Avisynth: Increasing Compression Through Preprocessing - Page 3

kwag · #41 01-16-2003, 11:24 PM

I thought I was joking too

( last night ), until I did some tests, and I couldn't believe it. But I missed

. I joked about 300 minutes

, and in reality, It's 360

-kwag

black prince · #42 01-16-2003, 11:37 PM

@Kwag,

The link to d/l Convolution3d on Doom9 is dead. Is there another
link that's active?

-black prince

kwag · #43 01-16-2003, 11:42 PM

Quote:

Originally Posted by black prince

@Kwag,

The link to d/l Convolution3d on Doom9 is dead. Is there another
link that's active?

-black prince

I just update the "What's hapening" announcement with this link: http://hellninjacommando.com/con3d/

-kwag

jorel · #44 01-16-2003, 11:55 PM

Quote:

Originally Posted by kwag

I thought I was joking too

( last night ), until I did some tests, and I couldn't believe it. But I missed

. I joked about 300 minutes

, and in reality, It's 360

-kwag

what?

i use only dvd2svcd to encode.
can you help me to do this?

at now i don't learn how to use avisynth,sampler,prediction,fitcd..

waiting for muaddib or a complete tutorial !

thanks in advance!

kwag · #45 01-17-2003, 12:02 AM

Hi jorel,

You can set DVD2SVCD with the ULBR (

) parameters.
Set your MIN bit rate to 64kbps. Your MAX to 900Kbps. Set the KVCD BETA-1a "Notch" matrix. The one on all PLUS templates right now. Set the GOP to 1-24-2-1-24.
Then simply add the C3D ( Convolution3D ) .dll and the noisev4.dll (FaeryDust) to your .ini file, just the same way you added SansGrip's filters. You're done

Remember, this is for small TV's ( 5" to 13" displays ) and portable devices. Don't expect VCD quality on large TV's

-kwag

jorel · #46 01-17-2003, 12:26 AM

Quote:

Originally Posted by kwag

Hi jorel,

You can set DVD2SVCD with the ULBR (

) parameters.
Set your MIN bit rate to 64kbps. Your MAX to 900Kbps. Set the KVCD BETA-1a "Notch" matrix. The one on all PLUS templates right now. Set the GOP to 1-24-2-1-24.
Then simply add the C3D ( Convolution3D ) .dll and the noisev4.dll (FaeryDust) to your .ini file, just the same way you added SansGrip's filters. You're done

Remember, this is for small TV's ( 5" to 13" displays ) and portable devices. Don't expect VCD quality on large TV's

-kwag

it's a miracle!

:yippie:

a new saint: San(sgrip)Kwag=San Kwag!

ps:"we need more mad emoticons"
the atuals don't express my mind!

kwag · #47 01-17-2003, 01:35 AM

Here's what "Red Planet" will look like on one CD with KVCDx3 and Legalclip+GripCrop+FaeryDust+C3d: http://www.kvcd.net/redplanet-cq78-faery-c3d.mpg

jorel · #48 01-17-2003, 01:48 AM

GripCrop????

it's new for me...

where i get it?

kwag · #49 01-17-2003, 01:56 AM

Quote:

Originally Posted by jorel

GripCrop????

it's new for me...

where i get it?

It's part of SansGrip's GripFit_preview.dll

Here's the script for that sample:

Code:

LoadPlugin("C:\WINDOWS\System32\mpeg2dec3.dll")
LoadPlugin("C:\encoding\fluxsmooth.dll") 
LoadPlugin("C:\encoding\GripFit_preview.dll") 
LoadPlugin("C:\encoding\blockbuster.dll") 
LoadPlugin("C:\encoding\legalclip.dll") 
LoadPlugin("C:\encoding\sampler.dll") 
LoadPlugin("C:\encoding\vobsub.dll") 
LoadPlugin("C:\encoding\convolution3d.dll") 
LoadPlugin("C:\encoding\dustv4.dll") 
Mpeg2Source("k:\temp\VTS__01_P01_1.d2v")
LegalClip() 
GripCrop( width=528, height=480, overscan=2 ) 
vobsub("K:\temp\VTS_01_VSUB_P1")
FaeryDust()
Convolution3D(preset="movieLQ") 
GripSize() 
GripBorders() 
LegalClip() 

Sampler(length=24) 
## MPEG size = ((Total frames/MovieTimeInMinutes)/24) * MPEG sample file size ##

-kwag

GFR · #50 01-17-2003, 05:23 AM

Quote:

Originally Posted by SansGrip

Quote:

Originally Posted by kwag

What about Flux + FaeryDust

Well, they're both denoisers, so you don't really need to use them together. Faery should catch all the noise that Flux does, and all the noise it doesn't, too

.

Quote:

Or TemporalSmoother + C3D

And these are both smoothers rather than denoisers. I think we want:

Flux or FaeryDust

and a smoother, be it TemporalSoften, TemporalSmoother, C3D, or my as-yet-unnamed smoother/sharpener combo filter

.

I think C3D is more of a denoise/smoother combo, (2D denoising + the other D is temporal smoothing).

Are there any C3D alternatives for us poor souls that don't have a SSE CPU?

black prince · #51 01-17-2003, 08:06 AM

@All,

Tested new avs script for ULBR and here's what my results:

1) Movie is "Unfaithtul" and sampler gave me a file size for CQ=60
.....4,490,448. Used file prediction and it calculated a new CQ=114
.....tried CQ=100. "Unfaithful" is 120 minutes.

