Something must be wrong...

[kecci]

I'm trying to encode the Italian version of Trigun (an anime for those who don't know ;) ) DVD and I'm finding myself in VERY strong difficulties like never before...at 640x480 with XviD 1.0 RC2, cartoon mode, VHQ4 and all those things, i get a 560mb for a 24 min episode first pass :eek: :eek: when my aim was ~230mb/ep...the result is obviously horrible (with a simple script consisting of FieldDeinterlace, UnDot and Convolution3D).
I think that this is due to a strange interlacing of the source. It seems like there are some "edges" on the lines of drawings that move all the way making the codec go crazy. I'm trying to find some filters to solve this problems but I am a newbie in this field and I really don't know where to start from (if from deintarlacing or from edge-softening or deblocking or god-knows-what-ing :( ).
Please could some one enlighten me? :D :D
I'm linking you (http://digilander.libero.it/pmforum/cut.vob) a 10sec cut of the VOB just to make you realize...

TIA everyone for your support! :)

Bye!

EDIT: sorry everyone for my English...I'm Italian :)

[scharfis_brain]

Oh-Oh...

Again and again, fieldblended PAL.

This Anime is a conversion from NTSC to PAL.

the only way to get it correctly deinterlaced, is IMO to use restore24:

http://forum.doom9.org/showthread.php?s=&postid=451901#post451901

[kecci]

Great avatar! :D
Mant thanks, scharfis_brain...I'm trying right now. I didn't understand perfectly the arguments this restore24() (i obviously picked up your modified version).
[]
Tested right now :(
This script

setmemorymax(256)
setworkingdir("E:\divx\gordianknot\")

import("restore24.avs")
import("bobmatcher.avs")

loadplugin("masktools141.dll")
loadplugin("kerneldeint.dll")
loadplugin("avisynth_c.dll")
loadplugin("smartdecimate.dll")
loadplugin("mpeg2dec_.dll")
loadplugin("mpeg2dec3.dll")
loadplugin("unfilter.dll")
loadplugin("sangnom.dll")
loadplugin("tomsmocomp.dll")

mpeg2source("f:\trigun03\cut_proj.d2v",cpu=4,iPP=true)

converttoyuy2(interlaced=true)

restore24(tomsbobsoft(last),bobmatcher(last))

gives this error: "script error: there is no function named "overlay" (bobmatcher.avs, line 89)
(C:\Documents and Settings\Administrator\Desktop\try.avs, line 21)"
What's wrong?

[scharfis_brain]

please update your avisynth!

or replace the overlay-command in bobmatcher

with layer(out1,out2,"fast")

[kecci]

Thank you! With the layer command it seems to work! (updating avisynth to 2.54 I get a strange message about avisynth_new_c and avisynth_c.dll :confused: :confused: ) The whole deinterlacing quality is awesome, but it seems that there are still some duplicates frames...
Look at the cat in this clip (http://digilander.libero.it/pmforum/cat.avi) (encoded XviD 1.0 RC2 q2)...do I need to adjust some thresold or make some frame-rate variation or this is the best i can squeeze out of this DVD?
I used this script to encode the sample

setmemorymax(256)
setworkingdir("E:\divx\gordianknot\")

import("restore24.avs")
import("bobmatcher2.avs")

loadplugin("masktools141.dll")
loadplugin("kerneldeint.dll")
loadplugin("avisynth_c.dll")
loadplugin("smartdecimate.dll")
loadplugin("mpeg2dec_.dll")
loadplugin("mpeg2dec3.dll")
loadplugin("unfilter.dll")
loadplugin("sangnom.dll")
loadplugin("tomsmocomp.dll")
loadplugin("convolution3d.dll")
loadplugin("undot.dll")

mpeg2source("f:\trigun03\trigun_9.d2v",cpu=4,iPP=true)

converttoyuy2(interlaced=true)
restore24(tomsbobsoft(last),bobmatcher(last))
crop(20,6,688,564)
converttoyv12()
Convolution3D(preset="animeLQ")
UnDot()
UnFilter(-5,-5)
LanczosResize(640,480)

Again, thank you for helping me! :)

[scharfis_brain]

you will have to live with the blended cat.
(I've noticed this with your test-sample)

It is a just to small feature, compared to the whole picture.
This is, why it isn't catched correctly.


But I think, that this small issue won't be that noticable.

[kecci]

Ok...the deinterlacer is now working perfect! (far better than the FieldDeinterlace I used before :) ) but the problem is still there. First pass is ~510mb (for a 24min ep with no audio :eek: :eek: ) using Convolution3D (animeLQ preset). Could you advice me some filters like Conv3D to reduce the filesize by cut off blocks and artifacts and/or sharpen the image?
Right now I'm messing up with MSmooth and later I'll give Dup a try :rolleyes: :rolleyes:

[manono]

Hi-

One way to improve the quality is to lower the resolution to 512x384. Even the NTSC version doesn't compress so well. In my opinion that beats filtering the hell out of it, and it'll encode faster as well.

[Bogalvator]

Hey kecci check out the 'Speed and Compression of filters' sticky in this forum.

Seems like Faerydust() and PeachSmoother() might be what you looking for to increase compressibility.

Also manono's suggestion is a good one - sometimes you just have to cut your losses and lower the resolution and 512x384 will still look fine.

[kecci]

@all: Thanks to everyone for your advices! :) :)
I'm almost there :D ...I've achieved a ~280mb first pass using MSmooth and BilinearResize-ing (instead of Lanzcos) to 512x384px! The video I want should be around 195-200mb, so I'm very very close.
Just to inform you...during my maaaaany attempts I tried using DeShaker to make the image more still and save bits...and the results were not bad at all!! :) :) Now I've discharged this option because of its slowness (it would require 4-three-hour-passes to make a episode!) but someday I will try it combined with Dup and I'll see where I can get :)

@Bogalvator: I already tried PeachSmoother() but at default values, MSmooth gives better compressibility. So now I'm sticking with it, but I'll give FaeryDust() a try ASAP! ;)

Bye and thanks again!