1 Topic by travolter (edited by travolter 06-04-2011 16:23:42)

  • Thanks: 8

Topic: my fav script

I know that you have tested tons of scripts.. but I want to post here my fav one. Maybe you find this movement interesting.

My goal is to get max smoothness in porn movies/sports. I can sacrifice some image artifacts if I get more fluidity in the movement. So this is my best after 2.5 years of tests( and Im sure that  can be improved)

(I convert source framereate to 25 and then apply mvtools to do 75fps. Using 75Hz monitor)

SetMTMode(2,7)
ffdshow_source()
assumefps(25000,1001)
SetMTMode(2)
super=MSuper(pel=2, hpad=22, vpad=22,  rfilter=0, sharp=0)
backward_1=MAnalyse(super,  chroma=false,  isb=true, blksize=8,levels=0, searchparam=3, pelsearch=0, plevel=1, search=4, badrange=(-24), pnew=50,lambda=1000,lsad=1200,global=true)
forward_1=MAnalyse(super, chroma=false, isb=false, blksize=8, levels=0, searchparam=3, pelsearch=0, plevel=1, search=4, badrange=(-24), pnew=50,lambda=1000,lsad=1200,global=true)
backward_2=MRecalculate(super, chroma=false, backward_1, blksize=4, searchparam=1, search=4, pnew=50,lambda=250)
forward_2=MRecalculate(super,chroma=false, forward_1, blksize=4, searchparam=1, search=4, pnew=50,lambda=250)
Setmemorymax(128)
MBlockFps(super, backward_2, forward_2, num=75000, den=1001, mode=0, thscd1=400,thres=100)
SetMTMode(1)
GetMTMode(false) > 0 ? distributor() : last

I was not able to reproduce same smoothness with SVP yet.. because I dont have enough cpu to use all settings that I need. (Im waiting for chroma=false and save cpu cycles, because no GPU ghere to help).

  • MAG79
  • SVP developer
  • Offline

Re: my fav script

travolter
Thanks for interesting information.

I was not able to reproduce same smoothness with SVP yet
The closer SVP settings to your script are:
http://www.svp-team.com/forum/misc.php?item=155

Try this. Your opinion is interesting to us.

And now let's have a look at your script.

1. assumefps(25000,1001)
Is it mistake? This function is changing the video speed. You could not change speed from avs-script in ffdshow. It can lead to loss synchronization hmm

2. blksize=8 and hpad=22, vpad=22
Why you need so big values for hpad and vpad?
I think 8 pixels is good enough for them.
SVP uses hpad=vpad=blksize.

3. chroma=false
SVP uses chroma=true. It is more difficult for calculation but all SAD-settings are working by another way.

4. plevel=1
SVP uses plevel=0. It adds more smoothness than plevel=1.

5. searchparam=3, pelsearch=0
SVP uses searchparam=pelsearch=Radius. I can't say good or bad about this. Just this is implemented so.

6. badrange=(-24)
SVP uses default (24). UMH vectors search instead of Exhaustive.

7. pnew=50,lambda=1000,lsad=1200
SVP uses the same values. These values are default for MVTools plugin.

8. lambda=250
SVP don't change this parameter. It it still 1000 by default.
This is a good idea. cool To search with value of 1000 and to repeat search with value of 250. It has to reduce coherence of vectors.

9. SetMTMode(1)
Why is this line?
Is this a some trick? For what?

As said Chainik you can change the main part of script situated in AVS\MSmoothFps.avs by adding or changing parameters.

Post's attachments

SVP3_settings_4travolter.png, 17.92 kb, 425 x 563
SVP3_settings_4travolter.png 17.92 kb, 677 downloads since 2011-04-06 

  • Thanks: 8

Re: my fav script

MAG79 wrote:

travolter
Thanks for interesting information.

I was not able to reproduce same smoothness with SVP yet
The closer SVP settings to your script are:
http://www.svp-team.com/forum/misc.php?item=155

Try this. Your opinion is interesting to us.

I dont have enough cpu to run these settings sad
In my opinion you are limiting SVP if some values cannt be changed.
Personally I trade quality for smoothess, and I turn off options to save CPU cycles

For example, Im using rfilter=0, sharp=0, search=4, chroma=false, mode=0 for saving CPU cycles, not for image quality

MAG79 wrote:

And now let's have a look at your script.

1. assumefps(25000,1001)
Is it mistake? This function is changing the video speed. You could not change speed from avs-script in ffdshow. It can lead to loss synchronization hmm

I notice that mvtools2 works better when you do x2 x3 x4.. etc instead of using x2.4 (25->60)

All the clips available are 24 25 or 30.. so I decided to set monitor to 75hz and convert all videos to 25fps and then apply mvtools. (25->75 x3)

I know that assumefps is not the best way to convert every video to 25 because some audio sync problems will appear. If you a better way to do .. plz tell me wink

MAG79 wrote:

2. blksize=8 and hpad=22, vpad=22
Why you need so big values for hpad and vpad?
I think 8 pixels is good enough for them.
SVP uses hpad=vpad=blksize.

you true. In fact these values that Im using here are random because I dont find many differences tweaking them

MAG79 wrote:

3. chroma=false
SVP uses chroma=true. It is more difficult for calculation but all SAD-settings are working by another way.

