I was perusing the Avisynth forum and read a post about a LimitedSharpen (http://forum.doom9.org/showthread.php?s=&threadid=84196) script for Avisynth that produces superior sharpening. Here is a blurb from that thread from Didée the LimitedSharpen author:

LimitedSharpen() can be used like a traditional sharpener, but producing much less artefacts. It can be used as a replacement for the common "resize(x4)-XSharpen-resize(x1)" combo, with very similar results (perhaps even better) - but at least 2 times faster, since it requires much less oversampling. And by chaining several instances, it can even be used for something like a "[very] poor man's deconvolution" - but only if one knows how to battle the noise

What's the problem with "normal" sharpening?

Traditional sharpeners like sharpen() or UnsharpMask() compare each pixel against the average of its neighborhood, and emphasize the difference between them. The results (something like "per-pixel contrast enhancement") are good as long as the strength is kept low enough. But artefacts will arise very soon.
XSharpen, being a non-linear sharpener, replaces each pixel with either its darkest or brightest neighbor, depending on which is nearer in range. By the nature of the method, XSharpen produces edges with maximal possible aliasing (jaggyness). So one has either to reduce the percentage, thereby weakening the overall effect, or to work with big supersampling, which makes it both extremely slow and less effective.

After seeing some of the results some were getting using it in pre-processing I decided to see if it could be use in post processing via ffdshow for DVD playback. This is a powerful sharpening script but I found if you set its parameters low it could be used on the fly with one draw back at least for me and that is re-sizing. If I try to resize it becomes to slow and stutters so the before and after pics are original size.

I found that using LimitedSharpen with Denoise3d and applying the Denoise prior to LimitedSharpen gives me the best results with between 53% &56% processor usage.

Here are some of my results so far what do you think?:

Empire Strikes BackNormal

http://img92.exs.cx/img92/927/No-sharp.jpg

Empire Strikes BackNormal with LimitedSharpen and Denoise3d

http://img23.exs.cx/img23/147/EMPIRE_STRIKES_BACK--Uber.jpg

Episode 1 Normal

http://img116.exs.cx/img116/7307/Normal.jpg

Episode 1 with LimitedSharpen and Denoise3d

http://img116.exs.cx/img116/8149/Uber-tone-down.jpg

Normal LOTR

http://img23.exs.cx/img23/1697/LOTR-normal.jpg

LOTR with LimitedSharpen and Denoise3d

http://img23.exs.cx/img23/6786/LOTR-uber.jpg

Here is a another before and after :


Pirates of the Caribbean Normal

http://img23.exs.cx/img23/6452/PIRATES_normal.jpg

Pirates of the Caribbean with limitedSharpen + denoise3dl

http://img23.exs.cx/img23/2483/PIRATES_OF_THE_CARRIBEANube.jpg

Open Range Normal

http://img23.exs.cx/img23/2497/OPEN_RANGE-normal.jpg

Open Range with limitedSharpen + denoise3dl

http://img23.exs.cx/img23/2402/OPEN_RANGEuber.jpg

Open Range with limitedSharpen + denoise3dl with modified hue & saturation tweak See 5 posts down for explanation.

http://img13.exs.cx/img13/7280/OPEN_RANGE-color.jpg

Um how does this work? using a avisynth code in ffdshow? I knew you could proccess avs scripts using ffdshow.

I don't know if I like the effect. It seems to wash out the colors a little. None the less it is an interesting way of doing it.

Originally posted by Nightanole
Um how does this work? using a avisynth code in ffdshow? I knew you could proccess avs scripts using ffdshow.

Ok this will take a little work for you;

Assuming you already have ffdshow you will need to download Avisynth 2.55 (http://sourceforge.net/project/showfiles.php?group_id=57023) and install it.

Next you will ned to download the pluginWarpsharp (http://www.avisynth.org/warpenterprises/files/warpsharppackage_25_dll_20031103.zip) and the Masktools 1.5.4 (http://jourdan.madism.org/~manao/MaskTools-v1.5.4.zip) if you have AMD or other non P4 processor or the Masktools for P4 dll (http://jourdan.madism.org/~manao/MaskTools-p4-3.dll) for P4 users and install both the Warpsharp and masktool dll files in avisynth's plugin folder.

Now go to this
thread (http://forum.doom9.org/showthread.php?s=&threadid=84196&perpage=20&pagenumber=1) go to Didée's second post and copy the script starting at "# LimitedSharpen()"
Copy it to notepad then save it to your avisynth plugin folder and call it LimitedSharpen.avs and close notepad.

Open your ffdshow config and click on avisynth, in the text box enter these two lines:

Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpen.avs")

LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=3,strength=40)


Put a check in the avisynth box and you are set!

Oh and enable denoise3d and adjust to taste for best results!

NOTE: If you installed Avisynth some place other than C:\program files you will need to change the first line accordingly.

You can change the ss_x and ss_y parameters up to 2.0 but it will slow it down to a crawl, the smode choices are 1, 2, or 3 and strength can be set from 0 to 100. The way it is now is what I use.


NOTE: This works with the latest non-SSE SSE2 version of ffdshow it will give you a script error if you try it in the SSE-SSE2 versions.

Originally posted by imf4
I don't know if I like the effect. It seems to wash out the colors a little. None the less it is an interesting way of doing it.

Yes, but you only really notice it in still captures watching it live I don't even notice it at all even looking for it, it's pretty subtle.

I have not tried it yet but you could probably adjust hue and saturation to compensate if it bothers you to much, I might experiment with that tomorrow if I have time.

*Hue & Saturation tweak

As imf4 pointed out this uber sharping seems to wash out a bit of the color and the easiest way I could find to fix it was in ffdshow by enabling the "Picture Properties" and adjusting the Hue and Saturation to taste. Once fine tuned you should not need to adjust it again while using the Avisynth LimitedSharpening because it seems to remove the exact same amount of color no matter the source.

I did a quick experiment I bumped the Saturation from 64 to 78 and the hue from O to 2 and got the sharpened Open Range pic (see a few posts up) pretty close to the color in the original Open Range normal pic. It's not perfect, I will have to adjust a little more but once done it should be set for everything.

Update!

No Hue-Saturation tweak needed!

Re-Sizing in 100% doable!


First, I found that the color fading was due to using VMR9 as renderer in Zoomplayer I switched to Overlay as renderer and get zero color fade with a pristine image. So unless you insist on VMR9 NO tweaking of the Hue & Saturation is needed.


Second, Re-sizing is doable!

I have found the best way to re-size while using this sharpening method is to do it from within Avisynth and keep as much of the image processing as you can in one procedure. Because this Limitedsharpen takes some processing power it is better to resize after it has processed the image. You can add either of these lines after the LimitedSharpen call and change the size according to your needs;

BicubicResize(1280,720)
LanczosResize(1280,720)
Lanczos4Resize(1280,720)

All do a nice job with Lanczos4 giving you the best quality but taking the most processing power.

You can also try this resizing method during the sharpening process by including dest_x= and dest_y= in the LimitedSharpen call.

For example here is 1280x720 resize:

LimitedSharpen(ss_x=1.0,ss_y=1.0,dest_x=1280,dest_y=720,Smod e=3,strength=40)

Another thing you can to save some processing power is to crop before the LimitSharpen call by entering this line

crop(0,0,-0,-0,)
or
crop(0,0,-0,-0,true) The true parameter is supposed to make it a little faster I think?


Here is what I am using now with Avisynth in ffdshow for trimming the black borders and resizing to 1280x720 and it looks awesome!


Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpen.avs")

crop(8,8,-8,-8,true)
LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=3,strength=40)
Lanczos4Resize(1280,720)

why it always shows error:import:couldn't open C:\Program Files\AviSynth 2.5\plugins\avsfilter.vdf.avs

Originally posted by chenzhaoyi
why it always shows error:import:couldn't open C:\Program Files\AviSynth 2.5\plugins\avsfilter.vdf.avs

What settings are you using?
What are you trying to import?

I have no such error and no such file in my Avisynth plugin folder.

I get instant death and ie resets. Running 20041012 non sse ffdshow. the screen shows up and then the program crashes. I have the odd LimitedSharpen.avs copied to my avs plugin dir and the dlls and all the other junk that went with them.

there is no function named " DEdgemask"

line 47


line47 =


edge = logic( tmp.DEdgeMask(0,255,0,255,"5 10 5 0 0 0 -5 -10 -5", divisor=2)
\ ,tmp.DEdgeMask(0,255,0,255,"5 0 -5 10 0 -10 5 0 -5", divisor=2)
\ ,"max").levels(0,0.86,128,0,255,false)

tryed to add:

LoadPlugin("C:\Program Files\Avisynth 2.5\plugins\MaskTools.dll")
LoadPlugin("C:\Program Files\Avisynth 2.5\plugins\warpsharp.dll")


to the avs script and it just pitched a fit. either says unable to load or aception.

First , all you need are the MaskTools.dll, warpsharp.dll, and the LimitedSharpen.avs, added to your plugin folder not the other stuff that came with the dll files you downloaded. They also have to be in the first folder not a sub folder.

What processor do you have P4? AMD? The Masktool dll will cause Zoomplayer to crash if you are using the wrong version.

Also, using Avisynth in ffdshow you do not have to load plugins it will load them automatically for you.

Update #2

VMR9 fix

Here is the fix for using VMR9 as your renderer and not having washed out color!

Just download the ColorMatrix (http://www.geocities.com/wilbertdijkhof/ColorMatrix_v15.zip) Avisynth plugin and install just the Colormatrix.dll in the Avisynth plugin folder then add this parameter to your setting in Avisynth in ffdshow "ColorMatrix()".

This is what mine looks like now:

[Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpen.avs")

crop(8,8,-8,-8,true)
LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=3,strength=40)
ColorMatrix()
Lanczos4Resize(1280,720)


Here are a couple of non-resized pics using VRM9:

Normal

http://img15.exs.cx/img15/6638/OPEN_RANGE-fix-norm.jpg


Using above settings except resize with VMR9 renderer

http://img15.exs.cx/img15/7235/OPEN_RANGE-fix-uber.jpg

Im running a dual xeon setup. Im not using the sse versions of ffdshow like you said. Im also runing avisynth 2.5.5. I have both dlls in the root plugins folder. For some reason they dont want to load.

Try this Masktools.dll (http://jourdan.madism.org/~manao/MaskTools-p4-5.dll) I have a P4 and it works for me. Just make sure you delete the old one in your plugin folder.

I only have 7 files in my Avisynth plugin folder they are and should be all you need:

DirectShowSource.dll
LimitedSharpen.avs
TCPDeliver.dll
warpsharp.dll
ffavisynth.dll
MaskTools-p4-5.dll
ColorMatrix.dll

If you have any thing besides those remove them to a backup folder or delete them. You don't want ffdshow to try to load any thing oddball that could make it crash.

That did it. thanks. Now i just have to figure out what its doing. it looks better, but i dont know why. Its not harshing the edges like unsharp mask used to do, but its also improving something else. Like its smoother or something, less edginess. Hell i cant tell anything from freeze frames. Its like explaining how dvi looks better then component, freeze frames look the same but dvi runs alot smoother.

Awesome!

Love to see some pics!

You are alo right, the pics are good but seeing it live it is even a hole lot lbetter!

Remember to try Denoise3d with it but low settings or it will wash out some detail and see what you think.

Mine is set to: L:0 C:0.78 T:3.98 with HQ enabled.

You can also try these settings seems to be a bit more cleaner.

LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=2,strength=60)

SNiff sniff it doesnt work on dvds. Im not fast enough. I have to turn off denoise to get it to full speed. i have dual 2.6ghz, but its not smp compatible. Avi's work great though. Maybe ill try it with another decoder other then sonic.

You can try Tom Barry's UnDot (http://www.avisynth.org/warpenterprises/files/undot_25_dll_20030118.zip) Avisynth plugin. Just instal it in to your plugin folder and add the UnDot() parameter to start of your settings should look like this, make sure to include the period;

UnDot().LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=3,strength=40 )

This is very light so it should not slow you down at all!

Originally posted by Socio
What settings are you using?
What are you trying to import?

I have no such error and no such file in my Avisynth plugin folder.


I had the odd LimitedSharpen.avs copied to my avs plugin dir and the dlls and all the other junk that went with them. now I delete all junk,it works,but the cpu
loaded 95% and i just use blurr and limitedsharpen,no resize! LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=3,strength=40)
why?

Not sure,

Try not using the blur and use the Undot plugin in above post it takes much less procesing power, or just try it with out either and see what is does.

Make sure you are using nothing in ffdshow but avisynth and double check your settings, if ss_x=1.0,ss_y=1.0 are set to anything above 1.0 it will slow you down to a crawl.

What processor are you using?

One other thing to try, set your ffdshow blur before Avisynth in ffdshow. One thing I noticed on my setup was if I did any thing other than re-size after LimitedSharpen it would slow me down big time. Besides that doing any denoise after the sharpen would just degrade the sharpening anyway.

This really is an excellent sharpening technique. :D Very happy with the results I'm getting here. I've been playing with the settings so see how this all works. So far my favorite results have been using this combo:

TT2 w/ NV 4 codecs
NVPP (Film mode)
FFDSHOW (July 9th SSE AndyIEG version):
1) AVISYNTH:
Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpen.avs")
crop(0,0,-0,-0,true)
LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=3,strength=40)
2) Resize: 1280x720 Lanczos4 w/ .6 chroma

For some reason doing the resize afterward in FFDshow is faster for me then in AVISYNTH. I'm am using this with the July 9th *SSE2 optimized* version of FFDshow from Andy. So maybe his optimizations are just that much better for the resize.

The filter does take some serious CPU power though. I would love to let it have at a 1280x720 input but that seems quite out of reach at this point w/ my 3.52 (OC'd) P4. Still messing with everything though. THANKS for the find...

-James

Originally posted by jpooton
I'm am using this with the July 9th *SSE2 optimized* version of FFDshow from Andy. So maybe his optimizations are just that much better for the resize.

The filter does take some serious CPU power though. I would love to let it have at a 1280x720 input but that seems quite out of reach at this point w/ my 3.52 (OC'd) P4. Still messing with everything though. THANKS for the find...

-James

You got the SSE2 version of ffdshow to work? I have tried it and all it would do is give me Avisynth script errors! I will have to give it a shot again.

I am running a P4 3.06 and can do 1280x720 nicely I am also using the latest reclock and it sems to help. I will post some re-size pics in a while I am still fine tuning my settings.

Glad you like it, It does need some processing power but the results live are remarkable.

Originally posted by Socio
You got the SSE2 version of ffdshow to work? I have tried it and all it would do is give me Avisynth script errors! I will have to give it a shot again.

Yep... the July 9th version seems to work fine for me (that is what I already had in installed). BTW the "newer" 1012 SSE version did give me script errors you are talking about on another machine. Personally I get better results with Andy's version anyway, so that's what is on my htpc.

Originally posted by Socio
I am running a P4 3.06 and can do 1280x720 nicely I am also using the latest reclock and it sems to help. I will post some re-size pics in a while I am still fine tuning my settings.

I should have been more clear above. I can (and am) resizing to 1280x720 after the AVISYNTH routines. I'd just love to see limited sharpen work on a 1280x720 input :) Reclock I haven't tried at this point. Things are smooth for me, just a somewhat heavy cpu load. It's there to be used though. :D

Originally posted by Socio
Glad you like it, It does need some processing power but the results live are remarkable.

No doubt, it's a keeper!

Originally posted by Socio
Not sure,

Try not using the blur and use the Undot plugin in above post it takes much less procesing power, or just try it with out either and see what is does.

Make sure you are using nothing in ffdshow but avisynth and double check your settings, if ss_x=1.0,ss_y=1.0 are set to anything above 1.0 it will slow you down to a crawl.

What processor are you using?

One other thing to try, set your ffdshow blur before Avisynth in ffdshow. One thing I noticed on my setup was if I did any thing other than re-size after LimitedSharpen it would slow me down big time. Besides that doing any denoise after the sharpen would just degrade the sharpening anyway.

AMD barton 2500+ oc 3200+,nvidia Ti 4200, I am using nothing in ffdshow but
avisynth,cpu still loaded 75%. ffdshow version 20041012 non sse.dscaler 5 decoder in zplayer,colourspace yv12.

I just get

Evaluate: Unrecognized exception!
(ffdshow_filter_avisynth_scipt. line 1)

What does this mean? I tried everything, followed the instructions carefully, toggled Hyperthreading, made sure I had the P4 dll, etc. but no matter what I do I get this message.

Can somebody help?

Thanks, Martin.

Originally posted by chenzhaoyi
AMD barton 2500+ oc 3200+,nvidia Ti 4200, I am using nothing in ffdshow but
avisynth,cpu still loaded 75%. ffdshow version 20041012 non sse.dscaler 5 decoder in zplayer,colourspace yv12.

Still not sure, looks like you have plenty of power to do it, perhaps OCing is interfering with processing.

I am not completely familiar with all the AMD processors but if yours is has SSE you can try the July 9th version of ffdshow with SSE/SSE2 (ftp://download.hirekmedia.hu/ffdshow-20040709_SSE2.rar) and see if that improves performance and as jpooton pointed out it no Avisynth script errors with this one.

