended :))
What offense? I never pretended to have found the holy grail and to have forged it in 2*64 bytes... *)

However, DRFanalyzer can be trained a little. IIRC somewhere in an INI file or somewhere else (too lazy now too look it up), you can specify what quantizers should give what ranking. So, if one uses a specific matrix very often, just adjust these values.


*) (I just know it is so, but keep the secret :D )

[Jerry_Sm@rt]

Thanks for all the replies.
It took me a long time to figure out all your replies ,due to my poor English.
Maybe DRFAnalyzer means something but not everything.