Avisynth: How about an Optimal Static Script?

audioslave · #1 11-02-2003, 08:34 PM

I've been having difficulties with getting the right video size for my latest encodings

so have been experimenting with some static scripts instead. It would be interesting to know if anyone exept me have been doing the same. And it woulde be nice to see your script and info on your results. The script I'm using for my experimenting at the moment looks like this:

Quote:

MPEG2Source("D:/Rippat/Below/Below.d2v")
ASharp(1, 4)
BicubicResize(448, 416, 0, 0.6, 11, 0, 698, 576)
VagueDenoiser(threshold=1.5, method=3, nsteps=6, chroma=true)
STMedianFilter(3, 3, 1, 1)
MergeChroma(blur(1.50))
MergeLuma(blur(0.1))
BlockBuster(method="noise", detail_min=1, detail_max=10, variance=0.3, seed=1)
DCTFilter(1, 1, 1, 1, 1, 1, 0.5, 0)
AddBorders(16, 80, 16, 80)

Yes, I know the BlockBuster filter isn't supposed to make any difference in this resolution but on this particular movie (Below) it does!

All suggestions on script improvements are welcome.

EDIT: I use ToK for prediction and use 64 for MIN bitrate and 2500 for MAX bitrate.
EDIT 2: How can I add the "scene change blur" from the MA script?

muhali3 · #2 11-02-2003, 09:14 PM

im thinking that you should probably lower the threshold in VagueDenoiser, because there is already some filtering with STMedianFilter. but maybe i could be wrong.

Dialhot · #3 11-03-2003, 05:13 AM

You can use the last script I made for AVI->KVCD and just remove the blockbuster lines. It gives good results.

Perhaps you should add the MergeChroma and MergeLuma to gain some points on the CQ.

Something like:

Code:

BlindPP(cpu=4)
Convolution3D(1, 6, 12, 6, 8, 2.8, 0)
MergeChroma(Blur(1.57))
MergeLuma(Blur(0.1))
GripCrop(WIDTH, HEIGHT, source_anamorphic=false)
GripSize(resizer="LanczosResize")
Undot()
TemporalSoften(2,7,7,3,2)
DCTFilter(1,1,1,1,1,1,0.5,0)
GripBorders()

incredible · #4 11-03-2003, 06:58 AM

Really "sexy" Avi script Dialhot!
Well Im not that much in Dvix/Xvid but I tested it yesterday

.... however I think this should refer to your "V4 Thread"

But in case of general use on DVD Sources I would delete the Convo3d line cause as DVD Sources in general aren't that noisy or encoded incl. produced artefacts, so as it is it plains too much the skins and hairs.
BlindPP() always gives andvantages when encoding from a 8 mod DCT based source.

So ....
(Number 1 (good source))

Code:

BlindPP(cpu=4) 
Deen("a2d",1,5,8) # giving a touch of spatial filtering
MergeChroma(Blur(1.5)) 
MergeLuma(Blur(0.1)) # also very good to give Tempsoften afterwards less to handle!
GripCrop(WIDTH, HEIGHT) 
GripSize(resizer="BicubicResize") # In case of DVD as the Source is sharp and detailed enough and this will also rise the CQ
TemporalSoften(1,5,8,3,2) # Light Temporal Soften in case of less noisy source
Undot() # Undot behind TempSoften to kill a bit the "static" dots in surfaces produced during the temporal soften process
#Blockbuster(method="noise",detail_min=1,detail_max=3,variance=0.1,seed=1) # also activate on DVD source encodings when encoding movies containing big fading surfaces like underwatermovies etc. to avoid edges & blocks in surfaces as known from Jellygoose's experience last week
DCTFilter(1,1,1,1,1,1,0.5,0) 
GripBorders()

... would be a nice Adaption of your scripts to perform on DVD quality sources

@ Audioslave

My experiences showed me that its better not to perform an asharp before cause the filtering routines have to work harder (=loss of details) after this and by the way in case of DVD sources the image is detailed and sharp enough ... so even to compensate scaling afterwards (which I think is the Intention) ... IMHO its not needed, ... by not using it, it'll even give you more CQ! But that's only my suggestion, others also have reported other results

Dialhot · #5 11-03-2003, 07:34 AM

That's true that a spatio-temporal filter followed by temporalsoften if probably too much.