What version of Masktools plugin are you using?

Originally posted by MHoefler
I just get

Evaluate: Unrecognized exception!
(ffdshow_filter_avisynth_scipt. line 1)

What does this mean? I tried everything, followed the instructions carefully, toggled Hyperthreading, made sure I had the P4 dll, etc. but no matter what I do I get this message.

Can somebody help?

Thanks, Martin.

Copy & Paste here what you entered into the avisynth text box in ffdshow, so we can see what your settings look like.

I followed your instructions. That's what I pasted into the text box:


Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpen.avs")

LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=3,strength=40)


I also followed the rest of your instructions carefully. Still I only get this exception.

Regards, Martin.

I just cleared the text box completely and still get this exception! It looks as if it was a general problem of AviSynth! What could I try to solve it?

Regards, Martin.

Originally posted by MHoefler
I just get

Evaluate: Unrecognized exception!
(ffdshow_filter_avisynth_scipt. line 1)

What does this mean? I tried everything, followed the instructions carefully, toggled Hyperthreading, made sure I had the P4 dll, etc. but no matter what I do I get this message.

Can somebody help?

Thanks, Martin.

That is the script error I saw on one machine that was using the 1012 SSE version of FFDshow. Make sure you aren't using an newer SSE version like that. The July 9th SSE version works for me, but for some reason that 1012 SSE version threw the error your seeing.

-James

jpooton: Thanks a lot, you were absolutely right! I just installed the non-sse version and everything works!

Thanks a lot!

BTW: I never saw any speed (or CPU load) difference between the SSE and non SSE versions of ffdshow. Some people report dramatic differences. How can this be? I do have a P4 3.2 Ghz so it should profit from the special builds...

Martin

Originally posted by MHoefler
jpooton: Thanks a lot, you were absolutely right! I just installed the non-sse version and everything works!

Thanks a lot!

BTW: I never saw any speed (or CPU load) difference between the SSE and non SSE versions of ffdshow. Some people report dramatic differences. How can this be? I do have a P4 3.2 Ghz so it should profit from the special builds...

Martin

Look up a couple of posts I posted a link for the July 9th ffdshow with SSE SSE2 that will not give you the script errors. The one thing I noticed with the SSE SSE2 version is re-sizing in ffdshow is much smoother and faster than the non SSE October version. In fact I changed my settings so that Avisynth does its processing but not re-size then I use ffdshows Lancozs resize and get better performance and I don't need to use Reclock.

Originally posted by MHoefler
BTW: I never saw any speed (or CPU load) difference between the SSE and non SSE versions of ffdshow. Some people report dramatic differences. How can this be? I do have a P4 3.2 Ghz so it should profit from the special builds...

Martin [/B]

As Socio points out right above this thread, you should see a different in the resize speed using one of the SSE versions that AndyIEG had modified. I think that is mainly the portion he was concentraiting on. In the resize settings panel you'll see some of his notes for increased speed. For some reason Andy's versions are different from the SSE versions that are currently in the FFDshow build on sourceforge. Really I think his versions are the quickest out there if you have a P4.

I'd suggest trying the July 9th version linked above and then put the AVISYNTH filter first (with whatever settings you settle on) and then follow it by a Lanczos 4 resize in FFDshow to your output resolutions with perhaps a little luma sharpening. So far that's the best picture I've seen. Very sharp with litterally no added ringing or halos. :)

I have a 3.2 also and was able to do this without stuttering. However it's being feed from then new NV4 codecs w/ NVPP which seem to be very effecient. After playing with the settings a while I've pretty well stuck with what I've posted above, and just overclocked the system by 10% to buy a little headroom. All seems well as I watched Adaptation last night without a hiccup and it looked great...

good luck...

-James

Update #3


I have found and tested Loren Merritt's version of a high quality 3-way low-pass Denoise3d filter for Avisynth, it is called HQ3dn (http://students.washington.edu/lorenm/src/avisynth/hqdn3d/) and can be used in conjunction with imited LimitedSharpen.

Just download and put the hqdn3d.dll in your Avisynth plugin folder and add the " hqdn3d(0,0,0,0)" parameter to your settings at the front of LimitedSharpen and use a period in between them, if you try to put if after Limitedsharpen it will slow you down big time.

Should look like this for example;

hqdn3d(0,1,1,4.5).LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=3,s trength=40)

Usage:
hqdn3d(clip, float ls, float cs, float lt, float ct, int restart)

ls: (default=4.0)
luma spatial filter strength.
cs: (default=3.0)
chroma spatial filter strength.
lt: (default=6.0)
luma temporal filter strength.
ct: (default=4.5)
chroma temporal filter strength.
restart: (default=lt+1

Note: This filter works only in the YV12 colorspace so you have to only have the YV12 box checked in Avisynth in ffdshow and uncheck the rest. It does not seem to take any more processing power than ffdshow denoice 3d and lets you keep all the image processing in one area instead of jumping between ffdshow and Avisynth.

It is also very powerful, the default settings are to me at least to much for most DVD viewing but if you are viewing Dvix or Avis with a lot of noise the default should be plenty if not you can crank each setting up to 255.

To just use default settings make it look like this:

hqdn3d().LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=3,strength=4 0)

Enjoy!

Originally posted by MHoefler
I just get

Evaluate: Unrecognized exception!
(ffdshow_filter_avisynth_scipt. line 1)

What does this mean? I tried everything, followed the instructions carefully, toggled Hyperthreading, made sure I had the P4 dll, etc. but no matter what I do I get this message.

Can somebody help?

Thanks, Martin.

where is this hyper threading toggle you speak of?

also got undot to work, but cant get hqdn3d to work. Just says cant load dll.

Originally posted by Nightanole
where is this hyper threading toggle you speak of?

also got undot to work, but cant get hqdn3d to work. Just says cant load dll.

He means turning off hyper threading in your bios.

For hqdn3d, First make sure the hqdn3d.dll is in your root or first plugin folder not a sub folder then make sure YV12 is the only thing with a check in it's box under Avisynth in ffdshow.

Its int the root of the plugins, right next to all the other dlls ive been gathering from you. got just yv12 clicked. Even have the msvcp dlls in there ( that corrected something). Still no go on hqdn3d. I dont get it, its setup just like undot, yet undot works and hqdn3d doesnt.

The only other thing I can think to check is the codecs in ffdshow I have Raw Video set to all YUV.

Maybe copy and paste your settings here so we can see them.

In the meantime;

I have been trying out a couple of other things, like DNR2 (http://www.avisynth.org/warpenterprises/files/dnr2_25_dll_20021225.zip) it is a pretty clean denoiser with relitivly small processing power overhead. Download, install it in to your plugin folder and add this parameter to your settings "DNR2()"

Should look like this:

DNR2().LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=3,strength=40)

You can also try temporalsoften it is built in to Avistnth so you don't have to download anything. This is heavy on processing power and you might not be able to resize with it.

Try this for light noise:

temporalsoften(2,3,3,mode=2,6).LimitedSharpen(ss_x=1.0,ss_y= 1.0,Smode=3,strength=40)

For medium noise

temporalsoften(2,5,5,mode=2,10).LimitedSharpen(ss_x=1.0,ss_y =1.0,Smode=3,strength=40)

Heavy noise

temporalsoften(4,8,8,mode=2,10).LimitedSharpen(ss_x=1.0,ss_y =1.0,Smode=3,strength=40)

Eh i dont think my system is fast enough although i dont know why. Even if i eliminate the other proc, im still at 2.6ghz 166 buss with ram at 2-2-2-2, and a gig of it at that. 20% cpu for normal dvd playback. turnon grad denoise and hq, 50%. Turn on this new sharpen and it bottoms out. I can use either or, but not both. I am doing zero resizing. Did get dnr2 to work, but it takes up to much cpu too.

Eh i like the denoise more then i like the sharpen, so guess im axin the sharpen.

Nightanole,

To bad ffdshow and Avisynth are not SMP optimized or you would have been able to run both denoise and limitedsharpen and then some with your Xeons.

This looks good on my computer desktop, but I think it might be just a little much on my big screen (Sharp 11k on 120" diagonal screen).

But more to the point how are you guys getting it to run smoothly. On either my desktop PC (P4 3.06GHz on Intel 845) or my HTPC (P4 3GHz on Intel 865) I cannot get smooth pans with no frame drops irrespective of what version of ffdshow I try.

Cheers
Eric

Originally posted by fishy
This looks good on my computer desktop, but I think it might be just a little much on my big screen (Sharp 11k on 120" diagonal screen).

But more to the point how are you guys getting it to run smoothly. On either my desktop PC (P4 3.06GHz on Intel 845) or my HTPC (P4 3GHz on Intel 865) I cannot get smooth pans with no frame drops irrespective of what version of ffdshow I try.

Cheers
Eric


Eh your only 400mhz faster then me. I would try to stick to overlay. media 9 overlay takes more cpu to do the same job ( though its nice for multi monitor). The 3.0ghz should have hyper 2.0 ( any p4 has hyper 2.0, only the old xeons had 1.0 that slowed you down if the program wasnt hyper ready).

Originally posted by fishy
This looks good on my computer desktop, but I think it might be just a little much on my big screen (Sharp 11k on 120" diagonal screen).

Cheers
Eric

OOOH! pump it out to your 120 in screen I bet it will be a phenomenal picture!

Use the SSE/SSE2 ffdshow July 9th 2004 version I posted a link to it on this thread, on previous page.

I would use the VMR9 if you are not going to re-size and as Nightanole overlay seems to be less of a resource hog so its better for re-sizing and keeping things smooth.

I am still experimenting with stuff but my 3.06 GHz will run with ffdshow Denoise3d (low settings)-Limited sharpen and re-size to 1280 x720 using overlay and its smooth but I can tell it is maxing out.

Originally posted by fishy
But more to the point how are you guys getting it to run smoothly. On either my desktop PC (P4 3.06GHz on Intel 845) or my HTPC (P4 3GHz on Intel 865) I cannot get smooth pans with no frame drops irrespective of what version of ffdshow I try.

Cheers
Eric [/B]

It also helps to have an effecient decoder, not sure what your using. I'm using TT2 (w/ NV's 4.0 set) and NVPP. It ends up being very fast and outputs YV12 which AVISYNTH and FFDshow want to see.

Limitedsharpens parameters will also make a difference for you. Mode 3 is easier on the cpu then mode 2 if I remember correctly. You should be able to run limitedsharpen by itself in FFDshow and output to overlay just fine. Beyond that, getting a SSE optimized FFDshow like the July 9th one should let you resize in FFDshow to your final output *after* limitedsharpen but that's probably about it for a 3ghz machine I would guess.

I'm running a 3.2Ghz machine at 3.52Ghz and it runs smooth but is at 80% with limited sharpen and resize (L4 1280x720 w/ chroma .6) and Meedio running in the background. (that thing always seems to take like 8-10% for me, not sure why).

good luck, and I'd definitly try it on the big screen...

-James

TT2 (w/ NV's 4.0 set) and NVPP


What is this?

tt2=theater tek 2 ( i dont have).
nv= nvidia

so your using theater tek and nv dvd somehow?

Originally posted by Nightanole
TT2 (w/ NV's 4.0 set) and NVPP

What is this?

tt2=theater tek 2 ( i dont have).
nv= nvidia

so your using theater tek and nv dvd somehow?

yeah.. it's just TheaterTek 2 which uses nVidia's 4.0 decoders and nVidia Video Post Processor (NVPP). For me it ends up being a lot more effecient then the older Sonic decoders or others I've tried, especially when NVPP is used before FFDshow. That ends up changing the output colorspace to YV12 also which helps. There are still a few flaws with the nVidia decoders but once they clear them up it will be excellent. They are the same decoders you would get with the package nVidia is now selling directly.

limitedsharpen does take some power though, that's for sure. It would be very interesting if someone could take the algorithm and make some sort of SSE optimized filter that was part of FFDshow as another sharpening choice. :) Maybe Santa (aka AndyIEG) is listening....

-James

I am still playing with my settings trying new stuff and different parameters but so far I am happiest with ZoomplayerPro + WinDVD6 +ffdshow July 9th version with ffdshows denoise3d L:0 C:1.07 T:3.98 & HQ + LimitedSharpen smode 3 and resizing to 1280X720 via ffdshow Lanczos 4.

Here is a before and after pic’s look at the rocks and pebbles and look at the face you will notice its sharper, clearer, and more detailed in the after pic, looks even better live. Plus these are jpeg’s so the quality you see here versus what you will see on your own system will even be better.

You pump this out to a big screen you will probably have the exact same look of astonishment!:D

Normal with resize

http://img122.exs.cx/img122/8312/2-Towers-2normal.jpg

Denoise3d+limitedsharpen+ resize

http://img122.exs.cx/img122/4101/2Tuber1.jpg

LimitedSharpen+Resize NO denoise3d

http://img57.exs.cx/img57/7763/No-denoise3d.jpg

Originally posted by Socio
Normal with resize
Denoise3d+limitedsharpen+ resize
I can see the difference. How about the same shot without the denoise and just with the limitedsharpen/resize?

Ok I will put one up in a bit. the nice thing about this sharpen method is it does not introduce much noice in fact seems the opposite when used by it's self!

Ok imf4,

As per your request I posted a new pic with out using denoise3d just LimitedSharpen and resize. I put it below the one with both LimitedSharpen+denoise ia couple posts up so it can be compared with the others easier.

If anything the limitedsharpen only pic has a little more detail so you can easily use it with out a denoiser.

I use denoise3d with low settings so I can have a general one setting fits all. I know for instance Starwars Episode1 does have some noise so denoise3d at my low settings takes care of most of it. While LOTR Two Towers which is where those pics came from has very little noise so having denoise3d on at my low settings may do little to enhance the image quality but does not hurt it either. So having it on all the time seems fine and I can just pop in a DVD hit play and not have to mess with settings.

I was surprised as well but both of my PCs were struggling. I wasn't too shocked that my desktop PC dropped frames it has virus checking software and other stuff resident, but the HTPC is much cleaner, faster and seemed to struggle just as much.

Both machines use ZP4.10b1. I have tried both overlay and VMR9 (HTPC graphics card Radeon 9700, desktop 6800GT) with no real difference. I have used both the nVidia decoders with NVVPP and the DScaler 5 0.0.4 alphas feeding FFDSHOW YV12.

Neither was smooth enough to be useable (I run the Sharp at 48Hz BTW and for reasons that are not clear to me the dscaler filters on their own seem somehow less fluid, the nvidia filters are champs here).

The version of avisynth was downloaded specially for the purpose and is the most recent I could find. I ran the .avs file as posted by Didier and the script in ffdshow as posted in this thread. It’s the limitedsharpen that does the damage as I cannot run this filter even without any resizing (in either avisynth or ffdshow) without dropping frames.

Both machines run with hyperthreading turned on, I switched it off on my desktop machine but that didn’t help (CPU utilisation pegged at 100%).

I have no experience with avisynth but I wonder if I have inadvertently hard coded some of the parameters into the .avs file. Could someone that is using this successfully could post the contents of their Limitedsharpen.avs file.

Cheers

Eric

I have just built my computer
AMD64 3500+ processor
1 GIG corsair ram
ATI 9800 Pro 256MB ram video card

I cant believe its dropping frames with this avisynth stuff. I dont know how to set ffdshow up either so it may be my own fault. Picture looks very good on my 61inch DLP but dropping frames and stuttering like crazy while playing back dvd's. Plus i remember i had zoom player installed before and i could use the mouse wheel to change the height and width of the picture and now that doesnt work. If i try to change width or height with the wheel it changes width and height together. How can i fix this?

First make sure your refresh rate is set to 60hertz! Higher than that can cause stutter.

If you are using the latest ATI drivers 4.11 full or the 4.12 beta it is my understanding you can go to display properties-settings-advanced-displays, toggle your display and then in the advanced tab and make it either 720p or 1080i and if it asks for FPS set it 30fps. This is supposed to improve video performance.

I am still running Cat 4.9's and just downloaded the 4.11's, I will install them tomorrow and see for sure what it does but if you already have them installed give it a shot.

You can also try Reclock it will correct lots of minor stutter issues, just set your refresh rate to 60Hz and set Reclock to 30fps.

When i hooked my computer up to the tv thru dvi it found and installed the tv as a plug and play monitor and automatically set it for the tv's native resolution of 720p. Its a Samsung DLP tv. Anyway i just checked and its 1280x720 at 60Hz. I did not see a FPS anything. What is reclock.
The other annoying thing is that i used to be able to change the height and width of the picture using my mouse wheel. In zoom player now when i use the wheel it changes both width and height at the same time and when i switch to height only the wheel has no effect is there a way to fix this?

Oh i really dont know how to set up ffdsshow this is how i have it set up.

I have 3 items checked and they are in the following order
avsynth
blur & NR
resize

Should i have all 3 checked and if so in that order?

ok disregard the zoom question i just figured it out. Man the video looks so good using the avsynth stuff. I wish i could get it to play smoothly without the minor stuttering. When i moved the resize under the avsynth it did move a lot faster now its a minor annoying stutter. Im surprised at how good this looks using Dscaler5, zoomplayer wmv pro, and ffdshow. If you have any good info on how to kill this stutter please elaborate. Thanks.

