digitalFAQ.com Forums [Archives] - Avisynth: SmootherHiQ Vs. Convolution3d

Page 2 of 4

Show 40 post(s) from this thread on one page

digitalFAQ.com Forums [Archives] (http://www.digitalfaq.com/archives/)

- Avisynth Scripting (http://www.digitalfaq.com/archives/avisynth/)

- - Avisynth: SmootherHiQ Vs. Convolution3d (http://www.digitalfaq.com/archives/avisynth/3099-avisynth-smootherhiq-vs.html)

kwag	03-30-2003 11:07 AM

Quote:

Originally Posted by muaddib

Try breaking those images and examining them in PicSwith...
8O You will see a huge difference. :wink:

You're right muaddib 8O, even the colors are messed up (washed out) by SHQ :!:

-kwag

jorel

03-30-2003 11:08 AM

Quote:

Originally Posted by muaddib

Try breaking those images and examining them in PicSwith...
8O You will see a huge difference. :wink:

hi muaddib :wink:

PicSwith...????
where i get it?

thanks!
:D

PyRoMaNiA

03-30-2003 11:43 AM

So SmoothHiQ wasn't so HiQ...back to Dust+Convolution I go! :lol:

kwag	03-30-2003 12:38 PM

Quote:

Originally Posted by PyRoMaNiA

So SmoothHiQ wasn't so HiQ...back to Dust+Convolution I go! :lol:

I guess it is HiQ, because the Hi(Q)uality details dissapear :lol:

-kwag

Jellygoose

03-30-2003 12:41 PM

hey, I'm not the guy who created this filter :D
however I guess this thread has led to something :

Code:

LegalClip() 

mergechroma(blur(1.58)) 

mergeluma(blur(0.05)) 

GripCrop(528, 480, overscan=2, source_anamorphic=false) 

GripSize(resizer="BicubicResize") 

SpaceDust() 

TemporalSmoother(radius=2, strength=3) 

TemporalCleaner(ythresh=5, cthresh=10) 

Convolution3D(preset="movieHQ") 

GripBorders() 

LegalClip()

if this is what you guys go for right now, i'll test it! wait right here! don't move! :wink:

KingTuk

03-30-2003 12:49 PM

test this too... :D

Use instead of C3D

http://www.kvcd.net/forum/viewtopic.php?t=3366&start=45

Jellygoose

03-30-2003 12:52 PM

I will!

Jellygoose

03-30-2003 01:24 PM

:lol: Heavy Testing Here! prepare for very interesting results!! 8O

kwag	03-30-2003 01:34 PM

Quote:

Originally Posted by Jellygoose

hey, I'm not the guy who created this filter :D
however I guess this thread has led to something :

Code:

LegalClip() 

mergechroma(blur(1.58)) 

mergeluma(blur(0.05)) 

GripCrop(528, 480, overscan=2, source_anamorphic=false) 

GripSize(resizer="BicubicResize") 

SpaceDust() 

TemporalSmoother(radius=2, strength=3) 

TemporalCleaner(ythresh=5, cthresh=10) 

Convolution3D(preset="movieHQ") 

GripBorders() 

LegalClip()

if this is what you guys go for right now, i'll test it! wait right here! don't move! :wink:

Change Convolution3D(preset="movieHQ") for Convolution3D(preset="movieLQ") . Details are also well preserved in HQ parameters, but file size is smaller :wink:

-kwag

Jellygoose

03-30-2003 02:03 PM

ok, here are my first results.
First of all : Forget SmoothHiQ :lol:

Kwag, you're script really did beat the SmoothHiQ one by far! In my tests: filesize was smaller, Q-Level was smaller and I did see less artifacts. however I still think that C3d blurrs the picture too much. I'm not talking about details like the grill or something ( :wink: ) I just think that the overall sharpness is gone whenever I use C3d. It's all a matter of taste I guess.

Then I tried STMedianFilter for C3d and in my eyes the results are AWESOME! 8O

Filesize dropped by about 2.5% compared to C3d.
Q-Level dropped by 0.2 points compared to C3d.
Speed increased a little.
And as far as I see it the picture looks simply sharper and less artifacts are seen when motion appears.

this is the script I used. Please try this out and compare it to the script that you think is best.

Code:

Mpeg2Source("F:\Genug\genug.d2v")



LegalClip()



GripCrop (544, 576, Overscan=0)



Gripsize()



TemporalSmoother(radius=2, strength=2)

TemporalCleaner(ythresh=5, cthresh=10)

SpaceDust() 

STMedianFilter(8,15,4,7,8,15)

#Convolution3D(preset="movieHQ")

#smoothHIQ(3,15,25,192,10)



mergechroma(blur(1.58))

mergeluma(blur(0.05))



GripBorders()

Letterbox(0,0,20,20)



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



LegalClip()

Notice that mergechroma and mergeluma are used after the smoothing, since this gives me a lower Q-Level in Bitrate Viewer.
Happy Testing!

PyRoMaNiA

03-30-2003 02:22 PM

In my (very limited :wink:) tests in avscompare, this newest script definitely looks cleaner than the Convolution3D one. Haven't compared file size or anything yet as I'm only viewing these scripts in AVScompare, but so far, looks good.

Jellygoose

03-30-2003 02:43 PM

what did you compare that script to?

kwag	03-30-2003 03:20 PM

Quote:

Originally Posted by Jellygoose

Please try this out and compare it to the script that you think is best.

I'm trying your script right now, one encode with STMedianFilter(8,15,4,7,8,15), and another one with Convolution3D(preset="movieLQ").
I'll post results here.

