AutoGK softmatrix test

[Sharktooth]

Starting from 2.07 AutoGK has a new hidden option: .newsoftmatrix
Please watch those 2 encodes, post your comments and your preference.

Old soft matrix (Jawor's 1CD):
http://www.aziendeassociate.com/test/jawors_test_autores.avi

New soft matrix (EQM V3ULR):
http://www.aziendeassociate.com/test/ulr_test_autores.avi

Thanks:)

UPDATE: V3ULR file was incomplete ($#!7 happens)

[BigDid]

Hi Sharktooth,

While your first avi is downloading, wanted to let you know I tried you matrix in AGK 2.07

Source: sat analog capture, 8000kbps vbr, lpcm audio (Indiana Jones)
AGK setting: 892Mb (1/5DVD minus 4Mb) audio 144 mp3 vbr (edited: 137 final), auto width
Agk resulting: 608*336, softer matrix softer resizer, 1010 kbps (video size 785 Mb)
Result comparing to encodes made with old softmatrix(jawor):
less blocks, more details, nearly no ringing -> great improvement

I will continue testing and also compare your 2 avis but already great improvement over Jawor at this bitrate specially concerning the ringing. Great work, appreciate it, glad it can be used in AGK.

Did

[BigDid]

Round2

Same source, 648Mb, 560x304, 832Kbps.
Resulting avi has many more blocks, ringing is back but not as disturbing as with Jawor CQM. I really like this matrix :)

Will try to encode with Jawor's at same lower bitrate, and also dl the second clip.

Did

[BigDid]

Round3

Sharktooth clips (Spiderman trailer)

with V3ULR I see:
+ less blocks, no or nearly no ringing (Bframe 50, left size Columbia woman, hope it is ringing and not ghosting, not sure)
- different saturation or light level (lighter)

I'am really not good to spot small differences, I had to look fullscreen and frame by frame; maybe with a smaller bitrate (700/800 kbps) could be easier to compare

Did

[len0x]

Anyone else?

[niamh]

I find them terribly similar myself. You could toss a coin? :p
Neither of them looks particularly "soft" to me either
Not much use this all is :)

PS: I can only say from the samples Sharktooth uploaded.

[len0x]

The biggest difference between Jawor's and V3ULR are I-frames. P/B frames are very similar but I-frames behave weird IMHO: they are actually trying to be more sharp that original! Its impossible to spot that unless you compare original AVS with final AVI. Jawor's I-frames are the most soft ones, then original, then ULR's which is trying to introduce more details that source actually has...

[Sharktooth]

A CQM cant introduce more details than the original... it can preserve them or drop them...
Btw, since AGK uses only 1 bframe and sets i-frame boosting to 0 i think ULR can be improved for that codec settings.

[yaz]

Originally posted by len0x
The biggest difference between Jawor's and V3ULR are I-frames. P/B frames are very similar ... no surprise. the inter matrices are (almost) identical ;)
Originally posted by len0x
... Jawor's I-frames are the most soft ones, then original, then ULR's which is trying to introduce more details that source actually has... imho, jawor's i-frames seem to be smoother because details are removed more evenly (in a hvs manner) while sharktooth's intra matrix removes much more fine details but much less 'harsh' ones. the latter may seem as a kinda 'sharpening'. in add, both try to to preserve more details in i-frames than in the inter frames, so it also supports this 'sharpening taste'.

btw, if 'softmatrix' why not h263 of hvs-good ?

the bests
y

[Sharktooth]

coz h263 has a much lower compressibility and will force autogk to use a lower resolution. h263 has blocking problems too.
HVS-good has a lower compressibility too and it's not so "good"... (that's why i made ULR).

Originally posted by yaz
no surprise. the inter matrices are (almost) identical ;)
I started from jawors inter matrix (it was doing what i needed) and modified it slightly to have a better compression.

[yaz]

Originally posted by Sharktooth
coz h263 has a much lower compressibility and will force autogk to use a lower resolution. h263 has blocking problems too.can't get what u mean exactly. in my tests these 4 matrices produced (more or less the) same 'compressibility'. see also some further results here (http://forum.doom9.org/showthread.php?s=&threadid=93539). however, they will produce frames looking quite differently, that's true. if 'softness' is considered, h263 a/o hvs-good will produce 'softer' pictures than yours or jawor's, imho.
Originally posted by Sharktooth
HVS-good has a lower compressibility too and it's not so "good"... (that's why i made ULR).w/what respect ? using it for reasonable compression (>=66%) will produce quite nice, soft-looking streams. imho ;)

the bests
y

[Sharktooth]

Originally posted by yaz
can't get what u mean exactly. in my tests these 4 matrices produced (more or less the) same 'compressibility'. see also some further results here (http://forum.doom9.org/showthread.php?s=&threadid=93539). however, they will produce frames looking quite differently, that's true. if 'softness' is considered, h263 a/o hvs-good will produce 'softer' pictures than yours or jawor's, imho.

Not in my tests. h263 forced agk to choose 512*xxx while with jawor's and ulr agk choosen 608*xxx.
Originally posted by yaz
w/what respect ? using it for reasonable compression (>=66%) will produce quite nice, soft-looking streams. imho ;)
Well, i dont like hvs matrices for low bitrates. I tried making an HVS soft matrix and it didnt please me (the first version of ULR was based on HVS matrix). It produced artifacts and smoothed out some "significative" details.