XviD, DivX 5.1, DivX 5.1.1, DivX 5.05, WMV PSNR test

[kalxas]

Hello everyone!

I 've been doing some PSNR tests lately. The test was performed over an Athlon XP 1700+ and the source video used was the Dodjo fight scene from the Matrix. I used a bitrate of 700Kbps. Let the numbers speak for themselves:

DivX Pro 5.05 - Slowest mode, no B 21.1 fps
Average
PSNR: 42.2774
----------------------------------------------------------------
DivX Pro 5.05 - Slowest mode, B 15.6 fps
Average
PSNR: 42.2555
----------------------------------------------------------------
DivX Pro 5.1 - Standard mode, no B 18.2 fps
Average
PSNR: 42.3460
----------------------------------------------------------------
DivX Pro 5.1 - Standard mode, B 14.6 fps
Average
PSNR: 42.3805
----------------------------------------------------------------
DivX Pro 5.1.1 Beta1 - Standard mode, B 17.4 fps
Average
PSNR: 42.1855
----------------------------------------------------------------
DivX Pro 5.1 - Standard mode,B,MV 17.7 fps
Average
PSNR: 42.2985
----------------------------------------------------------------
DivX Pro 5.1 - Standard mode,B,PV Fast 11.2 fps
Average
PSNR: 42.3427
----------------------------------------------------------------
DivX Pro 5.1 - Standard mode,B,PV Slow 9.9 fps
Average
PSNR: 42.2198
----------------------------------------------------------------
DivX Pro 5.1 - Standard mode,B,GMC 15.6 fps
Average
PSNR: 42.3378
----------------------------------------------------------------
DivX Pro 5.1 - Standard mode,B,Qpel 5.5 fps
Average
PSNR: 42.1700
----------------------------------------------------------------
DivX Pro Kauehi - Standard mode, B 10.7 fps
Average
PSNR: 42.3860
----------------------------------------------------------------
XviD Dev-Api 4 - VHQ off, no B 19.0 fps
Average
PSNR: 41.9031
----------------------------------------------------------------
XviD Dev-Api 3 - VHQ off, no B 21.8 fps
Average
PSNR: 41.8640
----------------------------------------------------------------
XviD Dev-Api 4 - VHQ off, B=1 16.4 fps
Average
PSNR: 42.0710
----------------------------------------------------------------
XviD Dev-Api 3 - VHQ off, B=1 16.5 fps
Average
PSNR: 42.0470
----------------------------------------------------------------
XviD Dev-Api 4 - VHQ Mode, no B 14.7 fps
Average
PSNR: 42.0776
----------------------------------------------------------------
XviD Dev-Api 3 - VHQ Mode, no B 13.7 fps
Average
PSNR: 42.0272
----------------------------------------------------------------
XviD Dev-Api 4 - VHQ Mode, B=1 14.4 fps
Average
PSNR: 42.2652
----------------------------------------------------------------
XviD Dev-Api 3 - VHQ Mode, B=1 13.6 fps
Average
PSNR: 42.2241
----------------------------------------------------------------
XviD Dev-Api 4 - VHQ Limited, no B 9.5 fps
Average
PSNR: 42.1873
----------------------------------------------------------------
XviD Dev-Api 3 - VHQ Limited, no B 9.1 fps
Average
PSNR: 42.1453
----------------------------------------------------------------
XviD Dev-Api 4 - VHQ Limited, B=1 10.9 fps
Average
PSNR: 42.3371
----------------------------------------------------------------
XviD Dev-Api 3 - VHQ Limited, B=1 9.9 fps
Average
PSNR: 42.3010
----------------------------------------------------------------
XviD Dev-Api 4 - VHQ Medium, no B 7.1 fps
Average
PSNR: 42.1966
----------------------------------------------------------------
XviD Dev-Api 3 - VHQ Medium, no B 6.8 fps
Average
PSNR: 42.1588
----------------------------------------------------------------
XviD Dev-Api 4 - VHQ Medium, B=1 8.9 fps
Average
PSNR: 42.3481
----------------------------------------------------------------
XviD Dev-Api 3 - VHQ Medium, B=1 8.4 fps
Average
PSNR: 42.3130
----------------------------------------------------------------
XviD Dev-Api 4 - VHQ Full, no B 5.5 fps
Average
PSNR: 42.2447
----------------------------------------------------------------
XviD Dev-Api 3 - VHQ Full, no B 5.2 fps
Average
PSNR: 42.2108
----------------------------------------------------------------
XviD Dev-Api 4 - VHQ Full, B=1 7.2 fps
Average
PSNR: 42.4025
----------------------------------------------------------------
XviD Dev-Api 3 - VHQ Full, B=1 6.6 fps
Average
PSNR: 42.3705
================================================================
DivX Pro Kauehi - Slow mode, B 1.9 fps
Average
PSNR: 42.6918
----------------------------------------------------------------
DivX Pro 5.1 - Slow mode, B 1.8 fps
Average
PSNR: 42.6977
----------------------------------------------------------------
DivX Pro 5.1.1 Beta1 - Slow mode, B 2.3 fps
Average
PSNR: 42.5038
================================================================
DivX Pro Kauehi - Slowest mode, B 1.5 fps
Average
PSNR: 42.7041
----------------------------------------------------------------
DivX Pro 5.1 - Slowest mode, B 1.5 fps
Average
PSNR: 42.7101
----------------------------------------------------------------
DivX Pro 5.1.1 Beta1 - Slowest mode, B 2.2 fps
Average
PSNR: 42.5186
----------------------------------------------------------------
Windows Media Video 9 - Complex 2.0 fps
Average
PSNR: 43.0437
================================================================