-kwag

kwag	03-30-2003 03:57 PM

Here it is:

http://www.digitalfaq.com/archives/i.../2003/03/4.png

The difference is almost indistiguisheable, but still, I see a little detail lost. Zoom in so you can see what I mean after you save the picture.

The script used was this:

Code:

GripCrop(528, 480, overscan=2,  source_anamorphic=false )



Gripsize() 



TemporalSmoother(radius=2, strength=2) 

TemporalCleaner(ythresh=5, cthresh=10) 

SpaceDust() 

#STMedianFilter(8,15,4,7,8,15) 

#Convolution3D(preset="movieLQ") 





mergechroma(blur(1.58)) 

mergeluma(blur(0.05)) 



GripBorders() 

Letterbox(0,0,20,20) 



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



LegalClip()

Of course one with STMedianFilter(8,15,4,7,8,15) and the other with Convolution3D(preset="movieLQ")
File size diff was 100KB ( lower with STM ). But I do see a little more artifacts on movement with the STM, but maybe it's my eyes :?

-kwag

Jellygoose

03-30-2003 04:15 PM

I see what you mean. It's barely visible, but it's there. what about the overall sharpness? do you see a difference there?
what does bitrate viewer say? same Q-Level?

kwag	03-30-2003 04:30 PM

Quote:

Originally Posted by Jellygoose

I see what you mean. It's barely visible, but it's there. what about the overall sharpness? do you see a difference there?
what does bitrate viewer say? same Q-Level?

Everything else is basically identical :!:, however that little fault I circled has me wondering what side effects could cause on overall picture quality :idea:. Maybe it can't be seen, but maybe it can be "feeled" in overall perception :?
The Q is about .02 lower (better) with STM, but that's not visually noticeable. I do see a file size difference by changing the order of filters 8O, so probably more tests should be conducted to find out the optimal position :idea:

-kwag

jorel

03-30-2003 04:57 PM

"I do see a file size difference by changing the order of filters."

yes Kwag,
i post (don't remember where..) a few months:

resize first, filters in the end of the script:
less time to encode,more size.

filters first ,resize in the end of the script:
more time to encode,less size.

changing the order of the filters:
everything changes too... is a big confusion,
depending of the filter and the order!!
:!:

dazedconfused

03-30-2003 07:29 PM

Quote:

Originally Posted by kwag

I do see a file size difference by changing the order of filters 8O, so probably more tests should be conducted to find out the optimal position :idea:
-kwag

Eagerly awaiting the outcome on this one! :wink: Gotta love these kinds of threads. :D

And since everyone seems to be in the "testing mood" :wink: , I thought I'd bring up a couple things mentioned by Sansgrip and GFR awhile back in JellyGoose's "Mergechroma to gain Compression?" thread that seem to have gone rather unnoticed:

Quote:

Originally Posted by sansgrip

As someone pointed out earlier, MergeChroma and MergeLuma can be used in combination with any other filters. It's often possible to apply extremely heavy filtering to the chroma components without losing noticible detail, because the eyes are much less sensitive to detail in the chroma components.

For example, while kwag finds PixieDust too strong for DVD sources, it probably isn't if it's only applied to chroma:

Code:
Mpeg2Source("blah.d2v")
...

# Do over-the-top chroma smoothing

chroma = PixieDust()
chroma = chroma.TemporalSoften(...)
chroma = chroma.Blur(...)
chroma = chroma.DctFilter(...)

# Do luma smoothing

FaeryDust()
...

# Now merge the two together

MergeChroma(chroma)
...

Obviously I'm exaggerating somewhat, but you get the idea.

Sansgrip wasn't aware at that time that we're limited to only 1 instance of Dust per script, but still, this could prove useful with other filters to gain a bit of compression, no? Anyone tried it?

also in that thread:

Quote:

Originally Posted by GFR

BTW you can use more complicated things with the Merge filters like:

MergeLuma(Convolution3D(preset="movieHQ"))
MergeChroma(FaeryDust())

:idea: I just thought a few compression-hungry testers might want to revisit that thread and try out these suggestions to see what they bring. :wink:

-d&c

Jellygoose

03-31-2003 12:05 PM

@dazed and confused: I already tried stuff out. Last week I made a couple of tests using heavy filtering on chroma and only light filtering on luma. The result was ok. It was actually the same as it would have been using moderate smoothing on both, and the time to merge the chroma and luma takes VERY long. however i'll start experimenting again later tonight!

Does anyone else find these values used in kwags latest script blur the picture too much? I'll try some more stuff, hang on!

KingTuk

03-31-2003 12:49 PM

This filter order got me the smallest filesize... :D
LegalClip()
BilinearResize(528, 478, 7, 0, 626, 480)
mergechroma(blur(1.58 ))
mergeluma(blur(0.05 ))
TemporalSmoother(radius=2, strength=2)
SpaceDust()
TemporalCleaner(ythresh=5, cthresh=10)
STMedianFilter(8,15,4,7,8,15)
AddBorders(0, 1, 0, 1)
DctFilter(1,1,1,1,1,.5,.5,0)
LegalClip()

you can also move...
mergechroma(blur(1.58 ))
mergeluma(blur(0.05 ))
to the end before you addborders, but the filesize increases a bit...

so keep these together in this order... at least for me...

TemporalSmoother(radius=2, strength=2)
SpaceDust()
TemporalCleaner(ythresh=5, cthresh=10)
STMedianFilter(8,15,4,7,8,15)

Page 2 of 4

Show 40 post(s) from this thread on one page