Originally posted by minime9us
Oh i really dont know how to set up ffdsshow this is how i have it set up.

I havsynth
blur & NR
resize
ave 3 items checked and they are in the following order

Should i have all 3 checked and if so in that order?

Should be in this order:

blur & NR
avsynth
resize

Here is something new you can try and see if this helps performance:

Download the fdecimate (http://neuron2.net/fdecimate/fdecimate101.zip) Avisynth plugin and install to your Avisynth plugin folder and add the "fdecimate()" parameter to your settings. You will need to not use blur & NR this plugin does not like it but seems to improve performance using Limitedsharpen and then resizing with ffdshow.

This is what my test settings look like:

fdecimate()
crop(8,8,-8,-8,true)
LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=3,strength=40)
ColorMatrix()

Give it a shot and see what you think.

less stutter but still i dont want to see it at all. I cant believe that this is happening to a amd64 3500+ processor. I can play doom3 in ultra settings a not a stutter in sight.

My settings in avsync copied and pasted below is something wrong?

Import("H:\Program Files\AviSynth 2.5\plugins\LimitedSharpen.avs")

fdecimate()
crop(8,8,-8,-8,true)
LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=3,strength=40)

minime9us,

Download Reclock (http://ogo.nerim.net/reclockfilter/ReClock.test.zip) , install it, set your refresh rate to 60 and set the fps to 30 in Reclock then chose Reclock as your audio playback device in Zoomplayer.

if you get a little green clock icon in your system tray when you start a movie that means Reclock is working and that should help smooth out the stutter.

Originally posted by Socio
I use denoise3d with low settings so I can have a general one setting fits all. I know for instance Starwars Episode1 does have some noise so denoise3d at my low settings takes care of most of it. While LOTR Two Towers which is where those pics came from has very little noise so having denoise3d on at my low settings may do little to enhance the image quality but does not hurt it either. So having it on all the time seems fine and I can just pop in a DVD hit play and not have to mess with settings. Thanks for posting that. I just wanted to see how much it was doing to the image. This opens up a new way of looking at FFDShow. I always knew AviSynth would work but I never looked into it because of the challenge. Plus there are just so many plugins to choose from. This is a great starting place.

What should be the reason that avisynth is not compatible with ffdshow sse2?

using SSE2 could be the solution to get rid off the stutter.
On my system with or without SSE2 gives a change of 15 % cpu usage
(P4 3.0)

I hope this will evaluate, because i like this very much!

Originally posted by hybride
What should be the reason that avisynth is not compatible with ffdshow sse2?

using SSE2 could be the solution to get rid off the stutter.
On my system with or without SSE2 gives a change of 15 % cpu usage
(P4 3.0)

I hope this will evaluate, because i like this very much!

You can use it with the July 9th SSE version created by AndyIEG. (You'll find a link earlier in this thread). It seems the script errors only come up with the SEE versions from sourceforge. I'm running it w/ the July 9th SSE version just fine (Lacznos4 1280x720 w/ luma .6) on a P4 clocked at 3.5Ghz with HT on.

-James

Update #4

Here are a couple more ways to boost performance

If you have a ATI video card download and install the new 4.11 drivers. Once they are installed go to display properties-settings-advanced-monitor and select 60Hz as your refresh rate and apply and keep settings. Next, while still in advanced settings, click on Displays tab, select your monitor and go to the monitors troubleshoot tab and put a checkbox in 720p, and keep settings. Double check that your refresh rate is still at 60Hz and if you are running Reclock disable it. If you have Zoomplayer just choose a different playback device other than re-clock and restart Zoomplayer and you are set!

If that does not work for you, you can try the 1080i. However 720p worked for me it and really seem to boost performance lot!



Another trick,

If you are still having problems with the LimitedSharpen you can try Xsharpen which is kind of like LimitedSharpen light. If you are already running Limitedsharpen you do not have to download anything extra just substitute the LimitedSharpen parameter with XSharpen ().

My test setting look like this:


Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpen.avs")

crop(8,8,-8,-8,true)
XSharpen()
ColorMatrix()


XSharpen does not do quite as nice of a job as LimitedSharpen but you can really only tell from still captures, live you might not notice a difference at all and it is a lot faster. So give it a try and see what you think.

ok thx.

I use Media Portal to play my dvd's. Media portal itself uses 12% cpu (HT on)

Ill tried the SSe version off AndyIEG, but no luck. I have to play with unsharp mask untill there is faster code availible :-(

OK in my case I think that the speed problems were user error. It seems that this is very sensitive to the ordering of the filters in ffdshow. I was leaving the avisynth filter towards the bottom of the list. Even with no previous resizing this was causing problems. Bump the filter up to second spot (after denoise and before resize) as recommended above seems to reduce CPU load. I also overclocked both PCs by about 10% so the HTPC is now running at 3.3GHz.

Anyhow I was able to watch "Proof of life" (region 1) yesterday with only a couple of glitches. 99.9% of the time I was getting the smooth pans I'm used to. This is with
ZP4.03b (I have menu navigation issues with 4.10b1)
nVidia 4 video
nVVPP
ffdshow20040907-SSE2(gradual denoise @10; Crop(8,8,-8,-8,true), Limitedsharpen @40% strength, ColorMatrix; Lancos4 to 1280x720)
VMR7
nVidia 4 audio (SPDIF via RME digi96/8)

And the good news is that some scenes look excellent on my big screen. In general low contrast facial close-ups really have “pop” to them. The bad news is that a lot of bright outdoor scenes have a jumbled look to them (basically to much edge enhancement messing up “natural” textures). A by product of this is that facial close-ups against a bright background look very artificial – as though the faces were superimposed on a fake background using 1970s technology.

It is clear to me that the extent of sharpening is too high for my installation (DVI to Sharp 11k DLP on 120 diagonal screen, viewing distance approx 12 feet). I have reduced the strength of the limitedsharpen filter down to 20 and then maybe 10 and will have another look this evening, but at the back of my mind is the feeling that I should really just drop limitedsharpen (and denoise) from ffdshow and just resize. Incidentally it makes the posterization problem of the nvidia filters much more visible.

It looks great @40 viewed at native resolution on my desktop (a 22 inch viewsonic monitor run at 1400*1050@85Hz) but the DLP is much less forgiving (of everything, noise, edge haloes). The best image that I have seen on the DLP is from the dscaler5 0.0.4 filters scaled to 1280*480 using either Lancos4 or spline but these are “jerky” at 48Hz (maybe the 0.0.5 alphas will sort it).

fishy,

Try the Xsharpen mode I posted about two posts up and see if that has more tolerable sharpening for you.

Here is one more performance tip for Zoomplayer users that use VMR7 or 9 as their renderer. This could help improve using Limitedsharpen with VMR 7 or 9 on some systems.

Download and install the latest Zoomplayer Beta 4.10 beta 2 get it from here

http://www.inmatrix.com/files/zoomplayer_beta.shtml

Install it then open Zoomplayer go to Player Options/Advanced Options / Filter Control / Settings) and "Enable YUV mixing"

@Socio
My post was badly phrased - I have yet to try LimitedSharpen at lower intensities. From my reading of Didiers original announcement post it does look an ideal sharpening technique, I just need to hit the sweetspot of intensity. The niggle at the back of my mind is that the sweetspot (and I think that it would take many hours of viewing to form a definitive judgement) might well be zero...

Now I have the horsepower I'm keen to try to find the sweetspot with limited sharpen. I've tried the ffdshow sharpen filters and never really found one that I liked long term. I presume that xsharpen in avisynth is a similar algorithm to that in ffdshow?

I have made a huge mistake can someone help me out. I was attempting to overclock my AMD64 3500+ so that the stuttering would stop. When i hit the F10 key to save settings in the BIOS the computer would not load windows. Ive restarted it over and over and its just on but nothing more. Is there a way to fix this. I know what i changed and i can change it back if i can just get back into the BIOS or is there a way to just reset the BIOS? Luckily i have 2 computers or i wouldnt be able to seek help tonight.

Disregard i went to the MSI site and found out what to do and it worked. Does anyone know where i can go to find out how to overclock?

Originally posted by imf4
Thanks for posting that. I just wanted to see how much it was doing to the image. This opens up a new way of looking at FFDShow. I always knew AviSynth would work but I never looked into it because of the challenge. Plus there are just so many plugins to choose from. This is a great starting place.

You are welcome,

You are right there are a lot of plugins available to play with and usally lots of settings and parameters to adjust for each and more being developed all the time. It is a great tool to have at your disposal in ffdshow, figuring out how it all works is the time consuming part.

]Originally posted by imf4
fishy

I presume that xsharpen in avisynth is a similar algorithm to that in ffdshow?

Yes it is Avisynth version of Xsharpen, comparing the two in side by side stills it does appear the Avisynth version lends more detail.

I've been following this thread all week and gave limitedsharpen a try. So far, everything has worked for me and the results have been impressive, except for one issue I have with some DVD menus and navigation. Basically, I am unable to navigate some DVD menus when I have Avisynth enabled in ffdshow. Both my mouse and streamzap remote will not move the cursor or highlights in the menu and I can't select any of the menu items. This is really bugging me because I am now addicted to the quality improvements this brings to my setup.

I am using Theatertek 2.0.2, overlay, auto mode, NVPP enabled, ffdshow enabled (denoise 3d, Avisynth LimitedSharpen, Lancos4 resize). System is a P4 3.2, July 9 ffdshow SSE2 version, forceware 66.93 drivers with the 5900SE card (basically a version of the 5900XT). This is driving a Sony VPL-VW11HT set to 1360*768@56hz

The problem will go away if I disable NVPP, but then I don't get the power of NVPP converting to YV12 so I get stuttering and have to reduce resize and play with ffdshow setting.

The problem mostly occurs on older discs, but newer discs too have their problems. Stepford Wives 2004 I can only navigate the main menu to sub menus, then the hangs occur. Other DVD's non of the menus will respond. And strangely, Shrek 2 I can navigate all of the menus !! What is with that ?

I think the real underlying problem is in Theaterek and / or NVPP and not Avisynth, but I am suprised others here using Theatertek / NVPP have not reported similar issues.

I am working around this by keeping NVPP and ffdshow enabled in Theatertek to watch the main feaure, but to navigate the menus and look at suppplements I uncheck Avisynth in ffdshow.

I would love to know if anyone else is experiencing any issues like this with Theatertek and specifically the latest Nvidia decoders and the post processing (NVPP).

Thanks ... Dave

Originally posted by dgama1237
The problem will go away if I disable NVPP, but then I don't get the power of NVPP converting to YV12 so I get stuttering and have to reduce resize and play with ffdshow setting.

yep it's addicting :) I haven't really navigated menus with it yet as I'm playing ripped features directly through meedio/tt2. (no menus). But I think you are correct in that it's likely an NVPP problem as another thread stated some menu problems with NV4 will be corrected in the next update.

In the meantime, I'd try disabling NVPP and then putting a "null" filter at the top of your FFDshow list. Basically a filter that doesn't do anything really but will do the conversion to YV12 before more hardcore functions take place (like avisynth or resize). People often activate "levels" with no changes to them, or put a sharpen=1 before the routines just to get things into FFDshows preferred YV12 space.

There may be a way to do something similar in avisynth also if that's your first filter. I'm not sure if this will help or not, but it's worth a shot at least...

-James

You can try adding "ConvertToYV12()" to your Avisynth settings, it believe it needs to be first so it would look something like this:


Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpen.avs")

ConvertToYV12()
crop(8,8,-8,-8,true)
LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=3,strength=40)
ColorMatrix()

Thanks James - I did try putting a null filter up in the list (tried both the levels first then tried unsharp mask @ 1) , but I still got stutters so I had to reduce my resize which unfortunately did not look as good. I am hoping perhaps the Theatertek 2.0.4 and NVPP updates will help the situation. If I am able to get any improvements I will report back. But this is great stuff. I am also going to demo Meedio - took me a while to get all the ffdshow and colorspace YV12 stuff in my head then I had to tweak further with the Theatertek 2.0 upgrade, but I am ready for the next level. The adventure is never over ...

Dave

Dave... like you, I'm hoping that TT2.04 w/ the new NV stuff will clear some things up. Socio's AVIsynth version above is worth trying, but for some reason I do think NVPP is sort of magic in keeping CPU use down. It almost gives back CPU cycles :) That may be why it's the cause of a lot of other problems at this point though.

In any event, things like Meedio should be a big enough diversion for you to get through things until TT2.04 is out. Be sure to check out the HDeeTV theme, IMHO it's the best thing going. Especially when you start customizing it to be what you want. You'll also want the PlayMee plugin to make it use TT2 easily. Lots of fun ahead...

Good luck..

-James

@dgama1237
I also have problems navigating menus (for eg the initial AC3-DTS selection screen in Gladiator (R1). In my case the menu is still active but it appears that the frame rate has ground to a halt (one frame every 20-30 secs). If I use the mouse to click the menu it works but obviously you get no visual feedback.

The root menu of gladiator itself is useable but has a frame rate that is much lower than 24fps btw.

When the actual movie starts everything thing is fine but using the nvidia filters and nvvpp fast forward and rewind is a bit of a lottery - sometimes it works sometimes the screen freezes and you have to guess how far you've gone.

Here is another plugin called RemoveGrain (http://home.pages.at/kassandro/RemoveGrain/RemoveGrain.zip) and like UnDot I posted about earlier in this thread it seems to need little processing power so it works well with Limitedsharpen for removing fine grain.

Just download and install only the RemoveGrainS.dll to your Avisynth plugin folder, then add one of these parameters to your settings and it has to be before cropping! The parameters will depend on your processors ability SSE2 is the fastest. Modes and be set from 0-9, I found Mode 4 and ModeU 2 gave me the best results.

RemoveGrain(mode=4,modeU=2)
SSE=RemoveGrain(mode=4,modeU=2)
SSE2=RemoveGrain(mode=4,modeU=2)

My test settings look like this:

Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpen.avs")

SSE2=RemoveGrain(mode=4,modeU=2)
crop(8,8,-8,-8,true)
LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=3,strength=40)
ColorMatrix()



NOTE: This plugin also requires the Microsoft (R) C Runtime Library msvcr70.dll which can be downloaded here (http://www.dll-files.com/dllindex/dll-files.shtml?msvcr70) . You just need to put the msvcr70.dll in your Windows/System32 folder.

i m using a dlp pj with a native res. 800x600.
I have the following options for resizing but i doubt about it which i do need to use:

1. resize lanczos4 to 720x576
2. resize lanczos4 to 1440x1152
3. resize lanczos4 800x600
4. NO resizing

and a few questions:
there is a discussion on our forum in Holland (www.htforum.nl) whether it s usefull for 800x600 pj's to resize or not.
My question is also: IF it makes no sense to upscale is it the an idea to resize to 800x600 (actually it stays the same then) but have the benefit of lanczos4?

ps. sorry for my bad English :p

Well, I tried these settings last night using the recommended FFDshow version (July 9th) with TT2.0.

I thought the results were horrible compared to using plain FFDshow's resize with Lanczos 8.

There was alot of pixelation on my setup (128" 2.35:1 screen, BenQ 8700+ DLP projector, P4 3.04GHz).

I wouldn't recommend this tweak to people with very large screens, as it tends to bring out alot of pixelation around text. The opening of Star Wars (any one) is a good test to show the results.

well i couldnt get very far :( when i open zoomplayer it pops up with an error window: "Import: couldnt open "D:\Programs\AviSynth 2.5\plugins\avsfilter.vdf.avs"
any ideas?

Originally posted by Joe Przybylski
Well, I tried these settings last night using the recommended FFDshow version (July 9th) with TT2.0.

I thought the results were horrible compared to using plain FFDshow's resize with Lanczos 8.

There was alot of pixelation on my setup (128" 2.35:1 screen, BenQ 8700+ DLP projector, P4 3.04GHz).

I wouldn't recommend this tweak to people with very large screens, as it tends to bring out alot of pixelation around text. The opening of Star Wars (any one) is a good test to show the results.

Hmm... I'm not sure what you're seeing, but I'm having good results on a 133" 16:9 setup. Here are FFDshow grabs with LimitedSharpen, method 3, 40. No pixelization that I would consider extreme. Although SW is quite excellent as it is.

http://www.moonshineridge.com/avs/grab00770.jpg
http://www.moonshineridge.com/avs/grab01354.jpg
http://www.moonshineridge.com/avs/grab13407.jpg
http://www.moonshineridge.com/avs/grab17953.jpg

Originally posted by jojomcbobo
well i couldnt get very far :( when i open zoomplayer it pops up with an error window: "Import: couldnt open "D:\Programs\AviSynth 2.5\plugins\avsfilter.vdf.avs"
any ideas?


This is a Zoomplayer error or one your seeing in FFDshow? If your following directions on this thread I would assume you're trying limitsharpen/avisynth within FFDshow. In which case you probably want to start with the obvious question of whether your filter really is at "D:\Programs\AviSynth 2.5\plugins\avsfilter.vdf.avs".

