Log in

View Full Version : Small XviD and VP6.2 test [due to VP62 quality fluctuations]


Pages : 1 [2]

Sharktooth
16th July 2004, 13:00
Ok, my theories about VP6 are almost confirmed (below you will find some screenshots that proves what i am saying).
Here's my findings in a schematic way:

XVID: I-Frame (max quality) -> P-Frame (near max quality, sometimes shows artifacts) -> B-Frame (almost good quality in low motion/mediocre quality in high motion, shows artifacts)

VP6: I-Frame (sometimes max quality, sometimes mediocre) -> P-Frame (increasing quality over time, usually more detailed than I-Frame)

Heres some screenshots so you can better undestand what i'm trying to explain:

VP6.2 - Keyframe
http://ebola.gamersrevolt.it/codec_comp/vp62-01.PNG
This time VP6 shows good details in the I-Frame

XVID 1.0.1 - Keyframe
http://ebola.gamersrevolt.it/codec_comp/xvid101-01.PNG
XVID also has a good amount of details, it is slightly better than VP6

DIVX 5.2 - Keyframe
http://ebola.gamersrevolt.it/codec_comp/divx52-01.PNG
This is DivX PRO 5.2 with MPEG matrix (Standard, no QPEL, no GMC, fast PSY, Multiple B-Frames). Better results may be obtained using SLOW mode.

VP6.2 - Keyframe
http://ebola.gamersrevolt.it/codec_comp/vp62-02.PNG
Washed out I-Frame. We will see that in the next frames VP6 will gradually rise the image quality until it gets more detailed than XVID!

XVID 1.0.1 - Keyframe
http://ebola.gamersrevolt.it/codec_comp/xvid101-02.PNG
XVID does always a good job with I-Frames.

DIVX 5.2 - Keyframe
http://ebola.gamersrevolt.it/codec_comp/divx52-02.PNG
As usual this is not indicative of the best divx52 quality but shows the detail improvements it got with MPEG matrix support but there are too much artifacts...

VP6.2 - P-Frame
http://ebola.gamersrevolt.it/codec_comp/vp62-02b.PNG
16 frames after the keyframe VP6 shows its ballz regaining the lost quality and showing superior details. Even more than xvid.

XVID 1.0.1 - P-Frame
http://ebola.gamersrevolt.it/codec_comp/xvid101-02b.PNG
Xvid mantains a good amount of details since this particular sequence has minimal motion.

DIVX 5.2 - P-Frame
http://ebola.gamersrevolt.it/codec_comp/divx52-02.PNG
Divx has the same behaviour as Xvid due to low motion but still cant catch with it quality-wise. There are still artifacts. I want to stress again the fact the encode was done in Standard mode and better results may be obtained with SLOW mode.

VP6.2 - Keyframe
http://ebola.gamersrevolt.it/codec_comp/vp62-03.PNG
Again a washed out Keyframe. We will se that in the next frames VP6.2 will rise the details gradually.

XVID 1.0.1 - Keyframe
http://ebola.gamersrevolt.it/codec_comp/xvid101-03.PNG
Xvid, as usual, has a good quality in every keyframe

DIVX 5.2 - Keyframe
http://ebola.gamersrevolt.it/codec_comp/divx52-03.PNG
Even Divx has more details than VP6 in this I-Frame.

VP6.2 - P-Frame
http://ebola.gamersrevolt.it/codec_comp/vp62-03b.PNG
Once again VP6 start to rise the details and this shot is 16 frames after the I-Frame. Once again VP6 shows more details than XVID.

XVID 1.0.1 - P-Frame
http://ebola.gamersrevolt.it/codec_comp/xvid101-03b.PNG
Xvid mantains an high quality due to the low motion scene.

DIVX 5.2 - P-Frame
http://ebola.gamersrevolt.it/codec_comp/divx52-03b.PNG
Divx loses ground and its rivals show more details.

VP6.2 - P-Frame
http://ebola.gamersrevolt.it/codec_comp/vp62-04.PNG
This P-Frame shows how VP6 copes better than xvid in high motion. Where xvid shows artifacts, VP6 is able to preserve details!

XVID 1.0.1 - P-Frame
http://ebola.gamersrevolt.it/codec_comp/xvid101-04.PNG
Blocks!!!

DIVX 5.2 - P-Frame
http://ebola.gamersrevolt.it/codec_comp/divx52-04.PNG
Blocks!!!

Conclusions.
XVID: Great quality and details in low motion scenes. Artifacts are showing around edges, moving blocks and massive quality degradation during high motion.

VP6: Low motion scenes start with a low detailed i-frame but the codec tends to raise details immediately but gradually. In about half a second the quality is restored to an incredible level, leaving the competitors behind. The visual effect is almost unnoticeable coz the human eye cant distinguish details during the scene change, it needs a small amount of time to spot them... but some purists may contest that. High motion is perfectly detailed, no artifacts at all.

DIVX: The codec behaviour is similar to XVID but still cant keep up with the details. Using SLOW MODE would have produced a slightly better encode but custom matrices support is needed to stay on par with the details provided by XVID+Custom matrices...

pernas
16th July 2004, 23:01
Oh gosh ! that was plenty of screenshots :)
(Where do you find time to do this ?)

