Xvid sharper than DIVX

[Morbo]

Im a newbie on the board,and would just like to thank the
Xvid team for a wonderful codec.I just spent all week on
one movie and after comparing 5 pass DIVX to Xvid 2 pass
(using the iago's guide)....theres no comparison.

Do a movie that has lots of trees,pavement,stark colors
(Austin Powers)and it makes DIVX look like Realplayer.

IMO of course,but Ive been doing video encoding since capture
cards have been avail and Xvid is my choice from here on out.

Now if you guys/gals can speed up chroma motion :p ,Ill be
thrilled!!!

Also just one question,
I use Gknot(resize(neutral BC),IVTC(decomb4)),AVS,Vdub(fast
compress).
Is 12-20fps normal for 1800+XP/w 256 megs of ram?

Also the movie that gave me a fit was Fire Walk With Me,all
the flashy scenes in the bar....red wall,blue strobes,people
dancing ...worst scene ever to encode w/o blocks somewhere.

Cheers!!

[cult]

did you encode in yv12?

[Morbo]

YUV in DVD2AVI

Is there another way?

[Sigmatador]

avisynth 2.07 or 2.5 alpha ? (search about YV12 in the forum, there're some interesting posts)

[cult]

you can use avisynth with yv12 support which will increase your fps and your overall quality

[Morbo]

Ill look into it,thanks.
I remember ,at one time, having to convert scripts into avi(psudo).
Is that what you guys are talking about?

Just wanna make sure;)

[cult]

yes,the avs file you feed to vd

[Morbo]

I read up on that,but I dont understand.Do I need VirtualDub MOD?
Also,for my use,I dont mind the xtra time to encode.I have a poor laptop for that.

Thanks for the tip anywho...

[lamer_de]

Do I need VirtualDub MOD?
For yv12 use: yes. If you use avisynth 2.5, and you use yv12 capable plugins (mp2dec3 from marcFD, decomb for avisynth 2.5), your output will be yv12 which can give speedups up to 20-30%

CU,
lamer_de

[cult]

now thats a interesting thing that happened to me.A used vd(not vdmod) for testing an avs of yv12.It worked,no crashes,no erros.ffsdshow reports yv12 as output.That made me curious.Tested some old already encoded movies.Ffsdshow reports again yv12 as output even with divx4..Whats going on?I am doing something wrong?

[Koepi]

Wua!

MPEG4 (divx3, xviud, divx4+5, wmv, rv,...) all store the data internally as yv12, so the decoded output is _yv12_ (unless your drivers/vga can't display that as overlay. It's converted to i.e. RGB or YUV2 then).

It's a matter of encoding we're talking about:
MPEG2 is stored as YV12 internally. But old avisynth is requesting YUV2, thus the YV12 data gets interpolated to YUV2. Then the processing chain is done in YUV2 too, and in the video codec it's once more transformed back to YV12.

So we have:
- at least 2 colour space conversions
- 2 times interpolation/downsampling

adding their erros to the video data.

I hope this makes things clear. If in doubt, go to the avisynth forum and read the AVS-YV12 sticky thread.

Regards
Koepi

[Morbo]

Thanks for the Sticky Info there,Im looking into.
Lots of testing to do,

Btw:The post title turned out incorrect,I used differing
resolutions between the two machines I set up batches on.

One was 704 other was 640.

640 had less blocks anyway(of course).

But,looks like Im gonna have to redo the test anyway.

[Morbo]

FYI

I got all the needed things for YU12,it looks more like original for sure.
Big speed increase(when I dont have to IVTC).

Also wanted to let anyone interested know that the XviD-26012003-1 build is superior to the newest one still.

I can run almost(no b frames and GMC yet)all setting with varying results but no anomolies.

I get cleaner files than with the stable build as well.

Cheers!!!

[aharden]

I'd also like to add my compliments to the XviD developers. I just recently upgraded to the AVISynth 2.5 beta with the intent to use the YV12 colorspace for DVD backups. I grabbed one of Koepi's recent unstable XviD builds, and after some crashes in VDubMod and some diagnostics, I figured out that checking "lumi-masking" in the XviD options was the culprit. (I didn't look to see if this was a known issue, but it didn't take too terribly long to diagnose.)

I used Qpel, GMC, and 3 B-frames and did an impressive-looking (relative to DivX 5.03) 1CD encode of "The Bourne Identity". On the AVIsynth side, I used Convolution3D() at "movieHQ" setting prior to a sharp BicubicResize(). It took quite a while to encode (to be expected) but the results were quite watchable. The action scenes in the movie were quite fluid, and in the slower scenes the color-blocking was there but didn't distract me too much. I haven't watched the entire movie yet, but I didn't see any show-stopping artifacts.

Thanks for the great work.

[PowerMacG4]

If you want a good resize, use lanczosresize(int, int)

[peaudepailles]

Hi all,

I've a question concerning Lanczos resizing filter:
I've read that it claims to save more details than Sharp bicubicresize
and therefore its not a good idea to use it for 1cd rips.

What do you think about it?

Thanks.

Peaudepailles

[Manao]

It depends of the movie ( as always ) and of the other filters ( especially fluxsmooth / c3d / dust / temporalsmoother ). Do a compressibility check on the movie with and without lanczos ( but with the other filters you plan to use ), if the result is sufficient ( for your tastes ) with lanczos, use it, else, go for bilinear instead. If you are using b-frames ( no compressibility check yet with gknot if I'm not wrong ), you have to do the entire first pass ( or make a script with avisynth to do a check )

You will find more explanation on this topic in the avisynth's forum.

[peaudepailles]

Thanks, I'll give it a try.
I didnt know that dust was avisynth 2.5 compatible.

Time to test now ! ;)

[Manao]

No it's not yet compiled for avisynth 2.5. In fact, I never tried it, since I switched to avisynth 2.5 two months ago. But it seems very promising, I can't wait to see it ported to avisynth 2.5.

[Koepi]

LanczosResize() is built-in into avisynth 2.5.

Read the documentation (forum rule 1).

Regards
Koepi

[Didée]

Originally posted by Koepi
LanczosResize() is built-in into avisynth 2.5.

Read the documentation (forum rule 1).
Sorry, but at last, Peaudepailles and Manao talked about dust, not lanczos, I think. So Manao's comment was valid.

[Manao]

I should have been clearer. I was speaking about dust.

[Koepi]

Oopsi, my bad, I got confused by the "Lanczos Resize" headline of the post.

My apologies for accusing you of not reading the docs.

Regards
Koepi