Originally posted by sanderheijnen
i m using a dlp pj with a native res. 800x600.
I have the following options for resizing but i doubt about it which i do need to use:

1. resize lanczos4 to 720x576
2. resize lanczos4 to 1440x1152
3. resize lanczos4 800x600
4. NO resizing

and a few questions:
there is a discussion on our forum in Holland (www.htforum.nl) whether it s usefull for 800x600 pj's to resize or not.
My question is also: IF it makes no sense to upscale is it the an idea to resize to 800x600 (actually it stays the same then) but have the benefit of lanczos4?

ps. sorry for my bad English :p

any ideas yet anyone?

Also the following problem:

i use ffdshow in the following order:
1. raw video yv12
2. avi synth
Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpen.avs")
crop(8,8,-8,-8,true)
LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=3,strength= 40)
ColorMatrix()
3. output collorspace yv12

GREAT PICTURE!

Only when i enable denoise 3d (settings 0.5,0.5,5 HQ) i get stuttering. I use the ffdshow 9july, have a p4 3.2 extreme ed. with 512mbddr2 and ati x600 card. cpu usage is 58% sop that cant be the porblem.
Any ideas?

:D AMAZING :D

Is crop(8,8,-8,-8) cropping 8 pixels from each side of the image? If so, why on earth are you doing that?

720x480 = 345600 pixels
after cropping
704x464 = 326656 pixels
You are losing 6.5% of the image

Its pretty noticeable in those shots of Gollum, a big chunk of his forehead is gone. Based on the big rock to the left of his head, it looks like a lot more than just 8 pixels are missing from the top too.

Originally posted by Li On
:D AMAZING :D

What are your settings Li On? Please copy and paste your script...

I'm still at the beginning stage with only the LimitedSharpen. In my early test this sharpen remove the ffdshow buildin unsharp/swscaler hash look, edge halo and some added noise artifact. With LimitedSharpen the picture looks much more smooth (analog/film feel) while having a even higher detail level.

I just came back from a friend's setup with a Sony G70 CRT projector running in 720p. The LimitedSharpen (same value to me as given in this thread) gave lots of pixelization! Weird!

So I still need to do more test as I output in 856x480 only and don't use Resize anyway...

regards,

Li On

Hi,

I have a weird stutter problem! My config:

Asus P4PE MB, P4 2.4G overclock 3.0G (533M FSB to 666M FSB), 1 DDR333 512M ram, Radeon 9500 128M, WinXP Pro SP1, DirectX9c, Cat 4.7, VGA output at 856x480.

NVDVD 4.0, DScaler5 0.004, ZoomPlayer Pro, FFDShow 20040709_SSE (YV12 in and out), with ONLY the AVISynth LimitedSharpen in FFDShow and nothing else.

On direct DVD-ROM DVD playback, I get heavy stutter with CPU at 70-80%.

If I copy the SAME clip onto harddisk and play the VOB file, I get SMOOTH playback while CPU at 80-90%.

The only different is the DVD Navigator in real DVD playback and the Elecard Demux in VOB playback. And the clip is exactly the same.

How can this be?

regards,

Li On

PS: the LimitedSharpen DVD playback does play smooth in my friend's setup with the SAME software config...

Originally posted by Li On
On direct DVD-ROM DVD playback, I get heavy stutter with CPU at 70-80%.

If I copy the SAME clip onto harddisk and play the VOB file, I get SMOOTH playback while CPU at 80-90%. Strange, I wonder if something can't keep up between the DVD drive and the processor. What happens when you use other settings in FFDShow, like just a denoise and resize? Same stutter from DVD with different settings?

The stutter only happen with FFDShow Avisynth LimitedSharpen in direct DVD-ROM playback. Without the use of LimitedSharpen, of course the playback is perfectly smooth, as usual.

Btw, I also tried rip the whole DVD to HD as a ISO file (instead of a VOB clip) and use DaemonTools to mount it as a virtual DVD. The playback also stutter like on a real DVD-ROM! It just make NO sense!

regards,

Li On

I am new to this forum, but have sucessfully been using a HTPC for a couple of years. My question is: How can you use the sharpening etc. mentioned in this forum when about 50% of my DVD's yield a Macromedia error message when trying to view them through my HTPC?

Have I missed something? or is there a way to play the "Protected" DVD's through my system?

thanks,
Keylkeys

I am new to this forum, but have successfully been using a HTPC for a couple of years. My question is: How can you use the sharpening etc. mentioned in this forum when about 50% of my DVD's yield a Macromedia error message when trying to view them through my HTPC?

Have I missed something? or is there a way to play the "Protected" DVD's through my system?

thanks,
Keylkeys

Unfortunately AviSynth is a no go for me. I get constant 100% CPU usage with it on in ffdshow, looks like a slide show, this with an a64 3500+ CPU, Geforce 6800 under VMR9, unbelievable. I'm guessing you need a freakin super computer to use these filters. I'm resizing to my desktop res of 1776 x 992 (Hitachi RPTV) Denoise 3d C 0.87, Avisynth then Lanczos 4 (1776 x 992) L 1.0.
Hmmm... but I also get about 80% cpu usage with just ffdshow too. Anybody out there with a similar CPU, I'd love to see your cpu usage numbers.

VC

Any idea for my stutter problem?

I don't want to watch any new DVD without this LimitedSharpen! I'm stuck! LOL!

regards,

Li On

Originally posted by Jeraden
Is crop(8,8,-8,-8) cropping 8 pixels from each side of the image? If so, why on earth are you doing that?

720x480 = 345600 pixels
after cropping
704x464 = 326656 pixels
You are losing 6.5% of the image

Its pretty noticeable in those shots of Gollum, a big chunk of his forehead is gone. Based on the big rock to the left of his head, it looks like a lot more than just 8 pixels are missing from the top too.

No, blackbars are part of that widescreen image so I just trimmed some of the black bars on top and bottom the actual image did not change except a little on left and right.

In fact to increase performance using limited sharpen you can remove the black bars completely so that Limitedsharpen only processes the image then add bars back afterwords to keep your AR.

Try theses settings for widescreen:

[Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpen.avs")
crop(8,60,-8,-60,true)
LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=3,strength= 40)
ColorMatrix()
AddBorders(0,60,0,60)


You can adjust the amount of crop of course just make sure the add border numbers match to keep your AR. Although my Addborders numbers are usually a little smaller than the crop like 54 instead of 60 so I get a little more image on screen with out any noticeable image stretch.

Also you might want to use ffdshows avisynth save and load feature for these settings and save this as Wide or something that way you can load it for widescreen movies only because it will mess 4:3 movies AR up.

WOW this looks amazing.

Only problem is, I only get smooth playback with "Only right half" enabled
for AVISynth and no other processing enabled, even with cropping .

Like LiOn I use P4PE 2.4GHz but with no overclocking.
And like LiOn I'll have to say, there's no going back.

I'll might have to throw in a highspeed cpu-fan and overclock it to the max.

Hopefully someone will assemble-optimize this, in the not to distant future,
and throw it into ffdshow as a standard sharpen feature........

Henrik

Hi Henrik,

I overclock the 2.4G to 3G. But this LimitedSharpen is super CPU sensitive as my PC will hard lock after a while (around 30mins). I need to back off a bit to 163M FSB at 2.93G to make my system stable under LimitedSharpen.

Anyway my BIGGEST problem is heavy stutter under LimitedSharpen in normal DVD-ROM DVD playback. On the other hand, I get perfectly smooth playback by playing those DVD directly as VOB file, using the exact SAME filters/codecs combination, even with a higher CPU usage.

This is 1 weird problem... :(

regards,

LI On

It looks like the magic marker is just above the 3GHz with a little more than 650MHz FSB for the P4 845 combo.

I tried it at 2.93G P4 163M FSB and 408M ram tonight and:
VOBs played fine
ISO's D-Tools mounted played with almost no stutter
DVD's in the drive with a lot stutter.

In a few days time I'll try and max the machine performance.
At the moment I use one 512M pc2700 and one 512M pc3200.
Maybe I should remove the pc2700 and max the ram speed as well.

My wife was also amazed from the quality of LimitedSharpen so maybe I can get new hardware in the not too distant future ;-)

Henrik

After many testing, my result are as follow:

- DScaler5 0.004 never give a completely smooth playback
- NVDVD 4.0 WITH VPP filter enable, give perfectly smooth playback

So I'm back to NVDVD 4.0 for good flag film and some bad flag film content!

regards,

Li On

Originally posted by Li On
After many testing, my result are as follow:

- DScaler5 0.004 never give a completely smooth playback
- NVDVD 4.0 WITH VPP filter enable, give perfectly smooth playback

So I'm back to NVDVD 4.0 for good flag film and some bad flag film content! Is this playing from the DVD drive or ripped? Did you figure that issue out?

That is smooth direct DVD-ROM DVD playback, with NVDVD 4.0 and VPP filter.

With DScaler5 decoder, heavy stutter DVD-ROM playback with May and up FFDShow. With April and before FFDShow, I get smooth playback in the first 10 mins or so, then a spilt "jumping" every few seconds. If I stop and play it will play smooth for 10mins and "jumping" again.

With NVDVD 4.0 and VPP filter, smooth DVD-ROM direct playback with all FFDShow versions.

I complain the heavy stutter problem in the beginning because I'd switched to use DScaler5 filter most of the time. Now I'm forced back to NVDVD.

I hope the upcoming DScaler5 0.005 release will solve this stutter issue.

regards,

Li On

Have you tried diabling HT Li On. Other P4P users have reported stutter when HT is turned on in the BIOS with this mobo.

I have NO HT in my old Northwood P4.

regards,

Li On

Btw, last night I watched the second half (the first half is watched on and off during DScaler5 testing!) of the movie "Collateral" with NVDVD + LimitedSharpen. And the playback, all the motion just feel SO SMOOTH! It actually feel like there is some kind of magic inter-frame processing or something. Or maybe it is just the way of this movie. Will check other DVD tonight!

regards,

Li On

I've been using Sonic Cinemaster video decoder for the last 1,5 years.
So last night I tried to install the Nvidia4 decoder, but with no luck.
I get the dreaded "pin connection error".
So I thought I'd try the DScaler5.004 decoder despite the alfa stage.
I also overclocked to 2.95 GHz.
Riped the LOTR Return of the King special extended edition to files on HD.
Fine adjusted the 1271 CRT.
Crossed my fingers, invited the family over and ......

WOW everything runs smoothly and with the greatest PQ I've ever seen.
The picture looks very......well, relaxed if you know what I mean.
And analog like, although very sharp at the same time.

This has truly been a night to remember.

Thank you Socio for bringing this to the AVS-Forum!!!

Henrik

i posted my problem about denoise a few times earlier but no response yet.
Problem is heavy stuttering when i enable denoise 3d (settings are now 0.5,0.5,3,98 HQ). My cpu says 58%
first i use denoise and then limitedsharpen with no risizing.

so its:
tt2.05, reclock 1.5beta6,vmr9 with nvpp and ffdshow 9juli.

specs:
p4 3.2 Ghz extreme ed. with HT enabled, 512mbddr2, ati x600 with 4.11 driver.

58 % sounds like very low for a P4 3.2GHz extreme ed,
given that a P4 northwood at 3GHz only manage to do limitedsharpen and nothing else with 90-100%.
Maybe the cpu reading is somehow wrong?

Henrik

i think it has to be something else what the stutter causes when i enable denoise 3d because when i ONLY enable denoise and for the rest nothing else i still get heavy stutter......:(

in response to Vidcan's earlier question... I too am having big time performance problems... i dont have the same chip but i have a similar speed AMD machine...

i've got a mobile athlon running at 2400mhz equivelant to around a 3.3 athlon ... with a gig of memory, radeon 9800 pro...

if I turn on the avisynth script on its own (just doing limitedsharpen thats it) i get 100% cpu usage and slow stuttery video... I will try some more but am a bit dissapointed.. ok its not a powerhouse, and it really does look like intel CPUs are good for this kind of thing.. ohh well :) maybe next upgrade..

I can manage to do denoise3d, high quality, lanczos resize to 1920x1080 with parameter set to 3 (can't handle 4) with about 80% cpu usage so to b e honest i was expecting being able to at least play with this thing..

the magic SSE version of ffdshow i downloaded and couldn't run.. even trying to configure gave an error.. my CPU has SSE 1 but not SSE 2.. i'm assuyming as sse2 is in the filename this is the only one for it.. also i assume as my ffdshow is empty just serving the avisynth script that it wont be affecting it too much...

anyone running it well on an amd machine around 3gig or more?

It looks ok, can't really test it till its working better :)

Originally posted by sanderheijnen
i posted my problem about denoise a few times earlier but no response yet.
Problem is heavy stuttering when i enable denoise 3d (settings are now 0.5,0.5,3,98 HQ). My cpu says 58%
first i use denoise and then limitedsharpen with no risizing.

so its:
tt2.05, reclock 1.5beta6,vmr9 with nvpp and ffdshow 9juli.

specs:
p4 3.2 Ghz extreme ed. with HT enabled, 512mbddr2, ati x600 with 4.11 driver.

Your CPU is pegged. Remember with HT enabled you are looking at a single processor split into two via software. So if your see task manager hit 55+ % you can be sure you are close to the ceiling with HT on. If you turn off HT for fun you'll see the two graphs change to a single graph and you be touching 100%.

-James

where can i check which sse version i got?
I have a Pentium 4 3.2 extreme edition

I disabled nvpp nowdays and only use ffdshow as postproc. in tt2. Now i CAN denoise 3d without heavy stutter so the problem was nvpp (strange i think?).

Super Performance Tip!


I could not help but notice a lot of people were still having a hard time running the LS+Denoise filters so I thought I would post how I get maximum performance. I actually did this to help speed up filtering and re-encoding of DVD's and found it also helps with on the fly DVD filtering and play back performance as well.


This will seem by most to be pretty drastic but if you are dead set on gaining as much performance from your system as possible to run the LS/Denoise3d with re-size then this might help.


Warning this I would not attempt this unless you know what you are doing and I would document what changes I made so you can easily reverse them if needed!

I went to this site and used their tweak guide:

Videoguys (http://www.videoguys.com/TweaksWINXPVE.html)

Then went here for even more in-depth information about Services and more tweaks:

Black Viper's (http://www.blackviper.com/)

I followed the guides on both sites and tweaked my XP pro by disabling every Service and stopping every process I could, set my desktop to highest performance, turned off messenger, removed all desktop icons, wallpaper. etc... I did every thing possible to squeeze every byte of memory and every bit of processing power I could out of my machine. I even take my cable modem offline and shut down my firewall software and antivirus while watching movies or re-encoding


Note: if you watch a lot of DRM content you will want to leave your networking stuff intact otherwise you will not be able to get licensing.


(Also make sure and defrag your hard drive and run anti-spyware software often this will also help keep your system running at full speed for optimum performance.)

Hope this helps some of you

Thanks Socio.

Although the DScaler is a no go on inserted DVDs thanks to my slimline Matsushita Drive's "features" to block any DVDIdle/AnyDVD help, I'm using Dscaler for ripped and Nvidia for inserted DVDs with the help of your Avisynth guides in here.

DirectShowSource.dll
LimitedSharpen.avs
TCPDeliver.dll
warpsharp.dll
ffavisynth.dll
MaskTools-p4-5.dll
ColorMatrix.dll

I don't have TCPDeliver and ffavisynth .dlls in my plugins, am I missing something? Haven't seen them referenced outside of the above quote. Also, why don't you use the Avisynth resize options anymore in your latest test settings?

I only use UnDot and LimitedSharpen. My Plugin folder has only the following files:

LimitedSharpen.avs
warpsharp.dll
MaskTools-p4-5.dll
UnDot.dll

regards,

Li On

Hi there,

is anyone able to use this filter and to do a Lanczos4 1920*1080 resize (using avisinth of ffdshow resize) ?

@Socio

From Blackviper's Homepage " Service settings are global, meaning changes apply to all users."

So what you are proposing should not be possible.

Cheers
Nima

Nima,

I checked my services settings for all three of my user accounts and you are correct, so I altered my tip, still a good tip though.

You still can use fast user switching to go from your normal desktop to a high performance desktop and gain some extra performance just create a new user and set that users desktop accordingly.

Has anyone else tried this with any Immersive products? I've got a Holo3D I and can notice a difference as well. However, it seems there is a lot of tweaking to be done as there is a lot of luminance added from somewhere. Also, I can now definitely tell I am watching an interlaced source because the processor is pegged out. Are any of these plugins (the script, mask edge, or color space) converting anything into the YV12 colorspace? I need to keep it in the YUY2 colorspace for the Holo3D. I was wondering if there was a YUY2->YV12->YUY2 conversion (have only YUY2 output in FFDShow) going on that was eating up clocks. What kind of system is it going to take to run this script smoothly and reliably? I have a 2.8 P4 non-HT w/ 512 PC2700. As stated by other people with faster P4's, we need more power Scotty! Would one of the premium A64's with dual channel DDR do the trick? Also, I have a Radeon 8500 as I've seen no real reason to upgrade as I stopped playing current PC games. If I am outputting VMR9, will I notice any speed improvement going to a DX9 card as the 8500 only does VMR9 in software?

