[Test] DivX 5.2 vs Other Codec

[Sagittaire]

Complete TEST (http://jfl1974.free.fr/HTM/32_Test_Codec_Video.htm)


DivX 5.2 is better than WMV9

DivXNetwork


Not better than XviD, WMV9, VP6 or RV10 with metrics ... :devil:
Sample in dl for 25%, 50%, 75% and 100% compressibility (q2 XviD) if you don't like metrics ... ;)

[V-tec]

@sagittaire

what decoder you use for each codec?

if you deactivate all decoder PP optimizations what result you get?

can you also try to compare DX 5.2 to 5.11 to see what sounds better?

[Sagittaire]

XviD dec for XviD
DivX5 dec for DivX5
FFDSHOW dec for DivX3

PP4 with MPEG4 ASP is the best setting for metric ...

[V-tec]

Originally posted by Sagittaire

PP4 with MPEG4 ASP is the best setting for metric ...

IMHO if you want to compare codecs with metric, it would be more realistic if no PP is used.

[Sagittaire]

- Bad result for VP6 without PP4 for low bitrate
- Bad result for MPEG4 ASP without PP4 for low bitrate
- Very bad result for DivX3 without PP4 for low/medium bitrate

RV10 and WMV9 PP0 without blocking for low bitrate ...

but CO-DEC test is a test with best setting for encoder and best setting for decoder ...

[nicco]

I'm wayting for divx 5 screens

What PP4 or PP1 means, how can know what settings they refers to?

[Sagittaire]

For MPEG4 ASP (and VP6?) Post-Process level

- PP0: No Post Process
- PP1: PP0 + Horizontal Deblocking Y
- PP2: PP1 + Vertical Deblocking Y
- PP3: PP2 + Horizontal Deblocking UV
- PP4: PP3 + Vertical Deblocking UV
- PP5: PP4 + Deringing Y
- PP6: PP5 + Deringing UV


For WMV9 VCM Smoother Post Process

- PP0: No Post Process
- PP1: Light smoother
- PP2: Medium smoother
- PP3: High smoother
- PP4: Heavy smoother (by default)

[CiNcH]

@ Sagittaire,

all four DivX5 Screenshots (from low to extreme) are the same:

http://jfl1974.free.fr/Test/Comparatif/DivX5-low-500.jpg

[nicco]

@Sagittaire
After looked at your tests (why didn't you write them in english?), and red your conclusions, I decided to do a test on my own, using similar settings to yours:
Avisynth

AVISource("\prov.avi",pixel_type="yuy2")
TomsMoComp(0,15,1)
LanczosResize(640,464)
Undot()
FluxSmooth(temporal_threshold=4,spatial_threshold=4)
ConvertToYV12
Convolution3D(0,6,8,6,8,6,0)

Divx5.2:
Encoding
1000Kbps
2pass@ standard-slow
Single Adaptive B-frames
GMC
PSV-slow

Decoding:
Divx decoder
PP5 + Deringing UV (full proc.mode)


*no more available*


RV10 Elysean
Encoded in RealAnime2
I followed your settings except for max b-frames=1 and avg Kb=1000Kbps

Captured in MPClassic (RealPlayer was installed)

*no more available*


It's smooth and there are no blocking artifacts

[Sagittaire]

interval EHQ [85-100] for RV10
Max bframe = 3 for RV10
Rate Control Elysean
and done all setting for RV10 (with job script for exemple)

capture isn't ideal compare mode ... sample svp ... with source if possible

source avi ??? ... for me bad quality for RV10 and DivX with this source and 1000 Kbps ???

Use accessible and clean source like DVD source if you want make comparaison ... !!!

RV10E is very very better than DivX 5.2 ... compare with my Matrix Reloaded Sample. I think that even DigitAl56K will say to you that the RV10 is very better than DivX 5.2 ... :D

[nicco]

I'm not as professional as you, I did this test only for a non-expert purpose

interval EHQ [85-100] for RV10
Max bframe = 3 for RV10
Rate Control Elysean
and done all setting for RV10 (with job script for exemple)


In rate control I have only "curve compression" and "streaming" (I used the first one)
EHQ "Extreme"
What is job script?

source avi ??? ... for me bad quality for RV10 and DivX with this source and 1000 Kbps ???