I never tested Deen in spatial mode only (a2d insteed of a3d). It can be a good solution, you're right. I'll give a shot to your script on my next DVD ripping

incredible · #6 11-03-2003, 07:52 AM

Deen("a2d") using light settings on good quality sources is very charming

But if you set it very high in case of high/big noise DVDs, it exponential gives a big punch to actors faces. Not only shaved ... really punched.

And also IMHO this will result in too much artefacts like (sorry) VagueDenoiser too.

But in a case of bad quality DVD streams I would recommend a different combination of filters as I already posted in another thread.

audioslave · #7 11-03-2003, 09:12 AM

Thanks for the feedback guys! I will give incredible's script a workout this afternoon.

incredible · #8 11-03-2003, 10:23 AM

In case of bad DVD Sources DialHot's one should give good/compressable! results
But you can give these ones also a try, Id like to know which of the static DVD source scripts you like (1. above in case of good source and 2. or 3. in case of bad/very bad sources) and it would be nice when posting 'before' and 'after'(still not encoded) Pics in here so I can see on which of "your!" source condition wich result came out, cause I tested it on my sources and it was nice

Number 2 (bad source)
(Color Conversion to YUY2 needed)

Code:

Mpeg2Source("xxxxxxxx.d2v")
BlindPP(CPU=4)  
undot() 
Limiter() 
GripCrop(WIDTH, HEIGHT) 
GripSize(resizer="BicubicResize")  
MergeChroma(blur(1.50)).MergeLuma(blur(0.1)) # in case of not enough CQ after prediction 
ConverttoYUY2()
TemporalSoften(2,3,5,3,2).peachsmoother().Unfilter(60,0) # again, yes I know - but I love it ;-)
Undot() 
# Blockbuster(method="noise",detail_min=1,detail_max=3,variance=0.1,seed=1) # in case of .... as already explained in my script above
GripBorders() 
LetterBox(0,0,16,16) 
Limiter() 
DCTFilter(1,1,1,1,1,1,0.5,0) 
#

Try to set the values of TempSoften optimal for the existing noise factor of your bad DVD source.

Number 3 (very bad source) - nice price/nice noise non-anamorph crap editions ... I encountered some of theese like "Black Rain" as shown in the "how funny" thread

(YV12 preserved)

Code:

mpeg2source("xxxxxxxxx.d2v") 
BlindPP(CPU=4) 
ATC() 
undot() 
Limiter() 
GripCrop(WIDTH, HEIGHT) 
GripSize(resizer="BicubicResize") 
MergeChroma(blur(1.50)).MergeLuma(blur(0.1)) # in case of not enough CQ after prediction
STMedianFilter(3, 3, 0, 0).TemporalCleaner(3, 5).Undot() 
# Blockbuster(method="noise",detail_min=1,detail_max=3,variance=0.1,seed=1) # In case of ... now you know ;-)
GripBorders() 
LetterBox(0,0,16,16) 
Limiter() 
DCTFilter(1,1,1,1,1,1,0.5,0)

NOTE! These last two scripts do work best if resizing not more than 480x576! For DVD (704 or 720) width-encoding there exist an addition cause the filter settings as shown above do work best when not getting over the amount of effective pixels of 480x576 (less the substracted borders recognised by gripfit) which have to be handled by the filters!

Im still developing a spatial/temporal adaptive routine to kill noise most effektive on the one hand and on the other hand to avoid theese "static" dots on slow movements ... we'll see