Oh I forgot to add that I can't see the extra luminance when the Holo3D software is wondowed. It only shows up fullscreen. It's like someone is shining a flashlight on everything. Am I correct in thinking it's luminance problems? Any idea why it disappears in windowed mode?

I asked a page back and no replies yet.. has anyone got anything like decent performance out of this plugin with an AMD (preferrably not FX/64) chip?? I'm still dissapointed with the performance of just the limited sharpen by itself on my system. (2400mhz (overclocked mobile 2600) ~ AMD 3300XP)

MoMeanMugs

You can try this:



ConvertToYV12()

Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpen.avs")

LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=3,strength=40)

ColorMatrix()

ConvertToYUY2()



Make sure you have installed and use the ColorMatrix Avisynth plugin if you are outputting via VMR7 or VMR9 otherwise colors will get messed up.

Also make sure you have both YV12 and YUY2 boxes checked in ffdshow-avisynth.

jimbobuk

If you are using Zoomplayer and you are not doing so already try using overlay as your video render and remove the ColorMatrix portion of the Avisynth script you do not need it with overlay and overlay is less CPU intensive at least for me.

using overlay and literally all i'm doing is limitedSharpen.. nothing else.. i was expecting to hit problems at some point, not on the initial sharpening when i'm able to compete so well with traditional ffdshow stuff :)

I'm happy with lanczos resize to 1920x1080 .. tbh i was happy with it resizing to my projectors native res of 950x540 before.. I could go back to this again.. to be honest its so hard to ABX this stuff i get bored of it.. the update doesn't happen when its paused and you can rarely turn things on and off without having to stop/start to the video (especially if a resize is involved) it just conspires to make it really hard to make a judgement call.. making screen grabs could be the way forward but its so different to seeing it flowing with the video..

I've given up for now, if someone else with an AMD 3000 or similar came on and said they could run it fine then i'd question it.. if not the P4 just seems to rule the roost in all this... strange really... the SSE2 stuff and general floating point power must help it out...

I have an a64 3500+ and cant run this stuff, so I'm not holding my breath. To be honest I've found that my graphic cards' (6800nu) sharpening under the display driver does a very good job and I'm happy with it. I'm running Lanczos 2 resize to 1776 x 992 with .9 Luma sharpening and that's it. If i have grainy video I'll turn on Denoise 3d but the newest DVDs these days are keeping the grain to a minimum and I don't see the need to use it. By the way going this route (graphics card sharpening) saves me about 10% CPU utilisation, I'm a happy camper. Honestly I think there's only so much you can do to a DVD signal, running all this stuff with a Cray supercomputer might get you that extra 1-2% of picture quality but come on....

VC

"I've given up for now, if someone else with an AMD 3000 or similar came on and said they could run it fine then i'd question it.. if not the P4 just seems to rule the roost in all this... strange really... the SSE2 stuff and general floating point power must help it out..."

Maybe AMD users should try the latest non-SSE-SSE2 version of ffdshow 2004-10-12 (http://prdownloads.sourceforge.net/ffdshow/ffdshow-20041012.exe?download) not sure if it will help but sure would not hurt to test it out.

i've ran the latest non-sse and came back to sse for when i do use lanczos resize.. the earlier SSE version that works with this didn't work on my chip, as i said earlier i think its SSE2 only perhaps?

The thing is my ffdshow is empty apart from running the simple 2 line avisynth script that does the limitedSharpen() call .. i dont think ffdshow is doing much, rather its avisynth taking the hit..

Vidcan can you not even manage to just do a limitedSharpen() on its own, nothing else in ffdshow, nothing else in the avisynth script, same as me.. if not then AMD users are truly screwed ;) your chip is faster and the 64 has better memory bandwidth i think..

I'm happy with my lanczos sharpening for now.. never done any other sharpening.. sometimes i see some halo-ing but its mostly subtle and i can put up with it.. may even sharpen some more at times... i've really not been able to look at this new sharpening method as the framerate is poor.. its hard to clarify the difference compared to lanczos... anyone care to take some screenshots so we can all look and see how it rates?!

enjoy it you lucky P4 owners :) at least in a year or two pcs will idle and do this stuff ;)

I have a AMD Athlon64 3000 and tried several ffdshow versions with and without SSE. I am not able to run LimitedSharpen, because framerate drops to 10-12 fps as soon as AviSynth is checked. Using Denoise 3d (with and without HQ) brings the same effect. I use Dscaler 5005 in Zoomplayer, VMR9, a Radeon 9600 with Catalyst 4.11 and I process mainly PAL material.

Originally posted by Socio
MoMeanMugs

You can try this:



Make sure you have installed and use the ColorMatrix Avisynth plugin if you are outputting via VMR7 or VMR9 otherwise colors will get messed up.

Also make sure you have both YV12 and YUY2 boxes checked in ffdshow-avisynth.

My understanding was that the Holo3D hardware only worked with YUY2. It seems there might be a YUY2->YV12->YUY2 conversion going on since the Holo3D input is YUY2 and output in YUY2. It seems there is a conversion to YV12 to use the script and ColorSpace. Am I correct? If so, this is eating CPU time, and I would like to know if there is a way to cut this out. Also, I'm gonna try hacking at the code as it seems inneficiency might be playing a key role here. Possibly eliminating some of the float and putting a set integer in would yield the same results and cut down on CPU time.

Colorspace conversion is an issue and not only at this stage. I think it is a big detriment in PIC Quality. Let me explain.

Remember that for each colorspace conversion, The pixel will have its chroma value filtered during the cubic covolution interpolation.

Now DVDs are encoded in YV12 (4-1-1 12Bpp) Mpeg2 format.
Most decoders decompress and deinterlace the frame and produce YUY2 (4-2-2 16Bpp) on the ouput pin.
Dscaler re Convert YUY2 to YV12 to present the Info to AVISynth.
AVISynth Scale the frame with LimitedSharpen then send the frame to the Video Renderer.
The video Renderer (Overlay/VMRx) convert frame color space to display surface Color Depth (usually RGB32 4-4-4) and apply scaling to user output window.

Thats 3 Colorspace conversion and 2 scaling done when only 1 of each is needed! Thus the pixel is processed and filtered 5 times (6 in case of Overlay with Luminance re Mapping) instead of twice.

Ideally, we would like the video decoder output YV12 to DScaler that convert the frame to AYUV (4-4-4 24Bpp) or RGB32 (4-4-4 32Bpp). Then AVISynth apply the efficient LimitedSharpen on full color depth space to user ouptut window size. Thus the ouput frame is directly 1 to 1 mapped to the Renderer surface. This way we have only 1 Scaling and 1 Color space conversion.

Picture details and color fidelity should improve by huge margin on the expense of higher CPU load due to more processing during 24bpp scaling and higher required bandwidth.

Why we don't have such solution yet?

"Vidcan can you not even manage to just do a limitedSharpen() on its own, nothing else in ffdshow, nothing else in the avisynth script, same as me.. if not then AMD users are truly screwed your chip is faster and the 64 has better memory bandwidth i think.. "

jimbobuk, I don't think I tried with just limited sharpen as I don't think it's worth dumping resize for just sharpening. Remember resizing a dvd is the single most important thing you do in improving quality.

Do I think Limited Sharpen is superior to what's out there now? Yes, but I think using some other sharpening tools like the one in the driver performance window gives me about 85% of its quality with almost no cpu hit at all. Using sharpening even in ffdshow causes a 5-10% cpu hit for me.

I think a64 users have been screwed from day 1. Almost everyone with a P4 with similar setups as mine are getting about 25-30% cpu utilisation while I'm gettting 70% and so are most others with an a64. I don't think this is AMD's fault though as they have a superior CPU to the P4 but the inefficencies of ffdshow code is really getting us. This will hopefully change though as I heaer Andy will eventually work on a native 64 bit version of ffdshow. Then we'll leave these p4s in our dust

VC

jimbobuk, I don't think I tried with just limited sharpen as I don't think it's worth dumping resize for just sharpening. Remember resizing a dvd is the single most important thing you do in improving quality.

Ohh i wasn't even suggesting it for a minute i was just interested if your powerful CPU could handle even the sharpening on its own as a CPU load test, nothing more.. I've given up now as its clear we're not able to play.. never mind :)

I don't think this is AMD's fault though as they have a superior CPU to the P4 but the inefficencies of ffdshow code is really getting us.

The AMD cpus are certainly better for some things, and that they can compete with intel cpus at far lower clock speeds shows they are more efficient per clock cycle.. to be honest i've never really looked that closely.. I believe the p4s have far superior floating point capabilities don't they!? and also possibly memory bandwidth, although dual ddr across both platforms helps a lot on this i guess.. For office apps and some games the AMDs generally lead equivelant speed P4s but i've spotted video encoding discrepencies before... perhaps avisynth is loving the SSE2 (3?) of the p4s and thats why they are comfortably handling all this..

Ohh well.. i was excited reading this thread a few nights back.. for now i'll have to leave it.. and i have to say i think the only difference between this sharpen and other sharpens is that it seems to manage to sharpen the images to a point with a much reduced incident of ringing etc.. correct me if i'm wrong on that, as i've not really played with it enough to judge it on my machine... baring this in mind and my often enthusiasm in my sharpening settings (my resize sharpens are both at max for luma and chroma!) I don't think i'm missing THAT much... i dont see the halos often enough to be bothered by them, and even this new sharpener would probably have some other artefact at the amount of sharpening i sometimes do..

I've only ever done resize sharpening, perhaps some unsharpen or other can help get even more sharpening!? i'm sure the resize sharpening amounts changed once when i updated ffdshow from being very clear on most scenes (along with more haloing) to being much more subtle and to th e point of almost not sharpening enough for my tastes compared to how it was before.. then again perhaps this was just placebo.. when i have a spare hour or two i'll have to try and make my way through the ffdshow settings thread and try to decipher some reasoning out of it for doing MORE sharpening :)

AMD users,

I posted about this earlier in this thread, you still might be able to make use of LimitedSharpen script buy just using the Xsharpen portion as sort of a LimitedSharpen light.

The call in avisynth would look like this instead:



Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpen.avs")

crop(8,8,-8,-8,true)
XSharpen()
ColorMatrix()



This does not do quite as good of a job as full LimitedSharpen but seems to perform like an enhanced version of ffdshow Xsharpen. Give it a shot and se what you think.

Originally posted by Joe Przybylski
Well, I tried these settings last night using the recommended FFDshow version (July 9th) with TT2.0.

I thought the results were horrible compared to using plain FFDshow's resize with Lanczos 8.

There was alot of pixelation on my setup (128" 2.35:1 screen, BenQ 8700+ DLP projector, P4 3.04GHz).

I wouldn't recommend this tweak to people with very large screens, as it tends to bring out alot of pixelation around text. The opening of Star Wars (any one) is a good test to show the results.

Yeah, I tried it tonight and my setup is similar except for projector.

I run a Marquee 9000 CRT with a fixed height setup for 2.35 on a 128" screen also.

Watching "Return of the Jedi" tonight, everytime I saw a red LED (about a zillion in this movie) you could clearly see the problem. I went back to standard ffdshow settings as mentioned in several setup guides and I'm happier.

This has been a great eye-opening thread from which I've learned a lot.

Thank you, and thank you to all the helpful posters here.

Quick question: This thread got me started on using avisynth and I'm now trying something out. My monitor is a HDTV/LCD set to 1280 x 768, it's native resolution.

It seems that by using Avisynth and setting my filters to run between the cropping function and the Add Borders function I save a decent amount of cpu cycles.

Is this a cheat (I'm losing out on image quality) or am I getting something for free here?

I'm running Avisynth after resize so the difference in cpu useage is noticeable.

crop(0,40,-0,-40,true)
RemoveGrain(mode=4,modeU=2)
UnDot()
AddBorders(0,40,0,40)

The above gives me an image that seems exactly equal in size to what I have with Avisynth disabled. This is with a widescreen DVD.

Edit: By "free" I'm referring to the lower cpu load when running RemoveGrain(mode=4,modeU=2)
UnDot()
between crop and AddBorders vs. running them after.

Originally posted by Babel-17
It seems that by using Avisynth and setting my filters to run between the cropping function and the Add Borders function I save a decent amount of cpu cycles.



That is exactly what you are supposed to do; by cropping you make the image to be processed as small as possible thus less CPU usage. Then after the image has been processed you add the borders back to keep your aspect ratio intact. Otherwise you waste a lot of time and processing power needlessly filtering the black bar area of the image.

After using limitedsharpen for some time now
and shifting back and forth between unsharp mask and avisynth
I find that with LimitedSharpen the picture is much more smooth and "calm"
and the colors seems more vivid or clean and alive
and of cause the halos are almost gone (maybe only 15% left)
It is very impressive.

Henrik

Thanks for the confirmation Socio.

I really gained a lot of valuable and useable insight from your posts.

Anyone who has had trouble running LimitedSharpen give this a shot it really seems to boost performance!

This works for widescreen DVD's;

Crop using ffdshows Crop, 56 off top and bottom, then use either denoise3d or try gradual denoise @40

Use the following in ffdshow avisynth:




Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpen.avs")

LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=3,strength=40)

Addborders(0,56,0,56)



Use ffdshow's resize, I use Lanczos4 @ 1280x720

The set overlay as your renderer in DVD settings in Zoomplayer, not sure how in TT2.

This makes the image as small as possible before denoise where cropping in avisynth still had the denoising procedure processing everything including black bars. This seems to improve performance a lot my, CPU usage is now only in the upper 40's with a P4 3.06. It runs smooth, no stutters and no need for reclock, I don't know why I did not think of this sooner.

Just make sure you do it in this order:

ffdshow crop
denoise
avisynth limitedsharpen
resize

Hope this helps

Great.
I can now do LimitedSharpen with nothing else on 2.4 GHz P4.

It seems that the "Addborders(0,56,0,56)" have to be removed though.
Ffdshow automatically adds the borders before sending to Rendering.

The world is suddenly a wonderful place :-))))

Henrik

Ah, thanks Sinobi, I was going just a tad mad messing with my settings just now! :)

Using Avisynth just to AddBorders along with ffdshow's crop feature and with Blur & NR set in between wasn't yielding me anything. In fact, it seemed like adding the borders was giving me a negative effect on my cpu cycles.

Fwiw, I'm using Andy's latest (Aug 2 2004) SSE2 optomized build and PowerDVD 6 as the decoder. Overlay mixer is checked in ZoomPlayer. I can't be sure but I'm wondering if Andy hasn't built in a sort of cropping feature into the way dnoise3d works.

Even using just the crop feature with no Avisynth AddBorders doesn't seem to give me a positive impact.

I've been doing a lot of testing lately and I've been trying to make sure I didn't hose my settings.


I can't swear to my thoroughness in testing though.

Originally posted by Sinobi
Great.
I can now do LimitedSharpen with nothing else on 2.4 GHz P4.

It seems that the "Addborders(0,56,0,56)" have to be removed though.
Ffdshow automatically adds the borders before sending to Rendering.

The world is suddenly a wonderful place :-))))

Henrik

You are right it seems the "Addborders(0,56,0,56)" only needs to be added when resizing at least for my setup if I don't the AR is off.

Update!

Introducing LimitedSharpen2

I have tested it out and works great, it has enhanced sharpening that produces virtually the same affect as super sampling but in real time.

Here are some before & after pic's:

Normal

http://img147.exs.cx/img147/3656/nosharpen6xy.jpg

LimitedSharpen

http://img147.exs.cx/img147/6754/limitedsharpen0co.jpg


LimitedSharpen2

http://img147.exs.cx/img147/5638/linitedsharp25bi.jpg

LimitedSharpen with aSharp

http://img28.exs.cx/img28/8233/lswithasharp6vl.jpg

Here is how to setup LimitedSharpen2;

Do every thing the same as with LimitedSharpen then download the SSXS-a-LIKE" zip (http://service.gmx.net/mc/LE8ieA9UDaLNXyzL10Kj3FUonin6zp) and install both asharp.dll and the aWarpSharp.dll to your Avisynth plugins folder.

Note: you have to hit the "GMX Mediacenter Starten" button on that page to get to download.

