Interesting Avisynth Resizing Thread at Doom9.org - Page 7

incredible · #**121** 03-13-2004, 06:16 PM

Thanx a lot Fupp! Can't wait to test!

kwag · #**122** 03-13-2004, 06:16 PM

Simulcasting Inc

-kwag

FuPP · #**123** 03-13-2004, 06:19 PM

Quote:

Originally Posted by kwag

You got it FuPP

I just scrolled through the section where the artifact was showing, and it's gone bye bye

Thanks,
-kwag

:P

FuPP · #**124** 03-13-2004, 06:21 PM

Quote:

Originally Posted by incredible

Thanx a lot Fupp! Can't wait to test!

Please do ! And let me know...

Regards,
FuPP

kwag · #**125** 03-13-2004, 06:28 PM

Anyone out there running MUTE, you'll find a very small sample called HybridFuPP_084_ffvfw_KDVD_sample.m2v
Or just search for KVCD, and you'll see it plus some of the old samples

Can't wait

, get it here: http://www.kvcd.net/HybridFuPP_084_f...DVD_sample.m2v (It's only 637KB)

-kwag

audioslave · #**126** 03-13-2004, 06:44 PM

Great work FuPP - again!

Will start testing right away.

P.S. Nice clip kwag!

kwag · #**127** 03-13-2004, 10:33 PM

Ok guys, here we go

I made three sample encodes, one with FuPP, another with Deen and one with modified MA script.
All samples encoded MPEG-2 (KDVD) at 704x480 using KVCD Notch Matrix, and TMPGEnc at a CQ of 70.

This are the scripts used:

FuPP script:

Code:

Mpeg2Source("K:\RED_PLANET\VIDEO_TS\red.d2v")
BicubicResize(704, 272, 0, 0.6, 11, 60, 698, 360)
HybridFuPP(704, 272, Preset="High", Chroma=True )
AddBorders(0, 104, 0, 104)
LetterBox(16, 16, 16, 16)

Deen script:

Code:

Mpeg2Source("K:\RED_PLANET\VIDEO_TS\red.d2v")
BicubicResize(704, 272, 0, 0.6, 11, 60, 698, 360)
deen()
AddBorders(0, 104, 0, 104)
LetterBox(16, 16, 16, 16)

Modified MA script (reduced to a single line) :

Code:

Mpeg2Source("K:\RED_PLANET\VIDEO_TS\red.d2v")
BicubicResize(704, 272, 0, 0.6, 11, 60, 698, 360)
ScriptClip(" nf = YDifferenceToNext()" +chr(13)+ "unfilter( -(fmin(round(nf)*2, 100)), -(fmin(round(nf)*2, 100)) ).TemporalSoften( fmin( round(2/nf), 6), round(1/nf) , round(3/nf) , 1, 1)  ") 
AddBorders(0, 104, 0, 104)
LetterBox(16, 16, 16, 16)

The results are interesting

Code:

FuPP - File Size: 2,444KB - Encode Time: 1m:16s
Deen - File Size: 2,410KB - Encode Time: 1m:13s
MA -   File Size: 2,342KB - Encode Time: 0m:48s

Edit: Don't trust encode times, because I have many things running in my PC
But file sizes are correct and consistent, no matter what CPU overhead there is.

Here are the samples:
http://www.kvcd.net/red-fupp.m2v
http://www.kvcd.net/red-deen.m2v
http://www.kvcd.net/red-ma.m2v

-kwag

jorel · #**128** 03-13-2004, 10:52 PM

cool!
i'm downloading to see the quality of each script!

seems off topic but :
i like too much of the deen filter with this parameters:
deen("a3d",3,4,1,4) # light cleaning
why don't use in MA and/or FuPP

kwag · #**129** 03-13-2004, 11:05 PM

Quote:

Originally Posted by jorel

cool!
i'm downloading to see the quality of each script!

I see the quality of FuPP slightly better than MA, and Deen and MA about the same.
If this difference is notable on a TV, I'm not so sure.
But for file size, specially on a long and very active (action) movie, I'm sure the MA will have the smallest file size.
On a drama type movie, the results of file size will probably be about the same with FuPP or MA, but quality of FuPP will be better.

Quote:

seems off topic but :
i like too much of the deen filter with this parameters:
deen("a3d",3,4,1,4) # light cleaning
why don't use in MA and/or FuPP

I just wanted to do a test by stripping down a many components as I could.
Really, the MA, is this single line of code:

Code:

ScriptClip(" nf = YDifferenceToNext()" +chr(13)+ "unfilter( -(fmin(round(nf)*2, 100)), -(fmin(round(nf)*2, 100)) ).TemporalSoften( fmin( round(2/nf), 6), round(1/nf) , round(3/nf) , 1, 1)  ")

-kwag

jorel · #**130** 03-13-2004, 11:25 PM

i see different:

MA and Deen are equal but better matizes,
look the little lights(red,yellow and green)
in the begining of the samples, the color of the skin (face),
the "brown" of the planet and the red light of the "aircraft" (nave,ufo,don't know the name)!

