GradFun2DBmod [v1.5 - Update 2009/12/30]

[Korben]

Nice feature. Why is it not possible to use YUY2? I got an error message:

---------------------------
VirtualDub Error
---------------------------
Avisynth open failure:

This is not an YV12 clip ! Please convert color space to YV12 before using GradFun2DBmod()

(GradFun2DBmod.v1.3.avsi, line 136)
(E:\Capture\Project_107\1tdecimate.avs, line 91)
---------------------------
OK
---------------------------

[Adub]

It has to due with it's supporting plugins. One(or more) of them do not support YUY2. Simple as that.

[Benus_MinO]

LaTo, how I can obtain results like this?:




Here is source:

http://pi.fastbighost.com/~benus/Banding.m2ts

[LaTo]

Originally Posted by Benus_MinO

LaTo, how I can obtain results like this?:




Here is source:

http://pi.fastbighost.com/~benus/Banding.m2ts

Something like this should do the job: GradFun2DBmod(thr=1.6,thrC=1.2,str=0.8,strC=0.4)

[Benus_MinO]

Thank you.

[VincAlastor]

very good filter, thank you La To. but for daily using to slow! - is there a direct compute plan?

[LaTo]

v1.4:
changed "temp" parameter (-1=off)
updated documentation

[LaTo]

Originally Posted by VincAlastor

very good filter, thank you La To. but for daily using to slow! - is there a direct compute plan?

For faster settings try:

GradFun2DBmod(adapt=-1,temp=-1)

It should be a big speed up


PS: For realtime processing (viewing) you can also add mask=false.

[Dreadkillz]

Hi, Lato.

Could you explain what "adapt" does? When I set adapt to high values, I see more noise on brighter areas than darker areas. Setting adapt to low values, I see more noise on darker areas than brighter areas.

[ryrynz]

Lato, loving your LSFmod, and am using it in conjunction with deband in ffdshow in most of my scripts. Would the settings you mentioned above be an adequate replacement for deband? Is there a setting that provides in your opinion better image quality with a similar performance impact? Awesome work man.

[markanini]

LaTo, could you suggest some settings for a deblocking type effect?

[LaTo]

Originally Posted by Dreadkillz

Could you explain what "adapt" does? When I set adapt to high values, I see more noise on brighter areas than darker areas. Setting adapt to low values, I see more noise on darker areas than brighter areas.

Yes it's true, this parameter adapts the strength of the grain depending on the brightness.

The goal is to be as close to reality and to reproduce natural film grain.

I think that adapt=64 is a good value, but it also dependent of the source... Pick the best value for your eyes

Originally Posted by ryrynz

Would the settings you mentioned above be an adequate replacement for deband? Is there a setting that provides in your opinion better image quality with a similar performance impact? Awesome work man.

GradFun2DBmod is particularly designed for encode to remove and/or prevent banding.

However you can use it in realtime, the quality will be better but it is slower... It's a tradeoff, I think that the ffdshow deband is faster and good enough for only viewing.

Originally Posted by markanini

could you suggest some settings for a deblocking type effect?

Never used for deblocking, but set mask=false and increase gradually thr/str parameters until blocks disappear.

[rkalwaitis]

Thanks for the update Lato.

[Dogway]

The filter only works in certain points of the filter chain. I dont know if Im doing something wrong. Here is an example where it doesnt work for me:

ffmpegsource("whatever.mp4")

trim(0, 900)

Tweak(sat=1.2, coring=false)
Levels(4, 0.96, 242, 0, 255, coring=false)

tnlmeans(ax=4,ay=4,az=10,sx=2,sy=2,bx=1,by=1,sse=false)
# GradFun2DBmod(thr=1.6,thrC=1.2,str=0,strC=0,radius=3,range=3,mode=2,temp=-1,mask=true,adapt=64) #here it works
Lanczos4Resize(858, 482)

ttempsmooth(maxr=7)
AddBorders(4,2,4,2,color=$000000)
mfToon(strength=0, sharpen=true, cwarp=false, wdepth=15, wblur=1, wthresh=0.5)
LimitedSharpenFaster(Smode=4, strength=20, edgemode=1, soft=-1)
Crop(6, 2, -6, -4)
GradFun2DBmod(thr=1.6,thrC=1.2,str=0,strC=0,radius=3,range=3,mode=2,temp=-1,mask=true,adapt=64) # here it has no effect

Although if I place it just before the resizer it does work. Its strange... And its not only in a certain video, it happens in others as well.