Use these settings in ffdshow-avisynth:



Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpen2.avs")

LimitedSharpen2(ss_x=1,ss_y=1,Smode=4,aSharpS=0.5)

ColorMatrix()



Then copy the following script and save it to your Avisynth plugin folder as LimitedSharpen2.avs




# LimitedSharpen2()
#
# A multi-purpose sharpener by Didée, hacked by Akirasuto n' Soulhunter... ^^
#

function LimitedSharpen2( clip clp,
\ float "ss_x", float "ss_y",
\ int "dest_x", int "dest_y",
\ int "Smode" , int "strength", int "radius",
\ int "Lmode", bool "wide", int "overshoot",
\ bool "soft", int "edgemode", bool "special",
\ float "aSharpS", int "exborder" )
{
ox = clp.width
oy = clp.height
ss_x = default( ss_x, 1 )
ss_y = default( ss_y, 1 )
dest_x = default( dest_x, ox )
dest_y = default( dest_y, oy )
Smode = default( Smode, 4 )
strength = Smode==1
\ ? default( strength, 160 )
\ : default( strength, 100 )
strength = Smode==2&&strength>100 ? 100 : strength
radius = default( radius, 2 )
Lmode = default( Lmode, 1 )
aSharpS = default(aSharpS, 1)
wide = default( wide, false )
overshoot= default( overshoot, 1)
overshoot= overshoot<0 ? 0 : overshoot
soft = default( soft, false )
edgemode = default( edgemode, 0 )
special = default( special, false )
exborder = default( exborder, 0)
xxs=round(ox*ss_x/8)*8
yys=round(oy*ss_y/8)*8
smx=exborder==0?dest_x:round(dest_x/Exborder/4)*4
smy=exborder==0?dest_y:round(dest_y/Exborder/4)*4

clp.isYV12() ? clp : clp.converttoyv12()

ss_x != 1.0 || ss_y != 1.0 ? last.lanczosresize(xxs,yys) : last
tmp = last

edge = logic( tmp.DEdgeMask(0,255,0,255,"5 10 5 0 0 0 -5 -10 -5", divisor=2)
\ ,tmp.DEdgeMask(0,255,0,255,"5 0 -5 10 0 -10 5 0 -5", divisor=2)
\ ,"max").levels(0,0.86,128,0,255,false)

bright_limit = (soft == true) ? tmp.blur(1.0) : tmp
dark_limit = bright_limit.inpand()
bright_limit = bright_limit.expand()
dark_limit = (wide==false) ? dark_limit : dark_limit .inflate.deflate.inpand()
bright_limit = (wide==false) ? bright_limit : bright_limit.deflate.inflate.expand()
minmaxavg = special==false
\ ? yv12lutxy(bright_limit,dark_limit,yexpr="x y + 2 /")
\ : maskedmerge(dark_limit,bright_limit,tmp,Y=3,U=-128,V=-128)

Str=string(float(strength)/100.0)
normsharp = Smode==1 ? unsharpmask(strength,radius,0)
\ : Smode==2 ? sharpen(float(strength)/100.0)
\ : Smode==3 ? yv12lutxy(tmp,minmaxavg,yexpr="x x y - "+Str+" * +")
\ : asharp(aSharpS,0,0).awarpsharp(3,1,bm=3,cm=0,thresh=0.99)

OS = string(overshoot)
Lmode == 1 ? yv12lutxy( bright_limit, normsharp, yexpr="y x "+OS+" + < y x "+OS+" + ?")
\ : yv12lutxy( bright_limit, normsharp, yexpr="y x "+OS+" + < y x y x - "+OS+" - 1 2 / ^ + "+OS+" + ?")
Lmode == 1 ? yv12lutxy( dark_limit, last, yexpr="y x "+OS+" - > y x "+OS+" - ?")
\ : yv12lutxy( dark_limit, last, yexpr="y x "+OS+" - > y x x y - "+OS+" - 1 2 / ^ - "+OS+" - ?")

edgemode==0 ? NOP
\ : edgemode==1
\ ? MaskedMerge(tmp,last,edge.inflate.inflate.blur(1.0),Y=3,U=1, V=1)
\ : MaskedMerge(last,tmp,edge.inflate.inflate.blur(1.0),Y=3,U=1, V=1)

(ss_x != 1.0 || ss_y != 1.0)
\ || (dest_x != ox || dest_y != oy) ? lanczosresize(dest_x,dest_y) : last

ex=blankclip(last,width=smx,height=smy,color=$FFFFFF).addbor ders(2,2,2,2).coloryuv(levels="TV->PC")
\.blur(1.3).inpand().blur(1.3).bicubicresize(dest_x,dest_y,1 .0,.0)
tmp=clp.lanczosresize(dest_x,dest_y)

clp.isYV12() ? ( exborder==0 ? tmp.mergeluma(last)
\ : maskedmerge(tmp,last,ex,Y=3,U=1,V=1) )
\ : ( exborder==0 ? tmp.mergeluma(last.converttoyuy2())
\ : tmp.mergeluma( maskedmerge(tmp.converttoyv12(),last,ex,Y=3,U=1,V=1)
\ .converttoyuy2()) )

return last
}
#

LimitedSharpen worked GREAT on my last projector Panny AE200 with the low 480p native output. But now I changed to a 720p Epson TW200H projector and LimitedSharpen gives TONS of "jaggies", "stair-stepping" and "anti-aliasing" artifacts around objects edge and diagonal lines. I can't use it anymore.

regards,

Li On

Sorry to hear that Li On not sure what could be causing those anomalies, sounds like the 720p projector does some internal scaling and does not like the output when using LimitedSharpen. Did you try it with re-clock to see if is just a timing issue between your HTPC and the Projector when using LimitedSharpen?

any news on whether this is cheaper or more expensive than limitedSharpen() that's still well out of reach for me.. i guess this will be more expensive and again out of reach :)

jimbobuk

Have you tried LimitedSharpen using ffdshow crop like what I posted about a few posts up? If not give it a shot and see!

no but i did try half image the other day.. it worked ok at about 50-60% usage ... thats the problem.. you guys are all doing a resize and sharpen (or not) on top... and i really dont want to lose the resize... so the cropping may just let me get by on realtime on its own.. but that just wouldn't do..

I did notice that as it was on its own the aspect ratio seemed wrong.. the picture was zoomed in slightly... which i found a bit bizarre..

Warning!

I have been testing LimitedSharpen2 and found it gives me some screwed up edge anomalies which appears to be caused by the aWarpsharp.dll.

However I also found while trying to figure what was causing the edge problems that the old LimitedSharpen can be improved upon using aSharp. So if you have already downloaded the zip with aSharp in it, have installed it to your Avisynth plugin folder and are still setup to run LimitedSharpen then try these setting in ffdshow-avisynth:



aSharp(0.4,0,0
LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=3,strength=40



This give me the cleanest image yet and no increase in CPU usage! See pics a few posts above to see results.

Disregard above warning!

Apparently some are not having any such problems with LimitedSharpen2 so give it a shot and see for your selves.

By the way for you AMD users and anyone else struggling with LimitedSharpen 1 or 2 if you downloded the zip I posted with aSharp and aWarpSharp and have them installed you can try these settings instead of limitedSharpen, I have seen some nice pic's when it is in use, with my aWarpsharp problem I can not adaquiatly test it my self so see what you think.


aSharp(0.4,0,0).aWarpsharp(3,1,bm=3,cm=0,thresh=0.99)


It is supposed to be much faster than LimitedSharpen as well!

A new hacked version of LimitedSharpen2 is available which was made by someone named SoulHunter. This version has a fix for the edge anomalies at least I was experiencing.

This is the new call script which has a new parameter called Thresh that can be adjust from 0.1 to 2 this controls edge sharpening and setting to high can cause edge problems at least for me anyway.

Call script:


LimitedSharpen2(ss_x=1,ss_y=1,Smode=4,aSharpS=0.4,aWThresh=0 .5)




New version of LimitedSharpen2





# LimitedSharpen2()
#
# A multi-purpose sharpener by Didée, hacked by Akirasuto n' Soulhunter... ^^
#

function LimitedSharpen2( clip clp,
\ float "ss_x", float "ss_y",
\ int "dest_x", int "dest_y",
\ int "Smode" , int "strength", int "radius",
\ int "Lmode", bool "wide", int "overshoot",
\ bool "soft", int "edgemode", bool "special",
\ float "aSharpS", float "aWThresh", int "exborder" )
{
ox = clp.width
oy = clp.height
ss_x = default( ss_x, 1 )
ss_y = default( ss_y, 1 )
dest_x = default( dest_x, ox )
dest_y = default( dest_y, oy )
Smode = default( Smode, 4 )
strength = Smode==1
\ ? default( strength, 160 )
\ : default( strength, 100 )
strength = Smode==2&&strength>100 ? 100 : strength
radius = default( radius, 2 )
Lmode = default( Lmode, 1 )
aSharpS = default(aSharpS, 0.5)
aWThresh = default(aWThresh,0.75)
wide = default( wide, false )
overshoot= default( overshoot, 1)
overshoot= overshoot<0 ? 0 : overshoot
soft = default( soft, false )
edgemode = default( edgemode, 0 )
special = default( special, false )
exborder = default( exborder, 0)
xxs=round(ox*ss_x/8)*8
yys=round(oy*ss_y/8)*8
smx=exborder==0?dest_x:round(dest_x/Exborder/4)*4
smy=exborder==0?dest_y:round(dest_y/Exborder/4)*4

clp.isYV12() ? clp : clp.converttoyv12()

ss_x != 1.0 || ss_y != 1.0 ? last.lanczosresize(xxs,yys) : last
tmp = last

edge = logic( tmp.DEdgeMask(0,255,0,255,"5 10 5 0 0 0 -5 -10 -5", divisor=2)
\ ,tmp.DEdgeMask(0,255,0,255,"5 0 -5 10 0 -10 5 0 -5", divisor=2)
\ ,"max").levels(0,0.86,128,0,255,false)

bright_limit = (soft == true) ? tmp.blur(1.0) : tmp
dark_limit = bright_limit.inpand()
bright_limit = bright_limit.expand()
dark_limit = (wide==false) ? dark_limit : dark_limit .inflate.deflate.inpand()
bright_limit = (wide==false) ? bright_limit : bright_limit.deflate.inflate.expand()
minmaxavg = special==false
\ ? yv12lutxy(bright_limit,dark_limit,yexpr="x y + 2 /")
\ : maskedmerge(dark_limit,bright_limit,tmp,Y=3,U=-128,V=-128)

Str=string(float(strength)/100.0)
normsharp = Smode==1 ? unsharpmask(strength,radius,0)
\ : Smode==2 ? sharpen(float(strength)/100.0)
\ : Smode==3 ? yv12lutxy(tmp,minmaxavg,yexpr="x x y - "+Str+" * +")
\ : asharp(aSharpS,0,0).awarpsharp(3,1,bm=3,cm=0,aWThresh)

OS = string(overshoot)
Lmode == 1 ? yv12lutxy( bright_limit, normsharp, yexpr="y x "+OS+" + < y x "+OS+" + ?")
\ : yv12lutxy( bright_limit, normsharp, yexpr="y x "+OS+" + < y x y x - "+OS+" - 1 2 / ^ + "+OS+" + ?")
Lmode == 1 ? yv12lutxy( dark_limit, last, yexpr="y x "+OS+" - > y x "+OS+" - ?")
\ : yv12lutxy( dark_limit, last, yexpr="y x "+OS+" - > y x x y - "+OS+" - 1 2 / ^ - "+OS+" - ?")

edgemode==0 ? NOP
\ : edgemode==1
\ ? MaskedMerge(tmp,last,edge.inflate.inflate.blur(1.0),Y=3,U=1, V=1)
\ : MaskedMerge(last,tmp,edge.inflate.inflate.blur(1.0),Y=3,U=1, V=1)

(ss_x != 1.0 || ss_y != 1.0)
\ || (dest_x != ox || dest_y != oy) ? lanczosresize(dest_x,dest_y) : last

ex=blankclip(last,width=smx,height=smy,color=$FFFFFF).addbor ders(2,2,2,2).coloryuv(levels="TV->PC")
\.blur(1.3).inpand().blur(1.3).bicubicresize(dest_x,dest_y,1 .0,.0)
tmp=clp.lanczosresize(dest_x,dest_y)

clp.isYV12() ? ( exborder==0 ? tmp.mergeluma(last)
\ : maskedmerge(tmp,last,ex,Y=3,U=1,V=1) )
\ : ( exborder==0 ? tmp.mergeluma(last.converttoyuy2())
\ : tmp.mergeluma( maskedmerge(tmp.converttoyv12(),last,ex,Y=3,U=1,V=1)
\ .converttoyuy2()) )

return last
}
#

Reduce aWThresh to around 0.2 seems remove most of edge "jaggies" but then the LimitedSharpen powerful effect seems goes away too...

regards,

Li On

Originally posted by Li On
Reduce aWThresh to around 0.2 seems remove most of edge "jaggies" but then the LimitedSharpen powerful effect seems goes away too...

regards,

Li On

I don't know what causes it but it seems to react different on different systems, I am not sure if it is video card, video card driver, CPU related or something else. I know I can set it to 0.5 and it gives me the best on the fly image yet, but as soon as I go to 0.6 edges start to show the "jaggies" while others can do 0.99 with out any problems.

the color matrix filter link seems dead, any mirrors? thanks!

Colormatrix (http://jourdan.madism.org/~manao/ColorMatrix.zip)

Here is another toy to try out it is called ModerateSharpen it is very light on resources so most should be able to run it;

You will need to download the

RemoveGrain package (http://home.pages.at/kassandro/RemoveGrain/RemoveGrain.zip) , inside will be three RemoveGrain.dll's and three Repair.dll's, you only need to install one of each to your Avisynth plugin directory. The the RemoveGrainS and RepairS seem to work best for me.

NOTE: You will need the runtime dlls posted on page one of this thread as well. Also should only be used with overlay as renderer because at least on my system even with " colormatrix " the color is still off using VMR7-9.

The following is the ModerateSharpen script just copy and save as ModerateSharpen.avs to your Avisynth plugin directory:



# a function by kassandro.
# (just made "repmode" a parameter / Didée)

function ModerateSharpen(clip input, float amount, int "repmode")
{
repmode = 1
sharpened = Sharpen(input, amount)
return Repair(sharpened, input, mode=repmode)
}




This is the call script for ffdshow-asvisynth, the maximum you can set it to is 1.



Import("C:\Program Files\AviSynth 2.5\plugins\ModerateSharpen.avs")


ModerateSharpen(0.2)

Nice, it looks good and doesn't seem to introduce any artifacting. The cpu hit from it is low. Trying it out with some other filters.

hello ;)
I have a problem in ZP. I have an error of scrypt if you know the solution ;)
I have the image with this scrypt ecris in green

Scrypt error: there i no function named "yv12lutxy"
(c:\programme file\aviSynth 2.5\plugins\LimitedSharpen2.avs. line 57)
(ffdshow_filter_avisynth_script. line 7)

my call:
-----------------------------------

Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpen2.avs")
LoadPlugin("C:\Program Files\AviSynth 2.5\plugins\MaskTools.dll")
LoadPlugin("C:\Program Files\AviSynth 2.5\plugins\aWarpSharp.dll")
LoadPlugin("C:\Program Files\AviSynth 2.5\plugins\asharp.dll")

LimitedSharpen2(ss_x=1,ss_y=1,Smode=4,aSharpS=0.4,aWThresh=0 .99)

-----------------------------------

my soft:

ZP 4.03 PRO
FFDShow 20040725
DSCaler5

it goes well with the call according to :
---------------------------------------------

Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpen2.avs")
LoadPlugin("C:\Program Files\AviSynth 2.5\plugins\MaskTools.dll")
LoadPlugin("C:\Program Files\AviSynth 2.5\plugins\aWarpSharp.dll")
LoadPlugin("C:\Program Files\AviSynth 2.5\plugins\asharp.dll")

aSharp(0.9,0,0).aWarpsharp(3,1,bm=3,cm=0,thresh=0.99)

---------------------------------------------

thank ;)
Aldo;) ;)

Hi there,

is anyone able to use this filter and to do a Lanczos4 1920*1080 resize (using avisinth of ffdshow resize) without heavy stuttering ?

Originally posted by BangoO
Hi there,

is anyone able to use this filter and to do a Lanczos4 1920*1080 resize (using avisinth of ffdshow resize) without heavy stuttering ?

No. My P4 3.4ghz system will not even do the LimitedSharpen without unacceptable (to me) jerkiness. DXVA is so smooth in comparison it is not even funny.

Aldo40


LimitedSharpen requires a plugin called "WarpSharp" you will find the link to it on page one of this thread.

It should look like this:

Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpen2.avs")
LoadPlugin("C:\Program Files\AviSynth 2.5\plugins\MaskTools.dll")
LoadPlugin("C:\Program Files\AviSynth 2.5\plugins\WarpSharp.dll")
LoadPlugin("C:\Program Files\AviSynth 2.5\plugins\aWarpSharp.dll")
LoadPlugin("C:\Program Files\AviSynth 2.5\plugins\asharp.dll")

LimitedSharpen2(ss_x=1,ss_y=1,Smode=4,aSharpS=0.4,aWThresh=0 .99)

BangoO

Try the "MpoderateSharpen" script I posted a few posts up with resize it should offer better performance.

thank you socio;)
I installed the warpsharp.dll version 2.5, I also put call Ci above but I have the same error except has the end it is more line 7 but the line 8

Scrypt error: there i no function named "yv12lutxy"
(c:\programme file\aviSynth 2.5\plugins\LimitedSharpen2.avs. line 57)
(ffdshow_filter_avisynth_script. line 8)

Aldo

Aldo,

Sounds like the MaskedTools plugin is not loading right go here (http://manao4.free.fr/) and download the latest version 1.56 and try that.

Also make sure you only have YV12 checked in ffdshow-avisynth and if you are using this script in ffdshow you do not have to use loadplugin by the way, because as long as they are 2.5 version plugins they will autoload so you just need:


Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpen2.avs")

LimitedSharpen2(ss_x=1,ss_y=1,Smode=4,aSharpS=0.4,aWThresh=0 .99)

ok it I walk is not more error of scrypt ;).
on the other hand I row :( 100% of resource. one is needed crowned PC for arrived has to make functioned all that I have a AMD 2600 and 512 DDR PC3200 you have what like materiel to make functioned this with fluidity? thank you ;)
Aldo

Originally posted by Socio
Second, Re-sizing is doable!