For me chroma=false is a must.. or Ill need to buy a new CPU and I cannt hmm

MAG79 wrote:

4. plevel=1
SVP uses plevel=0. It adds more smoothness than plevel=1.

thanks for the tip .. Ill fix smile

MAG79 wrote:

5. searchparam=3, pelsearch=0
SVP uses searchparam=pelsearch=Radius. I can't say good or bad about this. Just this is implemented so.

6. badrange=(-24)
SVP uses default (24). UMH vectors search instead of Exhaustive.

At moment I dont notice too much difference in smoothness using different values. Ill have to try different ones. maybe I can improve image quality without loosing smoothness



MAG79 wrote:

7. pnew=50,lambda=1000,lsad=1200
SVP uses the same values. These values are default for MVTools plugin.

I need to fix that too. Subjunk pointed me that before smile

MAG79 wrote:

8. lambda=250
SVP don't change this parameter. It it still 1000 by default.
This is a good idea. cool To search with value of 1000 and to repeat search with value of 250. It has to reduce coherence of vectors.

Im using the values that appear in the mvtools documentation: lambda=(blocksizex1000)/64

MAG79 wrote:

9. SetMTMode(1)
Why is this line?
Is this a some trick? For what?

In the AVS forum we are using that value since long time. Dont ask me why because Im not the creator.. but it does not hurt in our scripts wink

4 Reply by gaunt (edited by gaunt 07-04-2011 10:37:52)

  • gaunt
  • Beta Tester
  • Offline

Re: my fav script

travolter wrote:

,lambda=1000,lsad=1200

These are the basic values. To get a better fluency.
Try lsad = 2000-4000(blksize=16), will be better.
Reducing the lambda would be worse if SAD
SATD will be better, but "expensive".
Use lsad = 4000, lambda = 2000, dct = 5, sadx264 = 4
Or lsad = 800, lambda = 8000, if SAD

5 Reply by travolter (edited by travolter 07-04-2011 12:07:51)

  • Thanks: 8

Re: my fav script

hey!!! Thanks a lot gaunt!!!!!
I was testing lsad = 800, lambda = 8000 and the fluency is really nice.

I always loved fluency of my 8x8 script but using these lasd and lamdba combos I can go back to 16x16 or 16x8 blksizes..

Big thanks again for this help pal! wink

  • MAG79
  • SVP developer
  • Offline

Re: my fav script

travolter
Thanks for your answer.

I dont have enough cpu to run these settings
What CPU do you have?
I have core 2 duo (at work) and core i7 920 (at home).
See my System details.

you are limiting SVP if some values cannt be changed
Yes. We are. It is special for END users. And for GOOD quality.

Im using rfilter=0, sharp=0, search=4, chroma=false, mode=0 for saving CPU cycles
Thanks for this info. We uses search=4 too. Another settings you say are still untouched.
May be in near future. wink

mvtools2 works better when you do x2 x3 x4.. etc instead of using x2.4 (25->60)
You are right. It is because of saving all source frames in output. Special for this we have 1m- and 2m- coefficients. Take a look. You will be surprised. wink
Of coarse, true (not 1m or 2m) 24->60 interpolation are losses every second source frame. The result is more noticeable artifacts.

All the clips available are 24 25 or 30.. so I decided to set monitor to 75hz and convert all videos to 25fps and then apply mvtools. (25->75 x3)
It is strange to hear, especially about 30 to 25 slowdown. hmm
I use coefficients 2.4x, 2.5x not only doubling and ReClock for this purposes to save synchronization.

lambda=(blocksizex1000)/64
I need to look at source code. I'm not sure that this setting is not autoscalable by blocks size like thSCD1.

7 Reply by travolter (edited by travolter 08-04-2011 08:40:13)

  • Thanks: 8

Re: my fav script

>>>I dont have enough cpu to run these settings
>>What CPU do you have?

Here 9550 Quad Core.
Problem is that I dont have GPU. Im not a 3D gamer so I have the cheapest videocard.

>>>mvtools2 works better when you do x2 x3 x4.. etc instead of using x2.4 (25->60)
>>You are right. It is because of saving all source frames in output. Special for this we have 1m- and >>2m- coefficients. Take a look. You will be surprised. wink
>>Of coarse, true (not 1m or 2m) 24->60 interpolation are losses every second source frame. The result >>is more noticeable artifacts.


plz could you paste an example script using these 1m 2m coefficients? Im really interested to try this

Re: my fav script

travolter wrote:

could you paste an example script using these 1m 2m coefficients

Set "Framerate change" to "24->60 1m (2.5x)" or "25->60 1m (2.4x)" and see the script ("Show -> AVS-script from last play").