Ooops - DivX 5.2 - i must have missed it :scared:

I think the raise of details after a low detailed I-frame is visible when comparing the codecs, at least when going for lower bitrates.
My first encode with VP62 (Jesus of Nazareth apprx. 6hrs) showed this - but i was able to shrink it to 3 CD's ;)

My latest attempt (The Hulk, 2.12.25) gave me very good (enough) results with
VP62 on 1 (!) CD - only 555kb/s - and i must say that i have changed my mind
about details & artifacts/blocking vs. sharpness mixed with some smooth passages.

I'm aware of that my goal of putting (too much) on one CD could be a little unkind
way of treating codecs - but for me (at least now) it really works well with VP62 and some preprocessing :)

But since i find "the world of codecs" very interesting i will continue to follow the development of other codecs
(as much as my spare-time allows (and wife also :) )

Regards,

Sharktooth
17th July 2004, 02:32
I'm pretty happy with either xvid and vp6.
Usually i use VP6 for 1CD or high motion backups.
XVID is still the best for very high bitrates (or quant 2 encodes) but i like the idea of the efficiency of VP6.
I hope On2 will further develop the ideas behind VP6 coz i really like how it handles artifacts expecially in high motion scenes.
I also would like to see how VHQ5+ looks with the upcoming xvid release.
But we all know the real leap will be h264 and i really hope xvid devs will go that way...

Sharktooth
17th July 2004, 14:05
*Update*
Since i thought the DivX 5.2 encode could be enhanced with some tweaking, i've done a SLOW MODE encode with the following settings:
2 passes, No profile, 2000kbps, SLOW (it was STANDARD in my previous encode), FAST PSY, 1 BFRAME (source is high motion), MPEG matrix, I-Frame int. 250, NO GMC, NO QPEL.
Well... i cant tell the difference with my previous one...
I'm really disappointed by the amount of artifacts (blocking in particular).

However for the sake of completeness heres the DivX 5.2 encode:
http://valborg.hn.org/sharktooth/divx52-2000kbps-mpeg-slow.avi

SeeMoreDigital
17th July 2004, 16:30
Originally posted by Sharktooth
*Update*
Since i thought the DivX 5.2 encode could be enhanced with some tweaking, i've done a SLOW MODE encode with the following settings:
2 passes, No profile, 2000kbps, SLOW (it was STANDARD in my previous encode), FAST PSY, 1 BFRAME (source is high motion), MPEG matrix, I-Frame int. 250, NO GMC, NO QPEL.
Well... i cant tell the difference with my previous one...[/url] I'm gonna put my neck on the block and say... I don't think 2pass DivX5.2, even with it's new B-VOP implementation, is able to generate encodes to rival XviD, without B-VOP at between 500-1000Kbps...

Can anybody else confirm?

Sharktooth
17th July 2004, 17:04
Fully agreed. DivX is not on par with Xvid.
Even if the codecs were on par, xvid has big advantages: it is MUCH MORE configurable and supports custom matrices.
Until DivX network dont go AVC i doubt DivX could ever beat xvid.

Sharktooth
2nd August 2004, 19:58
Sorry to bring up this old thread but finally i had some time to make another encode.
Here's the DivX 5.2 PRO with slow mode, H263, GMC and single adaptive b-frame encode @ 2000kbps. Same usual source as previous tests.
This time i used GMC as in Xvid encodes.

http://valborg.hn.org/sharktooth/divx52-2000kbps-h263-slow-gmc.avi

As you may easily notice it is really better than the clip with MPEG quant matrix.
I believe, as many others, there's sometimg wrong in the DivX MPEG QM implementation.
Don't use MPEG matrix with DivX until they fix it...

Sharktooth
3rd August 2004, 14:14
Here we go with some screenshots:

DivX 5.2 PRO - H.263 and GMC:
http://ebola.gamersrevolt.it/codec_comp/dx52-h263-01.png

http://ebola.gamersrevolt.it/codec_comp/dx52-h263-02.png

http://ebola.gamersrevolt.it/codec_comp/dx52-h263-03.png

victorhooi
4th August 2004, 03:08
Hi,

OK, this has absolutely nothign to do with this thread :D, but what is that movie? Also, I haven't read this thread yet (kinda hairy here), but just from a cursory look, this looks like a damn detailed effort - great job! - also, why no RV10 (apologies if that's already been answered, or its obvious).

Thanks,
Victor

koszopal
4th August 2004, 07:08
@victor
animatrix
koszopal

Sharktooth
14th August 2004, 11:52
Originally posted by victorhooi
Hi,

OK, this has absolutely nothign to do with this thread :D, but what is that movie? Also, I haven't read this thread yet (kinda hairy here), but just from a cursory look, this looks like a damn detailed effort - great job! - also, why no RV10 (apologies if that's already been answered, or its obvious).

As koszopal said it's The Animatrix, and thanks for your kind words.
There's no RV10 clip simply because i dont want to mess with it:)
I prefer to keep Real networks stuff away from my PC.