Avisynth: Latest Script Discussion - Page 6

dazedconfused · #**101** 04-05-2003, 11:37 PM

Quote:

Originally Posted by kwag

Try it. I just dropped spacedust

Edit: This is what you get with the script above

: http://www.kvcd.net/sample.mpg

-kwag

Wha-wha-WHAT!!!!

No Dust

You're treading on blasphemy my friend!

But I'll take your word for it Kwag and give it a shot.

Boy, just when you think you know something, some wiseguy comes along and rewrites the whole darn book... :P Thanks,

-d&c

jorel · #**102** 04-06-2003, 01:41 AM

yeah.....
"Boy, just when you think you know something, some wiseguy comes along and rewrites the whole darn book... "

i think that 50,50 for unilter is "too much".

try a sample with 20,20 or 30,30....

i posted in "somewhere over the rainbow"

(in avisynth thread,some sample scripts i think...)
a few months ago that unfilter and temporalsmoother
was the best filters for me used in my samples!
my first kvcd 352x240 mpeg1 lbr was using this filters,
and the colors are better than the news encodes.

but we are trying to do the best,than i change the filters but,
when my son is watching monsters sa encoded with this filters,
i see fantastic brilliant colors....!

J-Wo · #**103** 04-06-2003, 09:39 AM

Quote:

Originally Posted by kwag

If you want to squeeze your video even more, and you don't have an HDTV, use BilinearResize and increase your mergeluma value a little. This will keep a very descent quality, while allowing a lower file size

Wow great tip Kwag. Any other possibilities for a guy with only a 23" TV without S-Video input?

BTW, I know you're an Avisynth 2.0x kinda guy Kwag, but I'd REALLLLLLY appreciate it if someone could maybe reply to your read-only sticky topic with the equivalent 2.5x scripts. Now that you'd dropped SpaceDust I'd think you'd see some real speed increases by switching avisynth versions. But I just got too damned confused with all the ConvertToYUV2 crap and other stuff.

kwag · #**104** 04-06-2003, 09:39 AM

Quote:

Originally Posted by jorel

i think that 50,50 for unilter is "too much".

try a sample with 20,20 or 30,30....

It would be too much if it was used alone, without any kind smoothing after it. But the lines:

temporalsmoother(1,2)
mergechroma(blur(1.58 ))
mergeluma(blur(0.2))

Take care of that, and it's part of the trick

Look at the sample

-kwag

kwag · #**105** 04-06-2003, 09:44 AM

Quote:

Originally Posted by J-Wo

Now that you'd dropped SpaceDust I'd think you'd see some real speed increases by switching avisynth versions.

Yes I agree, specially after reading the comments on 2.5. That's my next move. I'll try AviSynth 2.5 today. The quality I've been getting with the new script on one disk is just so good, that I'm going to re-encode some of my movies again

-kwag

jorel · #**106** 04-06-2003, 10:44 AM

Kwag posted:
"temporalsmoother(1,2)
mergechroma(blur(1.58 ))
mergeluma(blur(0.2))

Take care of that, and it's part of the trick

Look at the sample

"

ok,Kwag!

and i'm reading the comments on 2.5 too....is working better now.

waiting your results!

Kane · #**107** 04-06-2003, 01:19 PM

wow,

the sample looks awesome
i´ve some testings to do

Bchteam · #**108** 04-06-2003, 02:09 PM

I encoded a sample of the Movie "Spygame" using this script and CQ 63

Code:

Mpeg2Source("E:\spygame\spygameproj.d2v")
LegalClip()
GripCrop(480, 480, overscan=2,  source_anamorphic=false ) 
GripSize(resizer="BicubicResize")
VobSub("E:\spygame\SpyGameSubs\VTS_01_0")


TemporalSmoother(radius=2, strength=3) 
SpaceDust()
TemporalCleaner(ythresh=5, cthresh=10)

unfilter(70,70)
STMedianFilter(6,15,4,7,6,15)

mergechroma(blur(1.58)) 
mergeluma(blur(0.7))
 
DctFilter(1,1,1,1,1,.5,.5,0)
 
GripBorders() 
LegalClip()

The Filesize of the Sample was 11,3 MB.

The I used Kwag's script with the same CQ

Code:

Mpeg2Source("E:\spygame\spygameproj.d2v")
LegalClip()
GripCrop(480, 480, overscan=2,  source_anamorphic=false ) 
GripSize(resizer="BicubicResize")
VobSub("E:\spygame\SpyGameSubs\VTS_01_0")
unfilter(50,50)
TemporalSmoother(1,2) 
mergechroma(blur(1.58)) 
mergeluma(blur(0.2))
GripBorders() 
LegalClip()

And the Sample size was 13,7MB

The difference between the first and the second script is 2,4 MB.

