PDA

View Full Version : DivX's enhancements




Archer
27 Jun 2003, 12:41 AM
Hello!
Do GMC, Qpel, psychovisual enhancement work fine?

UncasMS
27 Jun 2003, 03:15 AM
i personally dont like them at all

Enchanter
27 Jun 2003, 10:09 PM
Same goes with me.

Archer
28 Jun 2003, 02:40 AM
Why d'you think so guys?
There must be a reason.

setarip
28 Jun 2003, 03:31 AM
Experiment with each of the features and arrive at your own conclusions...

Batman
28 Jun 2003, 03:54 AM
Depending on the bitrates/resolutions you choose QPEL, GMC, or psychovisual enchancements may or may not be helpful. You can read more about these features at the doom9 forums. I have only heard of limited success in some cases using QPEL and/or GMC. These features have not been fully developed.

Archer
28 Jun 2003, 06:16 AM
Originally posted by Batman
I have only heard of limited success in some cases using QPEL and/or GMC. These features have not been fully developed.


I guess the same is about lumi-masking & chroma motion of Xvid

Enchanter
28 Jun 2003, 09:56 AM
For example, QPel, while has a sharpening effect, creates disturbing effects on smokes and lights.

Archer
28 Jun 2003, 07:56 PM
So it seems that divx5 is not better then 3.11 alpha ? (because all it's advanced features are in the level of development)

Enchanter
28 Jun 2003, 08:09 PM
Originally posted by Archer
So it seems that divx5 is not better then 3.11 alpha ? (because all it's advanced features are in the level of development)
It depends on the type of video being compressed.

I personally find DivX 3.11a better than DivX 5 when it comes to animations. DivX 3.11a retains details very well as contrasted with the smoothing effect implemented into DivX 5.

For normal videos, the camp will be divided between those who prefer the good, ol' 3.11a and those for the newer DivX codec. This time I personally find DivX 5 to be better simply because it produces less artifacts (thanks to the smoothed and thus more compressible video) and is not prone to $hit artifacts (a problem with DivX 3.11a). Level of details is not as good as DivX 3.11a though, but overall the video should look better.

B-frames should be enabled all the time when encoding, unless you are using very high bitrates (i.e. around 2000kbps and above). Its usage allows you, for a given file size, to achieve higher quality than if used without.

p.s. Whenever referring to DivX 3.11a quality, I assumed this is done with nandub (which allows 2-pass routine with the codec).

UncasMS
28 Jun 2003, 08:23 PM
So it seems that divx5 is not better then 3.11 alpha ?

on the contrary

some of the latest options are useless to me, but overall visual impression is much better with divx5 imho!

Archer
29 Jun 2003, 01:45 AM
Thanks guys for your interesting mind

alucard83
29 Jun 2003, 03:39 AM
Quick question about Nandub, does it automatically do a 2 pass with Divx3.11a?

UncasMS
29 Jun 2003, 06:01 AM
it does

i wouldnt use it anylonger, though

use divx5 or xvid instead for better results

alucard83
29 Jun 2003, 06:24 AM
I mostly encode anime, and I notice that XviD blurs or darkens some of them while it pretty much does a great job on others. The ones that it doesn't do a good job on, I use DivX 5, but its been a while since I used Divx3.11, so I need to give it a run. I used Gnot for it a while back, but I'm running Ndub to see how it comes out

Enchanter
29 Jun 2003, 09:50 AM
Originally posted by alucard83
Quick question about Nandub, does it automatically do a 2 pass with Divx3.11a?
You need to make sure you have specified the STATS file first of all. After that, you need to run the First Pass (File >> First Pass) to generate the STATS file. Alternatively, you can run the Two Passes option which will perform the first pass, followed by the second pass as soon as the former is done.

Hence, if you have not specified the first pass STATS file and you run File >> Save as AVI, nandub will effectively be running the compression in single pass mode.

p.s. What Quantization type are you using in XviD?

alucard83
29 Jun 2003, 12:05 PM
I leave it at the defualt enchanter, which I believe is 2. Do you change around those numbers? If so, do you get better results? I usually play around with the the various 2-pass tuning section and curve aggression

Enchanter
29 Jun 2003, 01:38 PM
Originally posted by alucard83
I leave it at the defualt enchanter, which I believe is 2. Do you change around those numbers? If so, do you get better results? I usually play around with the the various 2-pass tuning section and curve aggression
Quantization = MPEG,, H.263, or MPEG-custom?

What you were referring to is the Motion Search Precision algorithm, which should be left at the default 5 or the even more accurate 6. I have not tried using something as low as 2, but I don't suppose the result will ever look good.

I usually tweak most of the settings around. The one thing I don't touch is Alternative Curve System. And since you are working with anime, don't enable any of the psychovisual settings, and most of the MPEG-4 advanced tools. What I personally do is enable B-frames (2 to 3 consecutive frames) and use a custom MPEG-matrix specifically meant for animation/CG. For decoding purposes, FFDShow does a beautiful job on XviD (and also DivX).

alucard83
29 Jun 2003, 04:23 PM
Heh, I'm talking about under the quantization tab with those min and max frame quantizers. I always put the MSP to 6. You say you use a custom MPEG matrix for animation/CG? Does this apply to AVIs as well? One more thing, would I get better playback using FFDShow to decode xvid than using the xvid codec by itself?

Enchanter
29 Jun 2003, 07:01 PM
For min/max quantisizer, I use 2 / 16-31.

"You say you use a custom MPEG matrix for animation/CG? Does this apply to AVIs as well?"

The question is not quite right. The encoding process will output to AVI, yes. Using two different matrices will normally result in different sharpness of the video. Using H.263 softens the picture while MPEG sharpens. The custom matrix I was referring to sharpens even more, but does not seem to suffer from artifacts as much as MPEG.

"would I get better playback using FFDShow to decode xvid than using the xvid codec by itself?"

FFDShow has post-processing filters that remove or hide away artifacts that result from compression. Few encodes do not suffer from compression artifacts and so having these filters around is handy.

alucard83
30 Jun 2003, 01:28 AM
Very nice E. Is it possible you can post your custom custom matrix here or send it to me so I can give it a try???? Thnx a bunch!:D

Archer
30 Jun 2003, 05:38 AM
Originally posted by Enchanter
What I personally do is enable B-frames (2 to 3 consecutive frames).

excuse me, but what value do you set in "maximum B-frames" & tell me about b-frames in xvid, please

Batman
30 Jun 2003, 05:41 AM
For xvid, some people have compiled custom matrices, which you can download then use for your encodes. Search the doom9 forum to find these matrices.

alucard83
30 Jun 2003, 06:52 AM
Originally posted by Batman
For xvid, some people have compiled custom matrices, which you can download then use for your encodes. Search the doom9 forum to find these matrices.

Schwanky:bglaugh2:

Enchanter
30 Jun 2003, 07:03 PM
"but what value do you set in "maximum B-frames"

As I said, 2 to 3 maximum consecutive frames.



"tell me about b-frames in xvid, please"

I can write pages on the technical aspects of B-frames (which will cause you to end up confused). Hence instead of that, let me just say that B-frames allows you to increase quality for a given filesize than if used without.