Quantcast Avisynth: Anyone Try Fmf? - Page 3 - digitalFAQ.com Forums [Archives]
  #41  
11-20-2003, 08:48 PM
ak47 ak47 is offline
Free Member
 
Join Date: Oct 2002
Posts: 168
Thanks: 0
Thanked 0 Times in 0 Posts
But doesn't it make sense that lower motion parts that don't generate any or much noise can be filter with something more simple and less costly of time and detail, and stuff with higher motion can be filter with something that focuses more on filtering in more then one way or better way that cost more time but removes more noise with almost the same detail so its still not noticeable and saves more space.
__________________
Later ak
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
  #42  
11-20-2003, 10:14 PM
digitall.doc digitall.doc is offline
Free Member
 
Join Date: Jul 2003
Location: Valencia (España)
Posts: 741
Thanks: 0
Thanked 0 Times in 0 Posts
Hi Kwag:
I'm pleased you came to help us . Even if after all this testing we come to the decission (most probably) that we'd better stay as we are now, we'll have learnt a lot, don't you think so.
I don't know how I can dare to discuss you a thing , me ignorant. And above all, because I don't have the knowledge, and I just speak from a theoretical point of view. But I think I disagree. In the MA script we have two motion levels: below and above SwitchThreshold. Below threshold we apply (linearly) TemporalSoften, and Unfilter above. And there's a sharp change between filters. I'm trying to reproduce the nice results of MA in FMF. Sometimes I make use of the same filter in low/medium action, making the parameter1_lm_max the same as parameter1_mm_min, so it's like a linearity (not linear at all because the proportion of the range applied to the filter is calculated with spline, that I think is not a linear function. But I'm not sure of this... even of anything) from low to medium. And I have the chance to heavy temporal in really low action (still testing this) and heavy spatial in really high action (and we could solve the problem I've noticed sometimes with MA: too blurry in some ocasions). All we need is some help to explore this way. But if you think it's a waste of time, because it have already been explored, with no success, maybe it's time to forget it .
Waiting your <<<always interesting>>> posts.
Reply With Quote
  #43  
11-21-2003, 12:59 AM
kwag kwag is offline
Free Member
 
Join Date: Apr 2002
Location: Puerto Rico, USA
Posts: 13,537
Thanks: 0
Thanked 0 Times in 0 Posts
If the MA script is too blurry for some, just increase the value of the trigger treshold (SwitchThreshold), so that it kicks in at a higher motion point. This way, only on very high action, blurring will occur.
As for the switching points, yes, the current MA script has only two levels. But that's far better that having three points, where filter glitches can occur, and create artifacts.
Ideally, there shouldn't be any switching points
That is, the filters would work "back to back", so the filters are all present at all times. That is, when low to high motion is detected, temporal is decreased as blurring is increased. When falling action, the reverse occurs.
I'm currently revising the MA script to do just that. This way, there is NO switching point, because all filters are "ON" at all times, but just working "inversely proportional". Sort of a "phase locked loop" design.
I hope this explanation is clear enough
If not, ask again

-kwag
Reply With Quote
  #44  
11-21-2003, 04:23 AM
digitall.doc digitall.doc is offline
Free Member
 
Join Date: Jul 2003
Location: Valencia (España)
Posts: 741
Thanks: 0
Thanked 0 Times in 0 Posts
Hi Kwag:
clear explanation, it,s just me, a little dumb .
I already tried to raise SwitchThreshold to 6, 8, even to 10. It improved, but in high action scenes, with still parts in the scene, it keeps the same (and I guess it will also happen with FMF).
I don't understand what filter glitches mean, that may create artifacts.
I followed your link to another KVCD thread (motion adaptive filtering now possible), and even I didn't read it hole, I think that wasn't the same as FMF. In FMF we can put the same set of filters in the three motion levels, with parameter range overlaping in the thresholds, so the filters are acting always in the three levels, at different parameters (I'm not exploring a lot this way, but I tried it once a looked well). And as you say, filters al always on, at different filtering level (I even tried unfilter in sharpening parameters in low action, but I think wasn't better, and of course bigger sized files).
I'm afraid it looks like I'm insisting too much in this kind of filtering. Don't misunderstand me, I'm just inviting you to explore another way , if you think worth it. My reference is always the MA script, even if I try FMF, I try it substituting the adaptive part, keeping the rest of the script (and I even employ the same filters as in the adaptive part: TemporalSoften and Unfilter). It's not I don't like MA script, it's just a question of attitude in life: always trying to improve (if possible) the very good. Not questioning how good MA is... since I tried it, I'm still .
For your answers I see you already tried it: any advise on how to adjust parameters, what filters to use, change thresholds,... ?
See you in forums
Reply With Quote
  #45  
11-21-2003, 06:14 AM
kwag kwag is offline
Free Member
 