I have found the best way to re-size while using this sharpening method is to do it from within Avisynth and keep as much of the image processing as you can in one procedure. Because this Limitedsharpen takes some processing power it is better to resize after it has processed the image. You can add either of these lines after the LimitedSharpen call and change the size according to your needs;

BicubicResize(1280,720)
LanczosResize(1280,720)
Lanczos4Resize(1280,720)

All do a nice job with Lanczos4 giving you the best quality but taking the most processing power.

You can also try this resizing method during the sharpening process by including dest_x= and dest_y= in the LimitedSharpen call.

For example here is 1280x720 resize:

LimitedSharpen(ss_x=1.0,ss_y=1.0,dest_x=1280,dest_y=720,Smod e=3,strength=40)



Does this work exactly the same as the resize algorithms in ffdshow? Ie., if I am resizing with Lanscoz 2 to 1920x1080, with Luma sharpening, can I disable this completely, switch to YV12 inside avisynth, and resize and sharpen inside the LimitedSharpen filter?

Also, I'm assuming that it would be mroe processor-intensive to have the sharpen come after the resize, even within LimitedSharpen, since you're working on a much bigger size - plus you'd need to up the values to get the same perceived results.

I can't wait to try this all out. Thanks to Socio for the great guide and all the information. Now all I have to do is wait for DSL to be hooked up in our new place so I can download all this stuff and try it out. Just when I thought I had the best picture I could get....... ;)

Originally posted by cyberbri
Does this work exactly the same as the resize algorithms in ffdshow? Ie., if I am resizing with Lanscoz 2 to 1920x1080, with Luma sharpening, can I disable this completely, switch to YV12 inside avisynth, and resize and sharpen inside the LimitedSharpen filter?

Also, I'm assuming that it would be mroe processor-intensive to have the sharpen come after the resize, even within LimitedSharpen, since you're working on a much bigger size - plus you'd need to up the values to get the same perceived results.

I can't wait to try this all out. Thanks to Socio for the great guide and all the information. Now all I have to do is wait for DSL to be hooked up in our new place so I can download all this stuff and try it out. Just when I thought I had the best picture I could get....... ;)

Yes the ffdshow Lanscoz and bicubic and Avisynth versions are almost identical so you can do it all inside Avisynth. You will for sure want to do any filtering before resizing as well.

This takes a lot of processing power to do, I have a P4 3.06 and can only resize to 1280X720 with LimitedSharpen using Overlay renderer. If you find you can't run it try the ModerateSharpen a few posts up it is a lot less CPU intensive.

I have a 2.8GHZ P4 with 512 ram, and I currently resize to 1920x1080, Lanscoz 2, and can use Luma Sharpen and/or Unsharp Mask after the resize, no problem, all with Overlay. Since all I'm doing is resizing and sharpening (I don't like denoise, on my 43" DLP anyway), when I get the chance I'll see if I can do it all inside avisynth.

Thanks again. :)

1. When using the cropping (FFDShow or AVISynth) and then AddBorders, what happens to this when playing a widescreen DVD who's anamorphic aspect ratio is closer to 1.75:1 and therefore has no black borders? Do you have to add and remove these croppings between each type of DVD?

2. Is there a difference in the pq AVIsynth provides when using the original LimitedSharpen option as opposed to the UnDot() option? If so, is the UnDot() option still superior to other sharpening techniques like LimitedSharpen is?

3. I know the following order is recommended in FFDShow:
ust make sure you do it in this order:

ffdshow crop
denoise3D
avisynth limitedsharpen
resize

but I am using the 20040801preview version and if I have Denoise3d before
AVISynth, ZP crashes to desktop everytime. I have a P4. So the question is, if I run Denoise3D after AVIsynth, then do I need to use FFDShow crop, or can I use the crop in AVISynth as it would apply to the image from that point down?

4. fdecimate() - I understand this boosts performance, but does it deteriorate the effect of Limitedsharpen?

5. Does LimitedSharpen2 require more CPU than LimitedSharpen?

6. Do you need to put lines in AVISynth for loading plugins like masktools, etc.? I thought the limitedsharpen.avs file loads all these automatically so all you need is the "Import:...." line

7. Is it better to not run any of this and resize to 1440x960, or to use Limitedsharpen and resize only to 1280x720?

A few things I can answer...

If you have a 1.78:1 or 1.85:1 DVD, you need to disable crop and add-border. You might have two Avisynth scripts, one with the add borders, one without.

The crashing may be due to the fact that Avisynth might require a YUY2 input, and running denoise converts the signal to YV12. If you can tell LimitedSharpen to accept YV12, that might help. However, I don't know if this is the cause, so don't hold me to it.



#7 - can you only do one or the other with the power of your system? I'd say the LimitedSharpen would be better, because you are sharpening the image. If you read through the thread, you'll also find that you can run resize within the LimitedSharpen script, which should help with CPU usage.


Currently I'm resizing to 2000x1600 with Lanscoz 2 with .80-1.00 on Luma Sharpen, waiting to get my DSL modem at my new place to download this stuff...

Thanks Cyberbri.

I have to test, but I think I cannot do 1440x960 & LimitedSharpen.

Another question:

Is using Alt-F in Zoomplayer and comparing the bmp files the best way to compare the same screenshot for various settings? Or is there some other way to do it that's better.

Anyone able to shed some light on my questions above? Socio?

I know that on my setup, I can't take screenshots post-ffdhow using Overlay. I think you may need to use VMR (and fix the levels/saturation to match your Overlay pic) to see the differences if you want to compare bmp files.

The only way I know to grab pics in overlay mode is to use ffdshow's grab feature. Other than that you need to use VMR as your renderer to use the alt-f function. Most all the pics I did in this thread using VMR as the renderer

I'm now settle on the following Avisynth config:

Import("C:\Program Files\AviSynth 2.5\plugins\ModerateSharpen.avs")
UnDot()
ModerateSharpen(0.1)
Lanczos4Resize(720,720)
Lanczos4Resize(1280,720)

DScaler5 decoder YV12 to FFDShow (Andy's preview version), Avisynth YV12, YV12 to Overlay on a Radeon.

For good flag film DVD that is.

regards,

Li On

A few questions about your config:


1) UnDot - is this a denoise filter?

2) Why only a setting of .1 for ModerateSharpen?

3) Why the two resizes?

Originally posted by Socio
The only way I know to grab pics in overlay mode is to use ffdshow's grab feature. Other than that you need to use VMR as your renderer to use the alt-f function. Most all the pics I did in this thread using VMR as the renderer

So if I'm using Alt-F, would that provide the best comparisons (i.e. bmp files).

Also, do you have any answers regarding the 7 questions I asked a few posts back?

UnDot is a very light denoise filter, I think. It remove a tiny bit of noise layer and has NO side effect whatsoever IMO.

0.1 is the ONLY value that has NO abnormal (jaggies) edge, IMO. Still better than any other sharpen effect in FFDShow, IMO.

2 steps resizes give the best overall picture quality with the least scaling artifacts, IMO.

This is my today setting. Tomorrow may be different! :D

regards,

Li On

Originally posted by JoeFigueiredo
1. When using the cropping (FFDShow or AVISynth) and then AddBorders, what happens to this when playing a widescreen DVD who's anamorphic aspect ratio is closer to 1.75:1 and therefore has no black borders? Do you have to add and remove these croppings between each
type of DVD?


If is does not have black bars do not try cropping unless you just want to trim the edges slightly, and if you do that you do NOT use the “addborders”. You only need to use the “addborders” if you have a widescreen movie that has black bars and you are resizing.



2. Is there a difference in the pq AVIsynth provides when using the original LimitedSharpen option as opposed to the UnDot() option? If so, is the UnDot() option still superior to other sharpening techniques like
LimitedSharpen is?

As Li On stated undot is just a very light denoiser it’s results will solely depend on the source on most sources you probably will not notice unless you do a still capture and really study it.

3. I know the following order is recommended in FFDShow:
ust make sure you do it in this order:

ffdshow crop
denoise3D
avisynth limitedsharpen
resize

but I am using the 20040801preview version and if I have Denoise3d before
AVISynth, ZP crashes to desktop everytime. I have a P4. So the question is, if I run Denoise3D after AVIsynth, then do I need to use FFDShow crop, or can I use the crop in AVISynth as it would apply to the image from that point down?


Yes that is the order you need to do it in for best performance if you try to run denoise3d after it will slow you down big time. I do not know why denoise3d makes your system crash. Make sure the ffdshow raw codec is set to YUV only maybe that will help.


4. fdecimate() - I understand this boosts performance, but does it deteriorate the effect of Limitedsharpen?


fdecimate() is supposed to remove duplicate frames and keeps them from having to be rendered, I am not sure it really has a impact or our application one way or the other.


5. Does LimitedSharpen2 require more CPU than LimitedSharpen?

Yes, it seems to take a little more CPU to pull off


6. Do you need to put lines in AVISynth for loading plugins like masktools, etc.? I thought the limitedsharpen.avs file loads all these automatically so all you need is the "Import:...." line


Ffdshow autoloads 2.5 plugins if you try to use any thing previous you would have to load a “loadplugEX.dll and then the pre 2.5 plugin.

7. Is it better to not run any of this and resize to 1440x960, or to use Limitedsharpen and resize only to 1280x720?

Resizing depends on your processing power I have a p4 3.06 and can barely pull off denoise3d-LimitedSharpen-resize 1280x720 and I mean barely. If you have a P4 3.8 I am sure you could pull off more.

Thanks Socio for your detailed response.

Here's some more clarification I need on the above questions:

2. Is Undot still superior to other sharpening including resize sharpening?

3. Isn't it better to set RawVideo to YV12 rather than YUV, as that is what DVDs are encoded as? This would therefore prevent a colour conversion in software.

4. So everyone should be using fdecimate then, right? If it doesn't hurt.

7. What version of FFDShow are you running? the 040801preview version seems very quick and on my 2.8 @ 3.04, I can do what you can do no problem.

Originally posted by Li On
UnDot is a very light denoise filter, I think. It remove a tiny bit of noise layer and has NO side effect whatsoever IMO.

0.1 is the ONLY value that has NO abnormal (jaggies) edge, IMO. Still better than any other sharpen effect in FFDShow, IMO.

2 steps resizes give the best overall picture quality with the least scaling artifacts, IMO.

This is my today setting. Tomorrow may be different! :D

regards,

Li On


I know the feeling.

Are you running Moderate Sharpen instead of Limited Sharpen for some reason?

And have you tried a setting of 2 instead of 4 for Lanscoz resize? That's what I'm running now, with a resize of 2000x1600 in ffdshow's resize (still waiting for modem at new place to download all this new software). I wonder if that would give you less scaling artifacts (have you read the thread here about Lanscoz vs Bicubic scaling? -- http://www.avsforum.com/avs-vb/showthread.php?s=&threadid=460922 ).

Originally posted by JoeFigueiredo
Thanks Socio for your detailed response.

Here's some more clarification I need on the above questions:

2. Is Undot still superior to other sharpening including resize sharpening?

3. Isn't it better to set RawVideo to YV12 rather than YUV, as that is what DVDs are encoded as? This would therefore prevent a colour conversion in software.

4. So everyone should be using fdecimate then, right? If it doesn't hurt.

7. What version of FFDShow are you running? the 040801preview version seems very quick and on my 2.8 @ 3.04, I can do what you can do no problem.

2) UnDot is a slight denoise tool, not a sharpening tool. So you probably don't need it at all, unless you have the extra CPU cycles and you want to feel that you are getting rid of a tiny bit of noise.

3) I think "ALL YUV" instead of "ALL" will not allow RGB or something like that - I'm not sure...

4) Don't know...


EDIT:
7) The August version seems to be what a lot of people are using. I am using the 10-12 version myself, from here:
http://sourceforge.net/project/shownotes.php?release_id=274595
But the notes say that the Avisynth scripting doesn't work, so it looks like Andy's August version is your best bet.

Hi cyberbri, READ my posts in this thread! :D

On my last projector the Panny AE200 with it's 856x480 panel the LimitedSharpen was as great as it gets. And there is no scaling artifact as there is no need to resize on a 480p output.

Then during last Christmas I changed to the Epson TW200H a 1280x720 projector. Now the LimitedSharpen (and the LimitedSharpen2 as well) at ANY setting still gives tons of JAGGIES edges. Together with scaling artifacts on (or without!) resize process. But that is no news as I already saw such effect on testing other 720p setup before I changed my own projector.

So after a month of tweaking, here comes my current config. Of course, YMMV!

Btw, I'm NOT using the FFDShow resize. I'm doing Lanczos resize INSIDE Avisynth. There is NO Lanczos2 resize in Avisynth. Besides I like Lanczos4 better anyway. Except's Andy's Preview version, Avisynth Lanczos resize works FASTER than the FFDShow resize. Andy's Preview version Resize does work faster than Avisynth, BUT it won't accept a >480p image to begin with.

Well, it seems quite complicate now! I hope you get the idea.

And I also tried over resize to bigger than output resolution and let the display card to downscale. In the end I still MUCH prefer the output resolution matching resize.

regards,

Li On

Tweaked and tested, and here's my final setup, and it looks phenomenal:

Sony GWIII over DVI
DScaler5, VMR9 Windowless
FFDShow (040801preview version)
RawVideo=YV12
FFDShow Crop (Top=40, Bottom=40)
AVISynth
Import...
fdecimate()
LimitedSharpen @ strength=20 (40 was too much & also edges were jagged)
ColorMatrix()
AddBorders(0,40,0,40)
Denoise3D (0,0,5, FAST)
Lanczos4 resize to 1280x720 with Luma=0.6; Chroma=0.0
Output=YV12 only

All of this is being done on a P4 2.8 @ 3.05, 800FSB and an Nvidia FX5900XT video card. I am right at the limits, and only get a couple dropped frames on wide pans, but fast moving action is relatively smooth.

Questions and further observations:
1. It was suggested several pages back that we need to boost to Saturation=78 and Hue=2 to compensate for what AVISynth does to washing out colour, but I actually find that I actually had to lower Saturation to 61 and increase Hue to 2 in order to get DVE calibration screens right. Any comments on this?

2. Does ColorMatrix() fix the problem about boosting saturation with LimitedSharpen? I do notice the colours look different than without it when I compare screenshots (reds are not as intense for example).

3. How much do you think the planned "renderless" mode in Zoomplayer is going to help with this sharpening technique?

I must say, this is the sharpest, clearest picture I have ever been able to attain. Hopefully someone can make this more efficient so we can have some headroom (CPU-wise) when using it.

Anyone able to answer any of my questions above?

Disregard the Saturation and Hue tweaking it is irrelevant now.

If you use Overlay as your renderer with LimitedSharpen then the color should be the same as the original.

Essentially ColorMatrix corrects the colors of mpeg2 streams of dvds so
if you use VMR as your renderer then you will need to use the Colormatrix plugin. Otherwise VMR will not render colors correctly after the LimitedSharpen process. Why VMR reacts to LimitedSharpen that way I have no idea.


As for how LimitedSharpen will work in ZP using renderless VMR mode remains to be seen. I would speculate a nice improvement but only time will tell.

Originally posted by Socio
I only have 7 files in my Avisynth plugin folder they are and should be all you need:

DirectShowSource.dll
LimitedSharpen.avs
TCPDeliver.dll
warpsharp.dll
ffavisynth.dll
MaskTools-p4-5.dll
ColorMatrix.dll



I'm about to try all of this and am checking what I have. After Avisynth install, and extracting what is needed from the other referred downloads I have everything except the ffavisynth.dll Where does this come from???

Thanks in advance...

I am stuck so far trying to load "LoadPluginEx.dll" from the IIPXcallscript.txt It is in the plugins folder with all the rest. Any ideas?

Well. my modem finally arrived, and I was able to download this stuff and test it out. Luckily I was able to get it up and running without a hitch. (My wife's probably been wondering what the hell I've been doing for nearly the last 3 hours). BTW, this was done on Star Wars Ep 5.

But after some testing, I found that the amount of sharpening was just too much, especially since this is so CPU-intensive it has to come before Resize.

I found that I like my 2000x1600 Lanscoz 4 w/ 1.00 Luma Sharp better than Limited Sharpen with a smaller resize. I guess the picture is just cleaner, because the sharpening isn't emphasizing the noise in the picture. I tried to figure out how to change the strength within the script, but nothing seemed to do anything except the one in the ffdshow call lines. But even at a strength of 2, it was so CPU-heavy I couldn't run it after a smaller resize.

So I guess I'm sticking with my previous setup. I did decide to try the new version of the DScaler 5 codec, and I think it may have an ever so slight edge over WinDVD 6, although I didn't do any caps and A/B them. But they're close enough I'll try it for a while.

Originally posted by BGPGuy
I am stuck so far trying to load "LoadPluginEx.dll" from the IIPXcallscript.txt It is in the plugins folder with all the rest. Any ideas?