[Stephen R. Savage]

GradFun2DB, which is the key filter underlying all of GradFun2DBmod, fails silently if the input width is not divisible by 8. After cropping, you have a width of 846, which is not divisible by 8 (nor by 4). What should be done is that the mirroring stage of GradFun2DBmod be changed to pad to a mod8 width.

Alternatively, you could crop/resize to width 840 or 848 instead of 846.

Edit: I missed the AddBorders call. The width in that script is 860, so the point still stands.

[LaTo]

Originally Posted by Stephen R. Savage

GradFun2DB, which is the key filter underlying all of GradFun2DBmod, fails silently if the input width is not divisible by 8. After cropping, you have a width of 846, which is not divisible by 8 (nor by 4). What should be done is that the mirroring stage of GradFun2DBmod be changed to pad to a mod8 width.

Alternatively, you could crop/resize to width 840 or 848 instead of 846.

Arf, I will add a warning in GF2DBmod

[LaTo]

v1.5:
added MOD4 restriction
changed script to accept MOD8 clip
updated documentation

[Stephen R. Savage]

That is unnecessary. Enabling GradFun2DBmod to work on mod2 and mod4 (width) video is as simple as a two line fix. Also, I think GF_Padding is the only edge protection mode necessary. It offers better performance than GF_Mirrors at the corners while still being faster. GF_Borders doesn't actually work (no dithering across sharp black edge), so it doesn't fit either.

Function GF_Padding(clip clp, float thr, float thrC)
   {
   ow = clp.width
   oh = clp.height
   w = int(round(ow/8.0)*8)  + 32
   h = CLP.height() + 32
   
   RDY   = CLP.pointresize(w,h,-16,-16,w,h)

   LUM = RDY.gradfun2db(thr).Crop(16,16,ow,oh)
   CHR = RDY.gradfun2db(thrC).Crop(16,16,ow,oh)

   GFP = thr==1.0 && thrC==1.0 ? CLP
   \   : thr==thrC             ? LUM
   \   : thr!=1.0 && thrC==1.0 ? LUM.mergechroma(CLP)
   \   : thr==1.0 && thrC!=1.0 ? CLP.mergechroma(CHR)
   \   :                         LUM.mergechroma(CHR)

   Return GFP
   }

I think this can be worked into the other modes, but I'll leave that for you to do.

[Dogway]

Thank you Stephen. I spent a lot of time trying to figure out.

Being mod 8 isn't too strict? limiting to 848 pixels width either messes up the AR, or I have to crop 3 pixels each side.

I normaly prefer pixel to pixel sampling with no scaling involved, I wouldn't mind encoding anamorphic unless desktop divx players respect custom AR(?)(ie. 848x480 vs 720x480)

sorry for the offtopic bit.

[LaTo]

Originally Posted by Stephen R. Savage

That is unnecessary. Enabling GradFun2DBmod to work on mod2 and mod4 (width) video is as simple as a two line fix.

Yes, you're right indeed... So, now it accepts mod4 input.

Originally Posted by Stephen R. Savage

Also, I think GF_Padding is the only edge protection mode necessary. It offers better performance than GF_Mirrors at the corners while still being faster. GF_Borders doesn't actually work (no dithering across sharp black edge), so it doesn't fit either.

Yes, it's why GF_Padding (mode=2) is the default mode





/!\ I HAVE UPDATED THE SCRIPT FOR MOD4 INPUT
YOU SHOULD DOWNLOAD AGAIN THE V1.5 /!\