One comment over the results: DivX 5.1 at slowest seems visualy to be BY FAR BEST of all.

[ookzDVD]

You should include the RV9 for the test, and you will know the answer.
:) I think you know why.

[Sirber]

@ookzDVD

Are you on the payroll of a well known compagny too? ;) lol

[CruNcher]

Rv9 will allways be the best in PSNR tests. But that doesn't says anything because the detail preservation is bad but the smooth edges are fantastic so its a question of what you like more and not what a dumb value is telling you ;)

[Sirber]

RV9 EHQ keeps much more details than before, and edge are better in RV9 than MPEG4. Viva muskito noise!!! :(

[CruNcher]

RV9 EHQ keeps much more details than before


hehe that might be true but still less then i like :D

[kalxas]

There will be an update soon and RV9 will be included.
Thanks for your sugestions.

[Leolo]

Hi people,

I'm mostly interested in measuring detail preservation. I understand, from what CruNcher said, that PSNR tests aren't reliable for that purpose.

But what about JND metrix tests? Do they test detail preservation better?

Thanks.

[cipher]

Rv9 will allways be the best in PSNR tests.
I didn't know that...even in high bitrates?:-) I know rv9 is 2nd to none in low bitrate encodings(what I mean low bitrate is below 1000kbps), but with fairly high ones like above 1500 or 2000 kbps, can rv9 still beat XviD or DivX 5.1?

btw, thx for your massive hard work, kalxas! :-) And a little comment, when you try to compare psnr, please pay more attention to the PSNR for Y channel(imo, you may even compare psnr for Y only in many cases), coz luminance is more sensitive to human eyes. In addition, could you also include filesizes for every encoded clips? You know, a clip with larger psnr but also a larger filesize does not necessarily mean it has higher quality than a clip with lower psnr result but also a smaller size. :-)

[kalxas]

All clips were 21.9 MB in final size except WMV9 clip which was 22,2MB.
You are right, I should have mentioned it... Thanks :)

[bilu]

If RV9 had a grain filter for playback everybody would love it :)

Bilu

[Soulhunter]

"Rv9 will allways be the best in PSNR tests. But that doesn't says anything because the detail preservation is bad but the smooth edges are fantastic so its a question of what you like more and not what a dumb value is telling you"

- by CruNcher
"Not everything that can be counted counts, and not everything that counts can be counted"

- by Albert Einstein

LOL !!! I really like it when a qoute exactly matches my opinion... :D

Bye

[CruNcher]

Originally posted by cipher
I didn't know that...even in high bitrates?:-) I know rv9 is 2nd to none in low bitrate encodings(what I mean low bitrate is below 1000kbps), but with fairly high ones like above 1500 or 2000 kbps, can rv9 still beat XviD or DivX 5.1?

