iron maiden rock in rio, very low compressibility

[ssjkakaroto]

hi there, i'm trying to encode just the first half of this IM concert, which is just about 1 hour, to 650MB but i'm getting a 1st pass of 1.7GB so the quality of the 2nd pass is not so good. I never encountered such a problem in any of my previous encodes so i'm a little lost here. i've tried searching for "inc* comp*" (as in increase compressibility) but it would always time out, if anyone could give me any hints or point me to some threads i would be very thankful ;)

here is my avs:
mpeg2source("d:\im.d2v",idct=7,cpu=4,iPP=true)
fielddeinterlace(full=true, dthreshold=5, blend=false, chroma=true)
crop(10,0,-8,-0)
lumafilter()
undot()
convolution3d(preset="movieLQ")
convolution3d(preset="movieLQ") #I was desperate
unfilter(-5,-5)
BilinearResize(512,384)

my xvid settings (latest koepi build):
I hit Load Defaults and changed the following:
VHQ: 4
Chroma Motion: ON
Max BF: 2
BF Thresh: 40
Chroma Optimizer: ON

TIA

[Koepi]

Qpel can give you a sweet compressability gain, try that as well.

Use unfilter(-10,-10) or (-15,-15) and a SimpleResize(512,384).

The image should look better with that treatment.

Also, I don't believe that a conv3D 2 times wil give you any better result than manually tweaking the values. n very hard cases I use

Convolution3d (0, 16, 22, 16, 22, 0)

or

Convolution3d (1, 16, 22, 16, 22, 0)

_before_ the Undot().

I hope this helps,

regards
Koepi

EDIT: with the default settings for the 2406-build your first pass can get up to 1.3GB, the 2nd pass will still be acceptable (erm, if everything's tweaked right of course). You could also head for SimpleResize(448,336) *whistle* in case nothing else helps...

[Kurosu]

[removed the useless stuff about Convolution3D - see assault post right below this one]

You could try Quantified Motion Filter by Homie FR: your live show is likely to have plenty of fast motion scenes that may not loose to much quality during real time playback with stronger filtering.

[Assault]

@ Kurosu

No that's wrong. Temporal influence and temporal filtering are two different things. Convolution's temporal influence "is used especially to speed up a little this filter and to avoid using temporal informations when not needed.(scene change, fade, ...)" (copied from convolution3d's readme ;))
There's temporal filtering in both the avisynth 2.0x and the avisynth 2.5 version. The YUY version is only a little bit faster.

You'll find more information about that in the avisynth forum. :)

Assault

[ssjkakaroto]

thx a lot guys, i'll try your settings koepi and hopefully i can post the results in about 14 hours when the 1st pass is done :p

[ssjkakaroto]

the first pass is done but the file size was 1.9GB this time, i'll try doing the 2nd pass to see how it'll look but i think i'll need to go to 448x336

heres the script i used:

mpeg2source("D:\im.d2v",idct=7,cpu=4,iPP=true)
fielddeinterlace(full=true, dthreshold=5, blend=false, chroma=true)
crop(10,0,-8,-0)
lumafilter()
Convolution3d(1, 16, 22, 16, 22, 0)
undot()
unfilter(-15,-15)
SimpleResize(512,384)

the xvid settings were the same as before but with QPel ON

[Acaila]

For hard to compresses (noisy) sources I always use the WNR (http://www.avisynth.org/~warpenterprises/files/wnr_20_dll_20020109.zip) filter. You can manually adjust it to the noise level of the source in VirtualDub and use those same settings in Avisynth. If you don't know how to import a VDub filter into Avisynth read my post in this (http://forum.doom9.org/showthread.php?s=&threadid=55570) thread. Combine it with UnDot before and TemporalSoften after and you've got my personal favorite filter chain for difficult sources (no C3D needed with this). Although you do need some patience because WNR is really slow, but IMO it's definately worth it.

[ssjkakaroto]

thx i'll try that too!

edit: after looking in the vdubmod preview window it really looks very compressible now, i'll try it right after the 2nd pass is over

[bond]

you could also give temporalcleaner (http://forum.doom9.org/showthread.php?s=&threadid=37620) a try!
it's a very nice and quite fast temporal noise filter

TemporalCleaner(3,6) produces approximately the same results as conv3d_hq so you could try some stronger ones but too high values can produce ghosting (of course)...

[frodoontop]

Well, I also own the Iron Maiden DVD. And I think it's the hardest to compress video ever seen. With some clips Xvid makes a larger file than the original.

The problem with the source that there is a lot of movement and above that there occurs a scenechange about every two seconds. And I didn't even mention all the light changes and the zooms on the huge jumping public.

So, for reasonable quality you definetely need to go for 3 CD's @ 700MB. And don't use too much filtering: it hurts the image so badly :( .

[ssjkakaroto]

@bond: thx i'll give it a try
@frodoontop: i was amazed too, the original vob was 3GB and the first pass size was 1.9GB, but when i thought 700MB for 40min seemed a bit unreasonable to me so thats why i'm gonna try a few possibilities and see if i can get acceptable quality using only 2cds, if i cant then i'll go for 3cds with 640x480 res and very little filtering ;)