Join Date: Apr 2002
Location: Puerto Rico, USA
Posts: 13,537
Thanks: 0
Thanked 0 Times in 0 Posts
Hi digitall.doc,

Hey, there's nothing wrong with experimenting
And I'm always open to suggestions. So if the FMF works for you, go ahead and use it. If the current MA script works, use it too. It's a matter of choices
If you look at the latest MA script ( as of last night ), you'll notice some changes. It's now completely linear. It will be slower, but the compression should now be even higher than before, because the filters are now on for every frame. Just like in a static filter script. The only difference is that the more the activity, the more the "unfilter" part will work and the less the "TemporalSoften". And vice versa. The only thing left is tunning parameters to the filters.
I'm currently testing the new MA script, and now even at 720x480 I think "Red Planet" will fit on one CD-R
CQMatic is currently calculating at CQ=65.95 (started at 60), and if I recall correctly, my last "Red Planet" at 528x480 with the old MA script was encoded with a CQ of ~63.x
I'll put up a sample clip as soon as prediction ends.

-kwag
Reply With Quote
  #46  
11-21-2003, 06:53 AM
digitall.doc digitall.doc is offline
Free Member
 
Join Date: Jul 2003
Location: Valencia (España)
Posts: 741
Thanks: 0
Thanked 0 Times in 0 Posts
Thank you, feeling better now.
And, what about suggestions?. I know (well, I guess) you have lots of things in mind, I don't want to disturb. But it would be very useful if you suggested something about FMF parameters, or even less, something in general about filtering, and parameters. Although I suppose, from what you're changing in MA, that your advise is linear filtering with two filters. But, can you imagine a way to take profit of this three level system?.
Reply With Quote
  #47  
11-21-2003, 04:18 PM
ak47 ak47 is offline
Free Member
 
Join Date: Oct 2002
Posts: 168
Thanks: 0
Thanked 0 Times in 0 Posts
Kwag I need your opinion. Is working on the FMF script a waste of time and will never get better than MA script, or should I try some more to see all the capabilities of more optional motion adaptive to see that if its worth the having more options.
__________________
Later ak
Reply With Quote
  #48  
11-21-2003, 05:39 PM
kwag kwag is offline
Free Member
 
Join Date: Apr 2002
Location: Puerto Rico, USA
Posts: 13,537
Thanks: 0
Thanked 0 Times in 0 Posts
Quote:
Originally Posted by ak47
Kwag I need your opinion. Is working on the FMF script a waste of time and will never get better than MA script, or should I try some more to see all the capabilities of more optional motion adaptive to see that if its worth the having more options.
Hi ak47,

Of course it's not a waste of time
It's a matter of choices, as I posted earlier
Not every movie is the same, so probably specialized scripts will perform better on a movie than the MA script.
I've never said the MA script is better than other scripts. It's just a reference script that performs well, and it's purpose is to be a general purpose script

-kwag
Reply With Quote
  #49  
11-21-2003, 07:04 PM
ak47 ak47 is offline
Free Member
 
Join Date: Oct 2002
Posts: 168
Thanks: 0
Thanked 0 Times in 0 Posts
Cool I guess I will continue on my hunt, and try to be the almighty kwag .
__________________
Later ak
Reply With Quote
  #50  
11-21-2003, 08:18 PM
kwag kwag is offline
Free Member
 
Join Date: Apr 2002
Location: Puerto Rico, USA
Posts: 13,537
Thanks: 0
Thanked 0 Times in 0 Posts


-kwag
Reply With Quote
  #51  
11-21-2003, 10:02 PM
ak47 ak47 is offline
Free Member
 
Join Date: Oct 2002
Posts: 168
Thanks: 0
Thanked 0 Times in 0 Posts
__________________
Later ak
Reply With Quote
  #52  
11-22-2003, 05:39 AM
digitall.doc digitall.doc is offline
Free Member
 
Join Date: Jul 2003
Location: Valencia (España)
Posts: 741
Thanks: 0
Thanked 0 Times in 0 Posts
Kwag:
Quote:

-kwag
ak47:
Quote:
digitall.doc:

...you guys
Reply With Quote
Reply




Thread Tools



 
All times are GMT -5. The time now is 04:17 PM  —  vBulletin © Jelsoft Enterprises Ltd