i don't know that myself cause i don't do so called "high bitrate" encodes but from my experience i would say no in high bitrate situations RV9s agressive inloop filtering should lose vs mpeg4 but im not that sure about it @ least if detail preservation is the same for both situations maybe inloop filtering is triggered in Rv9 i dont know have no info about this. And the 2nd to none thing i think i said it 1000 times before, it's a question of what you prefer or maybe what you need if you need a clear detailfull picture and dont mind some sloppy edges and mosqito noise (old analog tv's blure it anyway) and also aren't offended by some blocks that occoureing from time to time (those btw you wouldn't recognize if you tried to preserve as much bits as possible on low motion scenes),(because they are alot more recognized their) mpeg4 is the way to go if you like more clear edges but don't need so much details in low motion scenes or the oposite need much more details in high motion scenes (no blocking) i think Rv9 will be your favorite or looking films in front of your monitor or flatscreen :sly: the more action stuff you have a longer time period say 3h nonstop action the more i would tend over to RV9 but if its still balanced i prefer mpeg4 this only reflects my personal opinion for very very low bitrate encodes 1 CD :)

ah btw before i forget it for those who prefer RV9 you should really follow the development of Nicolas Rududu Codec im still impressed about the encoding speed and the visual quality it produced detail preservation and edges are onpar with RV9 :)

[bilu]

Originally posted by CruNcher
ah btw before i forget it for those who prefer RV9 you should really follow the development of Nicolas Rududu Codec im still impressed about the encoding speed and the visual quality it produced detail preservation and edges are onpar with RV9 :)

How is it going on the bitrate against RV9 ?


Bilu

[CruNcher]

Originally posted by bilu
How is it going on the bitrate against RV9 ?


Bilu

last thing i know of was a simple rate controll but its not that good for low bitrates i had to use the node system to reach the 550 kbps as the rv9 sample but the result was pleasing :) and it was hella fast in encoding i think it was like 3x faster or even 4x im not sure anymore but its fast and for a wavelett codec it looks not that wishy washy more RV9 alike :) i hope Nicolas continues development

[ookzDVD]

I always rip DVD into 1 CD, so RV9 is my choice at this moment.
I've been using XviD before but after I do my personal research,
finally I decide to use the RV9.

Ripping DVD to 2 CD doesn't make sence for me since I can get the
p*rate DVD9 for only USD 1.0 :) the same copy as the original one.

Lol!

[Joe Fenton]

We do people continue to do PSNR tests? They mean nothing. As indicated in the initial post, the subjective winner was not the codec with the highest PSNR. You'll find that in test after test... people will pick a "winner" which scores only average on PSNR compared to other "better" codecs. I mean, look at the results! They are all so close to each other, the difference is meaningless in any event! It's like a track race... the winner beat the competition by 3/1000 ths of a second... big friggin' whoop.

That is one reason someone developed the SSIM test. It seemed to give a better "quality" indication than PSNR as demonstrated with test subjects. More time should be devoted to tests like that. Let PSNR die and fade away. It's time has passed.

[CruNcher]

@Joe Fenton

i personaly neither count on PSNR or JND or SSIM or whatever i count only on my eyes and brain best combination their is :)
but nevertheless most the times and even in the last one "new compare" thread i can say that those psnr values without pp reasamble my visual feeling of the video exactly even those litle difference can bee seen maybe its because i doing since several years nothing else as watching compressed video starting back with :P graspsprite :P
ok wasn't really a compression format but hey still moving pictures ;)
and the greatest disadvantage of SSIM is its not aware of motion as my brain is.

for example


qpel+vhq4+cm+trellis+b-frames: PSNR 40.1454, SSIM 0.952205 ,8.107.008 bytes, 00:08:35.730
gmc+vhq4+cm+trellis+b-frames: PSNR 40.1558, SSIM 0.954120 ,8.105.472 bytes, 00:08:34.047
qpel+gmc+vhq4+cm+trellis+b-frames: PSNR 40.2188, SSIM 0.952830 ,8.106.496 bytes, 00:12:05.997


i dont agree here for me the qpel+gmc combination looks more detailed resambles the original the most the GMC only is more blurry details preservation is not as good