Quantcast Deen vs. Convolution3Dyv12 ? - digitalFAQ.com Forums [Archives]
  #1  
11-07-2003, 03:10 PM
Boulder Boulder is offline
Free Member
 
Join Date: Sep 2002
Location: Lahti, Finland
Posts: 1,652
Thanks: 0
Thanked 0 Times in 0 Posts
Does anyone have any suggestions regarding Deen? C3DYV12 doesn't do any temporal filtering (you can verify this by a simple Subtract() thingie) so Deen would probably be more powerful there - and it has the same C3D matrices incorporated too.

I was thinking of something along the lines of Deen("c3d",0,6,12,3) but has anyone made any tests? The result shouldn't be too blurry if the thresholds are low enough and the filter is quite fast too.
Reply With Quote
Someday, 12:01 PM
admin's Avatar
Site Staff / Ad Manager
 
Join Date: Dec 2002
Posts: 42
Thanks: ∞
Thanked 42 Times in 42 Posts
  #2  
11-07-2003, 04:00 PM
Dialhot Dialhot is offline
Free Member
 
Join Date: May 2003
Posts: 10,463
Thanks: 0
Thanked 0 Times in 0 Posts
How can you view that ? I pretty sure it does but my impression is only based on the delay needed by convolution 3D before statring its effect when you put it in ffdshow. A spatial filter only starts immediatly, a temporal one need a lack of time (that is logical). And C3D enters in the second group.

And secondly, I tested other filters in the script and only spatio-temporal ones gave me good results. As C3D gives also good results, there is only one explanation for me.

Now, for deen, script V1 used it and you can try : it is too heavy on details.
I used it with default values, so you can try with some different one and tell us your impressions.

Note: Deen... fast ? humhum
Reply With Quote
  #3  
11-07-2003, 04:45 PM
Boulder Boulder is offline
Free Member
 
Join Date: Sep 2002
Location: Lahti, Finland
Posts: 1,652
Thanks: 0
Thanked 0 Times in 0 Posts
Deen is fast for an effective spatio-temporal filter, I meant

You can see that C3D doesn't do any temporal filtering with a script like this:

video=AVISource("path\clip.avi")
a=video.Convolution3d(0,6,12,24,24,3,0)
b=video.Convolution3d(0,6,12,24,24,-1,0)
Subtract(a,b).Levels(127,1,129,0,255)

There should be no difference between the two clips whatsoever, I just checked it with several clips.
Reply With Quote
  #4  
11-07-2003, 05:21 PM
Dav88 Dav88 is offline
Free Member
 
Join Date: Nov 2003
Posts: 3
Thanks: 0
Thanked 0 Times in 0 Posts
Found this bit of info at Doom9 under Avisyth Usage-Convolution3D results in yv12 or yuy2 error:



Be aware of the fact that there are three versions of Convolution3D:

1. Convolution3D for AviSynth 2.0.x, supports YUV2 only
2. Convolution3D for AviSynth 2.5.x, supports YUV2 only
3. Convolution3DYV12 for AviSynth 2.5.x, supports YV12 only

No. 3 has not been finished; the "temporal influence" parameter is not being used, and there are issues at scene changes, thus I wouldn't recommend it. Better use one of the YUV2 versions, depending on the AviSynth version you use (preferrably 2.5.x).

bb
Reply With Quote
  #5  
11-07-2003, 05:28 PM
Dialhot Dialhot is offline
Free Member
 
Join Date: May 2003
Posts: 10,463
Thanks: 0
Thanked 0 Times in 0 Posts
Quote:
Originally Posted by Boulder
You can see that C3D doesn't do any temporal filtering with a script like this:

video=AVISource("path\clip.avi")
a=video.Convolution3d(0,6,12,24,24,3,0)
b=video.Convolution3d(0,6,12,24,24,-1,0)
Subtract(a,b).Levels(127,1,129,0,255)

There should be no difference between the two clips whatsoever, I just checked it with several clips.
That's because there is no temporal influence in YV12 (as told in the doc) and the 6th parameter (the one you change from 3 to -1) is the temporal influence value. So you can change it however you want, there is no difference !

I think there is a temporal filtering or... C3d is the slower spacial only filter I know

But we have to find a good way to proof it. Someone has an idea ?

Edit: I didn't saw the post of dav. No proof needed it seems
Reply With Quote
  #6  
11-07-2003, 06:02 PM
incredible incredible is offline
Free Member
 
Join Date: May 2003
Location: Germany
Posts: 3,189
Thanks: 0
Thanked 0 Times in 0 Posts
Send a message via ICQ to incredible
Quote:
Originally Posted by DialHot
Quote:
Originally Posted by Boulder
video=AVISource("path\clip.avi")
a=video.Convolution3d(0,6,12,24,24,3,0)
b=video.Convolution3d(0,6,12,24,24,-1,0)
Subtract(a,b).Levels(127,1,129,0,255)
That's because there is no temporal influence in YV12 (as told in the doc) and the 6th parameter (the one you change from 3 to -1) is the temporal influence value. So you can change it however you want, there is no difference !

