New modifications

[Gaia]

Try to encode Matrix, it's impossible to make ok 1 cd rip with Xvid.

Dark scenes look real ugly, very blocky. I have really nice Divx3/4 1 cd rips ot that movie, but Xvid,impossible.

Also, i have been testing Xvid quite alot and find out that all these new things(modulated quantization/new 2 pass code)just make picture more blocky, not better. Sure, shorter movies look very nice but try longer ones. Try over 2 h long with lots of dark scenes, impossible to make ok 1 cd rip.

If you test Xvid with short clips or trailers you don't get right impression! Try bitrates under 600 and encode movie with lots of dark scenes and you know what i mean! Don't just check few scenes, whole movie! Bitrates over 700 no problems, try 600 or lower and you see what i mean.

Problems of Xvid is not sharpness(that's what these new modifications do) or speed but blockiness in lower bitrates. So i can't understand what's the point in these new things you guys(-h and Koepi)do. I have one nice 1 cd rip of almost 3 h long movie, i don't want't to try that with Xvid.

First i loved Xvid but then i tried to encode longer movies and actually watched them, i was very dissapointed!

I know Xvid is in early Alpha state so future looks very good but before this blockiness problem is fixed i stick with divx3/4 and upcoming 5.

Sorry about my bit crap english but i hope you understand what i mean.

I am looking forward seeing your comments about this.

[Actron]

i dont want to know how this "nice" 3h hour movie on one cd looks like !

its quite normal that some scenes have blocks if you have a bitrate of less than 600 !!

by the way xvid dont have postprocessing...if you change the fourcc code so your xvid movies use the divx4 decompressor, you wont see any blocks anymore...

[-h]

I'm afraid that a lot of people have been making 1-CD rips, and finding that they look quite nice...

XviD actually excels at lower bitrates, if you compare file sizes at fixed quantizers. But, if XviD isn't your thing, there are always alternatives :)

-h

[Nic]

Are you using the XVID or DIVX DShow filter/fourcc for playback? Use the DivX one for now (if using h.263 quantizer) (no post-processing yet in xvid, but im working on it :)

-Nic

[Ripe73]

no post-processing yet in xvid, but im working on it

I'm waiting:) how long to wait?

[Gaia]

First, i use always divx fourcc, i have tested Xvid.
Ok, that 3 h 1 cd rip is not nice, but not that bad you could expect it to be.

I have made about 10 1 cd rips so far with Xvid, most of them are decent but Matrix seems to be really hard one. Divx3 and Divx4 versions are alot better. Most scenes looks just fine but especially opening one is very blocky. I had similiar problems with Fight Club too.

Yes, but most of people just seems to encode trailers and short clips with over 700 bitrate. I used to test Xvid that way too. It's totally different thing than whole over 2 h movie.

Xvid is nice and i think i will countinue using it but lower bitrate and dark scenes is big problem. Like other people have said.

I have also find out that difference beetween motion search precision 5 and 6 is big, 6 is way better.

I still think that "modulated" quantization or new 2 pass think doesn't make picture quality any better, sure they make it more sharp but that makes blocks more visible too.

If you encode longer movies with darks scenes use Divx3 or 4 otherwise Xvid is ok.

Nic:
I just tried Xvid fourcc again and(i usually use h.263 quantizer) it looks better now.

[Nic]

@Gaia:
Thanks for the info & I hope you keep on testing XVID (I did a test with the film Ronin (slightly shorter than Matrix) and XVID did come out on top(from SBC, DivX4)...but thats about the only real test ive ever done(!))

@Ripe:
Im working on it every night, I have some implementations (my one last night was written from scratch & really improved very low bitrate video, more than DivX's filter did, but thats useless for us DVD rippers :) (Could be useful for real low bitrate streaming though))

[Koepi]

My matrix 1 CD looks very well compared to divx3.

Dunno which settings you used, but they weren't good I'd say :)

Regards,
Koepi

[Gaia]

I have tried lots of different settings and only some scenes are ugly. Really dark ones. Anyway have to test another movie.

[kastro68]

@Gaia

Turn off Lumninance masking!!!

I turned Luminance masking on with one of the early builds of Xvid...I think it was build 7/2/2002...and the grey scenes in Hannibal was totally unacceptable. I have never turned on Luminance masking since, but I do not know if this problem has been solved thus far in the later builds.


AND for one cd encodes, I recommend you use H.263.

For 2cd encodes, modulated works really well.

I would only use Mpeg quantization when you are using a really really high bitrate.


Ja

[Gaia]

Sometimes luminance masking ON gives better results sometimes really ugly.

[kastro68]

I can't say much there...I have only used Lum masking once and it gave me a bad result.


First impressions last, and in my case, for a loooooooooong time. I believe that if you must use lum masking, then you better turn on smooth quantizer fluctuation also.


The one time i used lum masking was for Hannibal...it was unusual what happened. When there was a grey wall, i think of a building, it would go all blocky...but when it showed some moving objects with a face it went all nice and smooth again.

But this was with build 7/2/2002, they prolly fixed up the problem already. But since you are experiencing problems, it would be a long time before i'll try lum masking again. Well at least until I hear some promising posts about it.

The only solution would be to do 3 passes. one 1st pass, 1 pass with lum on, and 1 pass with lum off...and keep what looks best.


Ja


Edit: I had to fix a grammar mistake; experience to experiencing; and i spelt building wrong

[Koepi]

I use no luma masking and now modulated quant_type even for 1CD (matrix).
I'm sure that with the latest additions my actual build will give brilliant results (even better than with my first try with XviD-13022002-Koepi, since now I get a payback delay etc.). I just hope, filesize will be in range - but we've to see that.
I'll keep you posted.

Regards,
Koepi

[Foxer]

I apologize for my messed up asymetric curve compression formula. I was very tired when I thought it up :)

I've made it much much more accurate and later, if -h drops by, he'll commit the new source I e-mailed him.

In the mean time, putting the same value in both boxes for symetric curve compression works fine.

[Foo]

@ Foxer

It's okay, as they say "stuff" happens ;). Thanks for the nice work on the curve treatments.

foo

[gruel]

Originally posted by Gaia
I have also find out that difference beetween motion search precision 5 and 6 is big, 6 is way better.
[/B]

Is it? I don't think so. I have tested both (actually I did the implemention) and from quality 5 to 6 was never difference of more than 1% in filesize at same quantizer. If you have a clip where there really is visual difference, could you please send it to me? Maybe I can use it to optmize quality 6 in general. 

gruel (gruel@web.de)