With Kwag's script, the picture is a bit sharper with the same CQ.But with the same CQ you won't be able to get the Movie on 1 CD.So you have to decrease the CQ and that will also decrease the Quality.That's why I will still use the first script.

Kane · #**109** 04-06-2003, 02:14 PM

i just made a sample with kwag´s newest script from a movie, which lasts 95 minutes. the result was the sharpest video i´ve ever seen with kvcd and the sample-size is only 10.9MB.
that´s unbelievable

great script

PS: i made this sample with ToK filesize prediction for 1 CD

Bchteam · #**110** 04-06-2003, 02:17 PM

Quote:

Originally Posted by Kane

i just made a sample with kwag´s newest script from a movie, which lasts 95 minutes. the result was the sharpest video i´ve ever seen with kvcd and the sample-size is only 10.9MB.
that´s unbelievable

great script

PS: i made this sample with ToK filesize prediction for 1 CD

How high was your CQ???

Kane · #**111** 04-06-2003, 02:22 PM

cq was 64,82

kwag · #**112** 04-06-2003, 02:38 PM

Quote:

Originally Posted by Bchteam

So you have to decrease the CQ and that will also decrease the Quality.

That's what I thought before too

. Go ahead an lower the CQ to match the file size, and your picture will still look better than the one with the temporal filters

The temporal filters do decrease the file size, but they also spoil the image when used that heavily

I have compared the results on my HDTV, and I won't go back to those temporal filters, unless I have a dirty (badly authored) source or VHS, capture, etc.

-kwag

Bchteam · #**113** 04-06-2003, 04:05 PM

Kwag, it's just unbelieveable.With your new script, I get better results with CQ 52 than with CQ 63 and the old script.

But I have one Question:

Can I retain the sharpness, If add DctFilter(1,1,1,1,1,.5,.5,0) and increase mergeluma(blur(0.2)) to mergeluma(blur(0.7))???Or will it be too blurry???

kwag · #**114** 04-06-2003, 04:19 PM

Quote:

Originally Posted by Bchteam

Kwag, it's just unbelieveable.With your new script, I get better results with CQ 52 than with CQ 63 and the old script.

But I have one Question:

Can I retain the sharpness, If add DctFilter(1,1,1,1,1,.5,.5,0) and increase mergeluma(blur(0.2)) to mergeluma(blur(0.7))???Or will it be too blurry???

I haven't played too much with DCTFilter. I'll have to put that on my schedule

. You're going to have to experiment with that, and see if the speed decrease/file size ratio is worth it.

Also, I think that beyond 0.5 on luma, it starts to blurr the picture too much. Part of the trick was to use a sharpener with heavy value to enhance details (beyond recognition

), then apply the blurr to kill the artificial artifacts. This works good, and I think that with the current value of unfilter and chroma/luma merge, the picture looks very natural and is not over enhanced.

So right now:

Code:

unfilter(50,50)
temporalsmoother(1,2)
mergechroma(blur(1.58))
mergeluma(blur(0.2))

I've found that it's a very good quality/sharpness/file size compromise

-kwag

VILLA21 · #**115** 04-06-2003, 04:41 PM

Quote:

Code:
unfilter(50,50)
temporalsmoother(1,2)
mergechroma(blur(1.5

)
mergeluma(blur(0.2))

Isn"t the image too blurry with these high values 50,50 ?

Just finished an encoding of a TV captured avi, only Unfilter(20,20) plus Bilinear resize gave me an amazing result in less than an hour. Previously i did the same avi with the script u posted on "Optimal scripts" , the one for VHS_Captures, the encoding time was 4.5hours and the result was blocky picture full of artifacts...
Seems Unfilter does a good job even with non-aggresive values...

kwag · #**116** 04-06-2003, 04:51 PM

Quote:

Originally Posted by VILLA21

Quote:

Code:
unfilter(50,50)
temporalsmoother(1,2)
mergechroma(blur(1.5

)
mergeluma(blur(0.2))

Isn"t the image too blurry with these high values 50,50 ?

No. 50,50 is sharpenning. Negative values make the image more blurry. Did you download the sample I posted

-kwag

Bchteam · #**117** 04-06-2003, 04:56 PM

What about this:

Increase mergeluma(blur(0.2)) to mergeluma(blur(0.7)) and Unfilter(70,70) to compensate it.