Fupp change the matizes to green and all colors are faded!
(less intensity,washed)
use the same scenes that i used for MA and Deen like posted!

for me, MA and Deen have better matizes than Fupp

kwag · #**131** 03-13-2004, 11:46 PM

Quote:

Originally Posted by jorel

i see different:

MA and Deen are equal but better matizes,
look the little lights(red,yellow and green)
in the begining of the samples, the color of the skin (face),

DAMN jorel, you have the eye of an eagle

I hadn't seen that. Here are screenshots of exactly what you mean:

FuPP:

Deen:

MA:

Quote:

the "brown" of the planet and the red light of the "aircraft" (nave,ufo,don't know the name)!

Fupp change the matizes to green and all colors are faded!
(less intensity,washed)
use the same scenes that i used for MA and Deen like posted!

for me, MA and Deen have better matizes than Fupp

You're correct

I need to go to your eye doctor

-kwag

jorel · #**132** 03-13-2004, 11:52 PM

now you know what means:

"nose on the screen" :0)
in my signature.

kwag · #**133** 03-13-2004, 11:54 PM

Quote:

Originally Posted by jorel

now you know what means:

"nose on the screen" :0)
in my signature.

HAHAHA Yes, now I do

-kwag

fabrice · #**134** 03-14-2004, 02:56 AM

Hi everyone,

And what about replacing the MA part with this lines?

Quote:

MM = last.MotionMask(thY1= 12, thY2= 25, thSD= 33, y=3, u=1, v=1).Expand()
DM = last.Binarize(threshold=40).Deflate()

Dark = last.blur(1.5)
NoMoNoEdge = last.TemporalSoften( 3, 4, 8, 10, 2)
Motion = last.blur(1.2)

layer1 = MaskedMerge(NoMoNoEdge, dark, DM, y=3, u=1, v=1)
layer3 = MaskedMerge(layer1, Motion, MM, y=3, u=1, v=1)

return layer3

(idea from hybridfupp, but with some modifications)
The speed is a lot better, and I have to compare the quality...

Cu
Fabrice

FuPP · #**135** 03-14-2004, 04:24 AM

Kwag : could you provide me the sample ? I'm quiet surprised about colors as I have used colorYUV(analyze=true) to check that and it was ok !

PS : I was wondering if this sample is not a little bit too dark to have an opinion about quality ?

Oh, and you could do a test putting parameters :
preset = "high", N2 ="deen()", Nresizer = 4

or N2 = temporalsoften(something for still scenes)

or (preset = high, ST_Str = 3) too.

and tell us about compressibility and quality of HybridFuPP for the same scenes in these cases ?

The point is that I use a very very light temporal softening, fluxsmooth()1,-1), while you use probably harder values (deen() is quiet strong for instance).

FuPP

jorel · #**136** 03-14-2004, 10:23 AM

FuPP wrote:
" I was wondering if this sample is not a little bit too dark to have an opinion about quality ?"

i could be wrong FuPP, but don't think so, correct me if i'm wrong:
the color can't change the matizes if was used the same scene that have the same luma
no matter if the source is dark or bright, right?

links to download the samples:
http://www.kvcd.net/red-fupp.m2v
http://www.kvcd.net/red-deen.m2v
http://www.kvcd.net/red-ma.m2v

FuPP · #**137** 03-14-2004, 10:29 AM

Hi Jorel,

oh, I was not talking about the chroma problem ! I was talking about the overall quality.

About chroma pb, I need the original sample to take measurements !

FuPP

kwag · #**138** 03-14-2004, 10:57 AM

Hi FuPP,

If you mean from the original VOB, here's a 2 second cut:
http://www.kvcd.net/rp2sec.vob

Also take note on the red area I pointed on the screenshot, there's an artifact that looks like an incorrect colored mapped macroblock, on the lower left part, which is not visible on the samples that were created with Deen or MA.

-kwag

FuPP · #**139** 03-14-2004, 11:35 AM

ok, got it.

It is caused by a bug in the internal command blur(), used to smooth chroma !

I'm now trying to find a workaround...

FuPP

FuPP · #**140** 03-14-2004, 12:59 PM

I'm definitevely stupid ! This not a bug : this is normal, and it happens only on very dark scenes. In that kind of scenes, bluring chroma planes can cause that kind of things (not easy to explain in english for me). I'm looking if I can do something...