HC .01 vs Quenc .59b2 vs CCE 2.67.0.27 vs Recode2 with interlaced source

[onesoul]

I used pal interlaced (edit: Robin Williams Live on Broadway) source encoded at average 4313 kbps, overall reduction of 61.2%.
It's very interesting to analyze the inicial logo of HBO. Don't forget to raise your monitor's contrast and brightness very high to see more details (especially on dark scenes).

I have put a screenshot comparison online here (http://pwp.netcabo.pt/onesoul/enctest.html), it will take some time to load though (around 4mb).

The original m2v file:
http://pwp.netcabo.pt/onesoul/enctest/original.zip (~22mb)

I can try put the encoded files online too, if it's needed.

[Nic]

I hope Nic sees this
Nic sees all :)

Glad to see QuEnc isn't too far behind the others. Did it actually hit the filesize it was supposed to?

-Nic

[dragongodz]

Did it actually hit the filesize it was supposed to?
is anyone besides Lamster having a problem with this ? since Lamster WONT check his bloody PM's it would be nice to find someone to do some testing. it needs to be reproducable though, meaning 0.59B2 ALWAYS undersizes on a clip or film.

[EDIT]
interesting test by the way. very nasty starting with so much coloured noise. :)

[DMagic1]

Nice test.

[onesoul]

Originally posted by Nic
Nic sees all :)

Glad to see QuEnc isn't too far behind the others. Did it actually hit the filesize it was supposed to? Hi, I'm glad you saw it :) The only weird problem Quenc was that fading to black. QuEnc was better on some frames, and HC better on others, but in this little test IMO QuEnc and HC were much better than CCE, (I put Recode there too to have some reference on transcoding, which looked nice except for the logo).

I selected the bitrate of average 4313kbs because it was the selected reduction by DVD-RB for that cell on doing a full dvd backup, so I guess it approaches a real scenario. I only encoded that cell test which came out with the following sizes:

- original: 22.608 kbytes
- recode2: 15.786 kbytes
- hc 0.01: 13.892 kbytes
- cce 2.67.0.27 3pass: 13.823 kbytes
- cce 2.67.0.27 2 pass: 13.803 kbytes (to me the 2 pass was not worse than the 3 pass encode)
- quenc 0.59b2 dc9: 13.578 kbytes (dc9 seemed better than default 8)

- quenc 0.59b2 qlb dc9: 13.650 kbytes (didn't use this encode (which had qlb checked) on test page, but looking more closely maybe I should!)
Originally posted by dragongodz
interesting test by the way. very nasty starting with so much coloured noiseThanks, it was nice to see hc and quenc being better on that nasty start.
Originally posted by DMagic1
Nice test. Thanks :)

[onesoul]

I tried encoding the same cell with cce 2.5 and the outcome of logo was very good, better than any encoder, although strangely sharper than original, even with anti-noise filter activated, although I wouldn't call it noise. For the rest of clip, cce 2.5 didn't do so well, noise was introduced.

The pale red output persists on cce 2.5 though (as in 2.67), is it because of yuy2 processing instead of yv12? I really don't know.

[mb1]

Matrix: mb1 interlaced DV
My matrix has been developed for a completely different scenario :rolleyes:

I'm sure this 'abuse' made CCE worse than it should be ...

[onesoul]

Originally posted by mb1
My matrix has been developed for a completely different scenario :rolleyes:

I'm sure this 'abuse' made CCE worse than it should be ...
Hi mb1, I value your input on this and you are most welcome to criticize it.

So what matrix would you recommend with interlaced pal dvd source? I believe that originally it was shot as ntsc interlaced and then converted to pal interlaced, I don't know if it is relevant or not.

I will run the test again. Maybe I will go with mpeg standard matrix then or even qlb unless you point me a better option.

edit: I see now that I forgot to mention that the original source is the Pal DVD Robin Williams Live on Broadway. Corrected it in linked page too.

[dragongodz]

onesoul - you may want to give the updated QLB matrix a go and see how that does. would be interesting to hear if it performs better/worse/the same as the old QLB(that is in QuEnc) and help me decide whether to keep the updated one or kill it. :)
http://forum.doom9.org/showthread.php?s=&threadid=91821

[freelock7]

The pale red output persists on cce 2.5 though (as in 2.67), is it because of yuy2 processing instead of yv12? I really don't know
I don't think It is through color conversion (4:2:2) that CCE is different (predominant red and blue).
By default, saturation is too high and luma too low.
You can adjust Luma & chroma (coloryuv) to look like QuEnc (QuEnc has warm tints, more than HC):
sat_v = 0.8
cv = - (1-sat_v)*256
ColorYUV(off_y=0, gain_y=7, gain_v=3, gain_u=0, cont_v = cv, opt="coring")

@Dragongodz

You know I like QLB matrix.
It works fine in CCE too.
THX for the update. I'll try it!

[onesoul]

-dragongodz

I tried QLB 1.1 on this source at 4400kbps but the difference was negligible, although zooming a lot on frame comparison along with original I was inclined to choose qlb 1.1 because it seemed to preserve more detail overall (there were some trade backs), but anyway, is it possibly to obtain a perfect matrix?

-freelock7

I tried your suggestion which was fun to try :) but the problem isn't resolved by that, although I could see the clip warmer (warmer than it should be), the particular detailed red isn't affected. Look at frame 357 of test, up center you see a red comercial board, the outcome of cce is always like that, paler, even with your custom settings.

[dragongodz]

at 4400kbps but the difference was negligible
well that bitrate may be a little high to really stress the encoder, depends on source of course. since you had to zoom the frames that suggests no seeable difference in actual play so both ok. :)

is it possibly to obtain a perfect matrix?
no. all matrices are a compremise of some kind.

i will do an updated QLB(1.2) probably tonight which is somewhere between QLB 1.0 and QLB 1.1 so people can try that.

[freelock7]

the outcome of cce is always like that, paler, even with your custom settings.
Did you apply the "converttoyuy2(interlaced=true)" at the end of the script to correct chroma upsample?

[video]

Originally posted by dragongodz
i will do an updated QLB(1.2) probably tonight which is somewhere between QLB 1.0 and QLB 1.1 so people can try that.

dragongodz is there any thumb rule how to design good quant matrices?

[dragongodz]

is there any thumb rule how to design good quant matrices?
thats a question that a long time matrix designer ,such as Didée, could answer much better than i. :)

[onesoul]

Originally posted by freelock7
Did you apply the "converttoyuy2(interlaced=true)" at the end of the script to correct chroma upsample?
Yes! You have the clip available for download, you can try it for yourself. Maybe I am missing something.

[gigaset2004]

@ video

you could find at EDv tip a topic from kika with several points to desgin a matrix.

Imho it is not realy esay -impossible - to find a all in one matrix for all movies.

Short intr. expl.

You can see the matrices like steps to a higher level ; normaly they go up in steps for 4 points - or multplier of 4 -.

At the end of the steps form left to right or from top to botttom they higher at the end of "99" for mpeg matrices.

H 263 go highet but you could only use them in newer multi encoder - FFmpeg -mencoder - hank´s.

Normally the "99" means that the frequency for the picture is "0" like an equilizer.

I could recommend the info at edv top - kika- and i have found zht it is best to use one of the known matrix like QLB.