Bchteam · #**118** 04-06-2003, 05:28 PM

OK,Kwag.

Now I tried this one

Code:

Mpeg2Source("E:\spygame\spygameproj.d2v") 
LegalClip() 
GripCrop(480, 480, overscan=2,  source_anamorphic=false ) 
GripSize(resizer="BicubicResize") 
VobSub("E:\spygame\SpyGameSubs\VTS_01_0")
 
unfilter(50,50) 
TemporalSmoother(1,2)

mergechroma(blur(1.58)) 
mergeluma(blur(0.2))

DctFilter(1,1,1,1,1,.5,.5,0) 
 
GripBorders() 
LegalClip()

With DCTFilter the quality remains the same and I even could raise the CQ from 52 to 56,5.

Kane · #**119** 04-06-2003, 05:44 PM

Quote:

Originally Posted by Bchteam

OK,Kwag.

Now I tried this one

Code:

Mpeg2Source("E:\spygame\spygameproj.d2v") 
LegalClip() 
GripCrop(480, 480, overscan=2,  source_anamorphic=false ) 
GripSize(resizer="BicubicResize") 
VobSub("E:\spygame\SpyGameSubs\VTS_01_0")
 
unfilter(50,50) 
TemporalSmoother(1,2)

mergechroma(blur(1.58)) 
mergeluma(blur(0.2))

DctFilter(1,1,1,1,1,.5,.5,0) 
 
GripBorders() 
LegalClip()

With DCTFilter the quality remains the same and I even could raise the CQ from 52 to 56,5.

damn , i still don´t know, what this filter exactly does, but decreasing file-size is good, i will try it tomorrow

dazedconfused · #**120** 04-06-2003, 05:53 PM

Quote:

Originally Posted by kwag

Quote:

Originally Posted by Bchteam

So you have to decrease the CQ and that will also decrease the Quality.

That's what I thought before too

. Go ahead an lower the CQ to match the file size, and your picture will still look better than the one with the temporal filters

The temporal filters do decrease the file size, but they also spoil the image when used that heavily

I have compared the results on my HDTV, and I won't go back to those temporal filters, unless I have a dirty (badly authored) source or VHS, capture, etc.

-kwag

Anyone have any idea yet what the new minimum "acceptable" CQ range is while using the latest optimal test script with the x3 template? (generally speaking of course). With the old script (using Space and TempCleaner, etc), I think the general consensus was somewhere around CQ=60 was the cutoff.

I did an encode last night of a ~90 minute movie with the x3 template (544x480). I also included some extras that came out to ~105MB muxed, so after all the extras + the movie's audio, my target for video was 607MB. I used the script below, and after prediction, ended up with a CQ of only 51.3!

:

Code:

Telecide()
#Decimate()
LegalClip() 
GripCrop(width=544, height=480, overscan=2 ) 
GripSize() 
Unfilter(50,50)
TemporalSmoother(1,2) 
MergeChroma(blur(1.58)) 
MergeLuma(blur(0.41))
DctFilter(1,1,1,1,1,1,.5,0)
GripBorders() 
LegalClip()

As you can see, I switched to BilinearResize and raised MergeLuma to arrive at my target CQ...but I haven't had time to do any comparison-tests/optimizations of my own yet with this new script to see what I consider acceptable on my Standard TV, so I didn't want to go too overboard with tweaks (this is also why I added DctFilter...the very low CQ level was really starting to worry me, so I started to add filters rather than further lowering CQ...maybe I shouldn't have).

Surprisingly, the resulting movie was very watchable, although I'm not certain whether I'd consider it one of my better encodes. Right now, I'm thinking that to get a 2-hour encode onto 1 disc with the new optimal script (@ 29.97fps and 544x480 as my player requires), it's definetly going to take some further optimization of the Unfilter and MergeLuma levels, and probably even some additional smoothing as well.

With the old methods (using Space, and TC or C3D), fitting ~2-hours onto 1 disc wasn't normally a problem, and I was pretty happy with the results at the time. I'm thinking I may have to use the old script with heavier filtering for 2-hour movies and the new "optimal script" for ~90 minute films. I hope I'm wrong...as always, only testing will tell I guess.

Man, do I hate testing!

If anyone figures out the new general lowest-acceptable CQ, or an optimally tweaked version of the current script for use on a Standard, non-HDTV television, please post your results. Thanks,

-d&c