The source is a PAL-DV-avi, dark and with a lot of noise, I work on PAL-Avi files so I don't care about DVD source...

[Sagittaire]

"Low Motion" and "High Motion" Diaporama ...

Test is complete ... :D

[nicco]

This "bad" quality is due to the imput video, cose I used your settings..
What do you think?

[chilledoutuk]

@nicco i think the screen grab you took of RV10 Elysean shows too much smoothing for my liking a lot of the texture obkects is lost and it looks like it has been artificially smoothed too much.

[nicco]

In fact, I noticed that RV10 smooths the pictures, I'm a newbye in this codec and I don't know so much, but I used the same settings of Sagittaire...
Can help?

[Sharktooth]

It's in the nature of RV to smooth out the details when bitrate is too low.

[nicco]

1000Kbps is too low?

Maybe for this movie...yes!

[crnagora]

Originally posted by nicco
Avisynth

AVISource("\prov.avi",pixel_type="yuy2")
TomsMoComp(0,15,1)
LanczosResize(640,464)
Undot()
FluxSmooth(temporal_threshold=4,spatial_threshold=4)
ConvertToYV12
Convolution3D(0,6,8,6,8,6,0)

It's smooth and there are no blocking artifacts [/B]

Maybe you should change some of Sagittaire Settings, because they may be good with a DVD-Source but not with your kind of video. Try without "FluxSmooth"...also 640x464 is strange...why not 640x480 ?

But you are right...RV10 smooths a lot there...considering the possibility of becoming blind watching the RV10 result you should choose DivX ;) :D

PS.:
Who is the hot chick from the Screenshot ? :D Can we see more of her ? Maybe some clips ? Please ! Doesn't matter which codec you choose :D

[Gannjunior]

@Sagittaire
Bonjour!
How did you get "regular" SSIM values in RV10 tests? i obviously read it by directshowsource...source and encoded file are syncronized, but obtained results are too much low...any suggestion please?
ciao!

[Sagittaire]

Salut ... :-)

Frames tested must be equivalent: you must synchronise video with trim ... better speed with source in lossless (FF1 or Huff in YV12) if it's complexe script

# --> Video Opening <--

source=AviSource("D:\LossLess.avi")
Source=trim(Source,0,35237)
source=ConvertToYV12(source)

video=DirectShowSource("G:\Mes dossiers\B.A\Matrix\RV10-1000.rmvb",fps=25)
video=ConvertToYV12(video)

# --> SSIM analysis <--
return SSIM(source,video,"results.csv","SSIM-RV10-1000.txt",lumimask=true)

++

[Sagittaire]

http://gannjunior.altervista.org/immagini/doom9/test.GIF

Interessing ... psy fast mode decrease PSNR but increase very well SSIM in q2 mode ... ;-)

Don't forget GMC ...

[nicco]

@ crnagora
I don't know


:rolleyes:

I only make video editing, I don't know about source material! :p

[Gannjunior]

@Sagittaire
Yes, P.E. test is very pretty looking and the SSIM evaluation is ok.PSNR in unuseful in presence of psycovisual model...
For problem of time, i have not considered GMC in my tests that now are finished (what yo've seen is only a little part..): so i have to prepare the message with all observations and conclusions with an unexpected surprise at very low bitrate where i had have to eliminate all tests (at 250 kbps) for that codec from comparison (later you'll find which codec... :( ), 'cause there were absolutely unlookable..
if you want i'll give you the link in my ita-forum and if i'll have enough time, i'll translate all in english, 'cause i believe there could be interesting starting points.. ;)

à bientot !

[Sagittaire]

Originally posted by Gannjunior
@Sagittaire
if you want i'll give you the link in my ita-forum and if i'll have enough time, i'll translate all in english, 'cause i believe there could be interesting starting points.. ;)


SSIM and PSNR are universal languge ... ;-)

Post the values which you obtain in a table without commentary ...


i have not considered GMC

GMC is a very good function for improve quality with MPEG4 ASP -> better PSNR and SSIM


Ciao

[Gannjunior]

Sagittaire,

here (http://www.quellicheilpc.com/viewtopic.php?t=16184)

you can find my complete comparison. ;)

ciauzz!