2) At Cq=100, picture looked good, just softer and fuzzie, but on
.....a 13 inch screen very good.

3) The time to encode using Sampler was 24 minutes, about 4 times
.....slower than usual. That would mean this movie would take ~23
.....hours to encode the whole movie.

4) I use a CQ of at least 70 or higher to get rid of Gibbs noise and
.....a 3+ hour movie would easily fit 1 CD, but the CQ would not
.... reach CQ=70.

ULBR has to get faster encodes and be able to use CQ 70+ to be useful.
(in my opinion)

-black prince

SansGrip · #52 01-17-2003, 10:45 AM

Quote:

Originally Posted by GFR

I think C3D is more of a denoise/smoother combo, (2D denoising + the other D is temporal smoothing).

It does denoise, but that's more of a side-effect of how it works. I make the distinction that denoisers actually look for noise and try to get rid of it, while smoothers apply various transformations to the clip in the hope that noise will be reduced in the process. Convolution3D applies smoothing in three dimensions, averaging each pixel with (depending on the matrix used) up to 26 of its spatial and temporal neighbours.

Actual denoisers include FluxSmooth, PeachSmoother and FaeryDust, amongst others.

Quote:

Are there any C3D alternatives for us poor souls that don't have a SSE CPU?

You could use TemporalSoften, but don't go too nuts with the parameters -- it's quite aggressive.

SansGrip · #53 01-17-2003, 10:48 AM

Quote:

Originally Posted by black prince

ULBR has to get faster encodes and be able to use CQ 70+ to be useful.

The reason it's so slow is nothing to do with the template itself, but the combination of FaeryDust (extremely slow) and Convolution3D (quite slow). The author of FaeryDust is currently redesigning the motion estimation algorithm to increase the speed.

It may take a long time, but the results are second-to-none.

SansGrip · #54 01-17-2003, 10:51 AM

Quote:

Originally Posted by kwag

Here's what "Red Planet" will look like on one CD with KVCDx3

Outstanding! What really gets me is how stable the backgrounds are... Yes, there are blocks, but when they're not moving around like they used to, one doesn't even notice them.

Makes me wonder if we can get that little bit of extra compression back by using the old matrix...

apoc · #55 01-17-2003, 11:38 AM

Quote:

Originally Posted by kwag

Code:

LoadPlugin("C:\WINDOWS\System32\mpeg2dec3.dll")
LoadPlugin("C:\encoding\fluxsmooth.dll") 
LoadPlugin("C:\encoding\GripFit_preview.dll") 
LoadPlugin("C:\encoding\blockbuster.dll") 
LoadPlugin("C:\encoding\legalclip.dll") 
LoadPlugin("C:\encoding\sampler.dll") 
LoadPlugin("C:\encoding\vobsub.dll") 
LoadPlugin("C:\encoding\convolution3d.dll") 
LoadPlugin("C:\encoding\dustv4.dll") 
Mpeg2Source("k:\temp\VTS__01_P01_1.d2v")
LegalClip() 
GripCrop( width=528, height=480, overscan=2 ) 
vobsub("K:\temp\VTS_01_VSUB_P1")
FaeryDust()
Convolution3D(preset="movieLQ") 
GripSize() 
GripBorders() 
LegalClip() 

Sampler(length=24) 
## MPEG size = ((Total frames/MovieTimeInMinutes)/24) * MPEG sample file size ##

I thought that filters (FaeryDust and Convolution3D) should be after the resize ? I'm wrong ??

SansGrip · #56 01-17-2003, 11:54 AM

Quote:

Originally Posted by apoc

I thought that filters (FaeryDust and Convolution3D) should be after the resize ? I'm wrong ??

They can be either before or after. Personally I'd run FaeryDust after the resize, because it's so slow and I want it to run on as few pixels as possible. Whether to run Convolution3D before or after will depend on the material -- before will give less smoothing (and a smaller compressibility gain) than after.

kwag · #57 01-17-2003, 12:08 PM

Quote:

Originally Posted by SansGrip

Quote:

Originally Posted by kwag

Here's what "Red Planet" will look like on one CD with KVCDx3

Outstanding! What really gets me is how stable the backgrounds are... Yes, there are blocks, but when they're not moving around like they used to, one doesn't even notice them.

Makes me wonder if we can get that little bit of extra compression back by using the old matrix...

You mean the previous GOP ( 1-5823-1-1-24 )

-kwag

SansGrip · #58 01-17-2003, 10:22 PM

Quote:

Originally Posted by kwag

You mean the previous GOP ( 1-5823-1-1-24 )

Nope, I mean the old matrix

. I just finished my two-disc tests with LOTR TT and I had previously used v2 matrix and CQ 71 with 112kbps audio. With the v1 matrix and CQ 71.1 I can use 160kbps audio

with only a tiny loss of quality -- noticible only when stepping through the frames, completely invisible when playing the clip

.

DaDe · #59 01-18-2003, 11:46 AM

Ajas!!

The red planet sample looks "PERRON" and that's for one cd?????

Amazing, but where i can get all the filters that aren't in sansgrip's site?

Thanks a lot!! Keep the good job!

Saludos,
DaDe.

KingTuk · #60 01-18-2003, 06:31 PM

with Avisynth 2.5...
has anyone gotten good results using fluxsmooth and
temporalsoften(3,10,10,scenechange=10,mode=2)
?

I don't think there is a version of dust for 2.5
is there?