Don't we still belive in WYSIWYG? ...so I think "...Subtract(...)...." is the proof.?
Reply With Quote
  #7  
11-07-2003, 06:32 PM
Dialhot Dialhot is offline
Free Member
 
Join Date: May 2003
Posts: 10,463
Thanks: 0
Thanked 0 Times in 0 Posts
Quote:
Originally Posted by incredible
Don't we still belive in WYSIWYG? ...so I think "...Substract(...)...." is the proof.?
It seems that you didn't understand what I said.

The two clips of Boulder example differ only with the 6th parameter (change from 3 to -1). But this parameter IS NOT USED by the filter in YV12 mode.
So that is normal that you do not find a difference between the two clips !

But this parameter is only the threshold that is normally used during temporal filtering to detect scene change. That's all !
Not using it does not mean "there is no temporal filtering at all". That's only means that the scene change sensibility can't be adjusted.
And that is what is explained in Dav88's post.
Reply With Quote
  #8  
11-07-2003, 06:44 PM
incredible incredible is offline
Free Member
 
Join Date: May 2003
Location: Germany
Posts: 3,189
Thanks: 0
Thanked 0 Times in 0 Posts
Send a message via ICQ to incredible
Thats well explained ....
Reply With Quote
  #9  
11-08-2003, 04:07 AM
Boulder Boulder is offline
Free Member
 
Join Date: Sep 2002
Location: Lahti, Finland
Posts: 1,652
Thanks: 0
Thanked 0 Times in 0 Posts
I based my assumption on this (from the C3D docs) :

Code:
Temporal influence :
	It's used especially to speed up a little this filter and to avoid using temporal informations when not needed 
	(scene change, fade, ...)
	I first build a limit = Temporal Luma Threshold * Temporal influence
	For each 2 pixel computed (due to MMX, 2 pixel at the same time), I first check this :
	if 
	(Abs (Y0 - Y0[Previous frame]) +
	Abs (Y0 - Y0[Next frame]) +
	Abs (Y1 - Y1[Previous frame]) +
	Abs (Y1 - Y1[Next frame])) > limit
	then
		do Spacial work (only 3*3 matrix)
	Else
		do Spacial and Temporal work (3*3*3 matrix)
	The lower it is -> the faster will be the filter but compressibility should be lower
	The higher it is -> the slower will be the filter but compressibility should be higher
	if temporal influence is set to -1 then only spatial work is done (high speed).
	This parameter is a float.
There was no difference between the values 3 and -1 naturally, but I wonder if there is a fixed value used every time? I noticed that changing the temporal thresholds do change the output so there is some temporal filtering done.
Reply With Quote
  #10  
11-08-2003, 04:47 AM
Boulder Boulder is offline
Free Member
 
Join Date: Sep 2002
Location: Lahti, Finland
Posts: 1,652
Thanks: 0
Thanked 0 Times in 0 Posts
I posted this on the Doom9 forum as well:

I did a small test with the YUY2 and YV12 versions, 2500 frames of a DivX source, conversion to YUY2 if needed.

Code:
YV12, time 1:29, size 10 662 659, temporal influence disabled
YUY2, time 1:37, size 10 609 956, temporal influence disabled
YUY2, time 2:11, size 10 300 208, temporal influence 3
When I look at those figures, I get a feeling that YV12 does spatial work only. The small differences between the encoding time and file size can be explained by the conversion to YUY2 and using different versions of the filter.

I tried doubling the temporal thresholds with the YV12 version and surprisingly I got a larger file, 10 868 414 bytes

Let's hope someone goes through the source code and finds out how the filter works. If nothing happens, I'll try to contact the author. Too confusing for me to understand right now
Reply With Quote
  #11  
11-08-2003, 07:48 AM
Dialhot Dialhot is offline
Free Member
 
Join Date: May 2003
Posts: 10,463
Thanks: 0
Thanked 0 Times in 0 Posts
By the way boulder, your orignal question was : why do not use deen insteed of c3d. To that the answer is :

1/ C3D gave me the best results so far, that's why "my" V4 script uses it.
No matter is C3D is temporal or not on that point, just the results are important.

2/ test deen, find us the best values and we will do script V5

I will do some test also when I will have a little time.
Reply With Quote
  #12  
11-08-2003, 12:03 PM
Boulder Boulder is offline
Free Member
 
Join Date: Sep 2002
Location: Lahti, Finland
Posts: 1,652
Thanks: 0
Thanked 0 Times in 0 Posts
Yes, that's what I should do. I'll try to find the time next week.

Basically I'm trying to get some extra compression from temporal filtering without removing excess details. Therefore Deen _might_ be a better choice but the testing should show whether it is or is not
Reply With Quote
Reply




Similar Threads
Thread Thread Starter Forum Replies Last Post
O que é Deen() maeliseu Conversão e Codificação de Vídeo (Português) 1 07-08-2004 11:45 AM
Avisynth: A better filter than Deen? Fluffbutt Avisynth Scripting 15 05-21-2004 09:47 PM
Avisynth: New version of Deen() Boulder Avisynth Scripting 0 12-07-2003 04:10 AM
Using the new optimal script with deen? bigggt Video Encoding and Conversion 1 09-07-2003 06:17 PM
Avisynth: Convolution3dYV12 2003/03/29 Beta 5 jorel Avisynth Scripting 2 03-30-2003 08:55 AM

Thread Tools



 
All times are GMT -5. The time now is 06:51 PM  —  vBulletin © Jelsoft Enterprises Ltd