If you try to run that IIPX script in ffdshow-avisynth like LimitedSharpen it would run about 2fps. Besides that ffdshow autoloads plugins so you would not need the LoadpluginEx.dll anyway unless you are trying to load older Avisynth 2.0 plugins.

Originally posted by cyberbri

But after some testing, I found that the amount of sharpening was just too much, especially since this is so CPU-intensive it has to come before Resize.




Well you can turn down the sharpening by adjusting the "Strength" parameter to 20, 10, or even 5 for very light sharpening or try the "ModerateSharpen" I mentioned a page or so back at 0.1 or 0.2.

I read some material about Avisynth and just want to know is it possible for someone can make some avisynth script to solve the lcd projector's problems? e.g. panel misalignment, vertical bending...

Has anyone tried the new ffdshow release???
Does it work with avisynth limitedsharpen ??
Or show the same problems that 12-10 release?
Thank you

Is anyone developing a standard "packaged" version of LimitedSharpen for ffdshow, so that the rest of us who would love to try it don't have to go through all the joy that y'all have been thru to get it to almost work?

Socio,

Have have been a great asset to us using ffdshow. so I was wondering if you can make a master folder that all one has to do is export it into the Avisynth file of ffdshow.

Thanks

Does last ffdshow 01-22 release work with Limitedsharpen ???

what about this pic?
compare it with the other yoda pics in this thread (mid to end of the 5th page).
AMAZING DIFFERENCE!

Twenty3,

good pic, what are you using and in what order??

Thanks

That is a great pic. I too would like to know what settings were used.

I realize there's less compression used in that jpeg than the ones on page eight and there might be differences in brightness, gamma and color saturation but it does look richer. Very very nice, almost like from a downsampled high definition source.

Thanks for posting it, looking forward to reading the settings.

i picked that 1 from another board.
topic for that pic was "a hacked version of LS2" (maybe socio can tell us more about that)

point is, like babel said, that the other pics seems to have something like a white haze over it. indicates that maybe gamma/brightness setting was not correct.
but i have to say, i hope it was the same with u babel, that i only noticed the diff after seeing the last 1 (the 1 i attached). u need - i needed - the comparison to realize it.
anyway i guess these settings only look great on that picture.
first of all its just 1 frame...i wanna see it in motion. further i think u can do these settings only with a good quality source (like sw in that case) otherwise u will see a movie which looks like sand strewed over it, especially in non-close-up scenes :)
and the last thing is: i think that pq is feasible with 5ghz cpu...otherwise u may see seconds per frame instead of fps :)

Originally posted by Babel-17
That is a great pic. I too would like to know what settings were used.

I realize there's less compression used in that jpeg than the ones on page eight and there might be differences in brightness, gamma and color saturation but it does look richer. Very very nice, almost like from a downsampled high definition source.

Thanks for posting it, looking forward to reading the settings.


That pic looks like one of mine from the Doom9 forum using a combo of LS2 and another script called IIP. You could never achieve that look in real time, the only way to do it is by backing up the DVD and re-encoding it using those scripts.

See this thread for details on how to do just that, also on the last page you will see other similar Yoda pics with variation of the IIP script as well.

http://www.avsforum.com/avs-vb/showthread.php?s=&threadid=478263&perpage=20&pagenumber=1

For anyone using "ModerateSharpen" posted a couple pages back a new version of the RemoveGrain filter has been released get it here (http://home.pages.at/kassandro/RemoveGrain/RemoveGrain.zip) .

It now contains better SSE2 optimizations as well as new SSE3 optimizations.

Thanks!

I love this thread. :)

FYI:

LimitedSharpen, LimitedSharpen2, and ModerateSharpen all work in the newest ffdshow, 20050303-sse.

http://ffdshow.sourceforge.net/tikiwiki/tiki-index.php?page=Getting+ffdshow

What items need to be chekced in AVISYNTH in FFdshow?
I have Avisynth checked and that is it

What items need to be chekced in AVISYNTH in FFdshow?
I have Avisynth checked and that is it
It's been stated before in this thread but her it is:
"Avisynth" of cause has to be checked.
"Add ffdshow video source" has to be checked.
"YV12" has to be checked.
And thats it.

Henrik

Here is what I am using:

Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpen.avs")

crop(8,8,-8,-8,true)
LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=3,strength=40)
ColorMatrix()

I only have 6 0f the 7 files in my Avisynth plugin folder they are:

DirectShowSource.dll
LimitedSharpen.avs
TCPDeliver.dll
warpsharp.dll
ffavisynth.dll don't have and where is this file from
MaskTools-p4-5.dll
ColorMatrix.dll

I get an error if i CHECK YV12.

Any help on this would be great.

Chad

When you install ffdshow you can specify something about avisynth to be installed.
It might be that this is the step to remember to do to get the ffavisynth.dll installed.
I think ffavisynth is the file that makes it possible for avisynth to receive input from ffdshow.

Henrik

I've moved away from using the crop function in ffdshow and use the crop in avisynth instead.
When the crop in ffdshow is set to less than 46 in top and 36 in bottom it producess very heavy cpu load. With 46/36 I have something like 70% cpu load and with 44/36 or 46/34 or anything else less than 46/36, I hit 100% cpu load with HEAVY stutter. This is not the case with NTSC (480) but only with PAL (576).
So it seems like the crop funktion in ffdshow has some error that make it produce heavy cpu load when more than 494 lines is passed on from the crop funktion.
With avisynth crop instead, I can go as low as 8/8 and stay around 94% cpu load.
That way I never have to go into crop to adjust for each DVD movie.

This is finally possible for me to do because I've moved to a P4 northwood 2.8GHz and overclocked it to 3.15GHz. I still use a P4PE MB.
Codec used is Dscaler 5 006 for video and powerDVD 5 for audio.
I've also updated ffdshow to 20050303-sse and use ZP 4.10 instead of ffdshow-20040709_SSE2 and ZP 4.03. Not because they give me more speed but they ar more stable when it comes to start playback of a DVD.
A strange side effect seems to be that Reclock doesn't work properly anymore and audio sometimes stutters and often gets out of sync.

Henrik

Hmm... P4PE! Me too! 2.4G Northwood 133 (533) bus with 1 DDR333 512M. Overclock to around 161 bus at around 2.9G. Radeon 9500 256bit softmod to 9700.

DScaler5 0.006 codec YV12 out. Andy's Preview version FFDShow, Avisynth 2.55. Overlay YV12 out. ZoomPlayer 4.1.

For normal good flag 24fps NTSC movie:
- Avisynth LimitedSharpen
- FFDShow Resize Lanczos 4 to 1280x720

Smooth playback, CPU at around 90-100%

But there is stutter for 30fps NTSC and 25fps PAL (576p) playback and CPU at 100%. For those content I need to use Avisynth ModerateSharpen, instead of LimitedSharpen.

So I keep thinking how to upgrade the HTPC for more CPU power. I'm now thinking maybe get a Prescott core P4 2.8G with a 533M bus and overclock to 666M (166M FSB) to 3.5G on the old P4PE. I wonder if the old board can handle it...

regards,

Li On

@Li On
You're doing better than me if getting a 161 bus speed to work. My 3.06GHz P4E blue screens at anything more than a FSB speed of 147 (or maybe it's the RAM). At max speed my machine runs limited sharpen fine on NTSC disks but can't handle PAL resolutions.

For the moment I've taken limited sharpen out. I can resize to 1280*720 and run removegrain and ffdshow's asharp after the resize which produces an excellent image with this overclock. There isn't a lot of headroom though. I'm using an 865 mainboard so I can put a 3.2Gig or 3.4Gig Prescott in there and overclock a little (if I could work out how to silently cool it).

Still thinking about the Denon A1XV btw, what's your opinion on picture quality?

DDR333 ram runs at 166M FSB standard, unless you use DDR266 ram. The 130nm Northwood tops at around 3.2G so you already have a excellent overclock at around 3.4G at 147 FSB.

The Prescott core goes much higher and many people report 4G+ overclock. I just hope that the old board can handle the new CPU. Then I don't need to break down the whole PC and no Windows reinstall.

The Denon A1XV is fine for DVD playback at 720p but I use the HTPC for much more type of content viewing. I'll always use a HTPC for primary viewing so I'll only get a cheap DVD player (such as a 1910) as a backup solution.

regards,

Li On

Li On you can bios upgrade your P4PE to use 200MHz FSB if you want to use a Prescott.
I stayed with the Northwood because it stays 10 degrees celsius lower than the Prescott for the same cpu clock speed, and I want to keep it quiet.
I use the MIST cooler with a Papst 12 dB fan so it's virtually silent.
A 19 dB Papst fan only gave me 2 degrees lower temps but was to noisy for my taste.
I have successfully (and stable) run my 2.8GHz Northwood at 3.25 GHz (153 GHz FSB) but not higher.
I use 512MB of pc3200 RAM running at 412 MHz (or is it 418?),
and use a Radeon 9800 Pro 128MB.
I changed the cooler on the Radeon with a large heatsink from a Pentium 1 or 2 (used the Arctic silver 2 component glue) and mounted a 60mm 20mm thick lownoise fan with a resistor to slow it down to almost silence.

Henrik

My P4PE is already on the latest 1007 BIOS more than a year ago.

I don't really want to use 200M FSB as I'm sure I can't overclock from that.

There is a Prescott P4 at 2.8G using 533M bus with a 21x multipiler. If I can overclock it to 166 FSB at 666M bus which is my DDR333 ram speed then I can get 3.5G from the CPU. I think that will be a pretty good result on the old P4PE if that's possible.

I use a Zelman 7000AlCu HSF for CPU.

regards,

Li On

Li On, is it possible to add TomsMoComp to the mix and still get smooth playback? I wonder how much CPU that would take...

The purpose of Crop is to reduce the amount of picture resize, etc. is working with. If it's getting you near 100% CPU usage with a high-end P4, why use it?

Also, the fasted FFDSHOW is going to be Andy's hand-coded August version (here (http://ffdshow.leffe.dnsalias.com/andy2222/))

And the newest version of Zoom Player is 4.50, which has VMR9 Renderless and a box to check for Exclusive Mode. That should theoretically give you the best performance.

I know the purpose of Crop. but as I say, there seems to be some "bug" in the ffdshow crop function.
It should be better to crop in ffdshow before sending the pic to Avisynth to reduce the bandwith of the stream but it turns out that when I crop less and less in ffdshow the cpu load rises. So far so good, this is to be expected, but suddenly when I reduce cropping a little more the cpu load jumps very far to beyond 100 % cpu load. It should have only elevated from 70 to 75 % but instead it went from 70 to 120 %.
Not using ffdshow crop and transfere the whole lot to avisynth and only crop a little here keeps the cpu load on 94%.
This indicates to me that crop in ffdshow has some bug.....

Henrik

Ok still get error if I check YV12 in avisynth.
Using latest zoomplayer beta version
Using july 2004 version of ffdshow.

In zoomplayer I am using Dscaler, reclock, YV12, vrm9 windowless.
I also attatched ffdshow settings.

chad

cburbs,
"Input colorspaces: YV12" HAS to be checked.
Also, Do you have the "ffavisynth.dll " in there now?
If not, then get it there.

Also, DON'T use all the others stuff like resize and stuff UNTILL you have this
LimitedSharpening working. It just complicate things.

Henrik

Yes I have the ffavisynth.dll file in the plugins folder.
I will try tonight to uncheck everything to see what I get.
Like I said my error pops up the minute I check yv12.

Thanks for all the help !

Chad

Are you outputting YV12 from your video codec? It may be in YUY2, so you would need a filter in front of it like Unsharp Mask @ 1, in order to force the conversion from YUY2->YV12.

In zoomplayer I am using Dscaler, reclock, YV12, vrm9 windowless

Anyone tried Avisynth 2.5.6 Beta3? Will it eat up less CPU loading when comparing with 2.5.5? Thanks.

I've found that LimitedSharpen can be speed up a little bit by including the "optimized" scripts directly into ffdshow. If your mpeg decoder output YV12 and you are using LimitedSharpen with parameters ss_x=1.0, ss_y=1.0, smode = 3, and strength=30, you may replace your original scripts in ffdshow config with the following:


tmp=last
dark_limit = tmp.inpand()
bright_limit = tmp.expand()
minmaxavg = yv12lutxy(dark_limit, bright_limit, yexpr="x y + 2 /")
normsharp = yv12lutxy(tmp, minmaxavg, yexpr="x x y - 0.3 * +")
yv12lutxy( bright_limit, normsharp, yexpr="y x 1 + < y x 1 + ?")
yv12lutxy( dark_limit, last, yexpr="y x 1 - > y x 1 - ?")
tmp.mergeluma(last)


The sharpen strength can be changed by modified the "0.3" of fifth line. For example, if you want sharpen strength to be 40 as suggested by Socio, you can simply replace "0.3" by "0.4"

Enjoy.

Sorry, could you please explain more detail? Where to put/change your "script" into? In FFDShow Avisynth script of the LimitedSharpen.avs script? After the modification, what the correct way to call LimitedSharpen in FFDShow?

Thanks in advance.

regards,

Li On

In your ffdshow config of avisynth, you should have two lines, such as


Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpen.avs")
LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=3,strength=30)


Simply replace the above two lines with the scripts of my last post.

The scripts are extracted from "LimitedSharpen.avs" and optimized with ss_x=1.0, ss_y=1.0, smode=3 and strength=30. With the optimized scripts, you can no longer change any parameters, except the sharpen strength. To change the strength, you have to modify the code directly as mentioned in my last post.

Hope this help.

Got it. But... if I remember it right the LimitedSharpen.avs is a very long script, how can you reduce it to just a few lines? Besides speed speedup, will there be any picture quality difference using the same "ss_x=1.0, ss_y=1.0, smode=3 and strength=30" parameters?

regards,

Li On

The original scripts got checking on color space and have many scripts for handling different parameters. When removing those scripts, only few lines are needed. There will have no different in image quality. Only speed up a little bit.

I cant find this SSXS-a-LIKE" zip any more...can anyone link me to it?

Why do I get not a ColorMatrix function exist or something like that? I have the ColorMatrix dll in the plugins as usual? Same to happens to any command like lancoz etc...I am able to run LimitedSharpen scripts though.

I don't know if there is anyone besides me still using LimitedSharpen with ffdshow for DVD and video playback but Didee over at Doom9 has released is a new version of LimitedSharpen get it here (http://home.arcor.de/dhanselmann/_stuff/LimitedSharpen_(modded-29Oct2005).rar) just rename it LimitedSharpen and put it in your plugin folder.

This version is faster then the old version at least for me and has a new smode which is smode 4.

Speed wise the modes now break down like this:

Smode 2 fastest
Smode 1 fast
Smode 3 highest quality
Smode 4 high quality and slightly faster than Smode 3

There are also new and improved softening parameters undershoot and overshoot, overshoot looks best to me.

Also grab the latest Masktools 1.58 (http://manao4.free.fr/MaskTools-v1.5.8.zip) and Colormatrix (http://www.avisynth.org/warpenterprises/files/colormatrix_25_dll_20050223.zip) again you only need Colormatrix if you use VMR7 or VMR9 rendering.

I have been using the new Limited sharpen for a few days now and took some pics however they are NOT the exact same frame but should be close enough for you to see the effects of LimitedSharpening. I am using Zoomplayer WMV Pro 4.51, NVDVD 2.55 and the latest ffdshow FFDSHOW 2005-11-03 (http://www.afterdawn.com/software/video_software/codecs_and_filters/ffdshow.cfm) and Avisynth 2.56a (http://sourceforge.net/project/showfiles.php?group_id=57023) .

In ffdshow I just use Blur & Nr Denoise3D Luma 0, Chroma 117, Time 3.88,HQ enabled and of course avisynth enabled with Limitedsharpen calls.

The first pic is normal (nothing enabled) the second uses the above settings with this call which has been pretty much the standard call through out this thread.:

Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpen.avs")

LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=3,strength=40)

colormatrix()


The Third pic uses the same call but with using the Overshoot parameter at 7, which seems to keep most of the great sharpening effects from LS but smooth it down so it is not so 3dish, ( this is what I now use on most all movies). You can adjust to taste the lower the number the more softening the higher the less softening and 0 is off.

Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpen.avs")


LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=3,strength=40,oversho ot=7)

colormatrix()

The next pic uses Smode 4 in the call:

Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpen.avs")


LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=4,strength=40)

colormatrix()

EDIT: added new pic of a tweaked Smode 4 (the last pic) using a tweaked new version of LimitedSharpen and newer version of masktools, links to both can be found on next page.

Here is the call I used for this pic, this call it actually came from Didee over as doom9 if you noticed this call uses a "soft" parameter instead of "overshoot" and looks much better.

Note:If it causes to much CPU usage when resizing you can lower the number from 150 to 50 and 34 to 12 or adjust to taste.

Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpen.avs")


LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=4,strength=150,soft=3 4)

colormatrix()


http://img224.imageshack.us/img224/31/warriorsnormal7cx.jpg
http://img224.imageshack.us/img224/5438/warriorslsmode39gy.jpg
http://img224.imageshack.us/img224/7483/warriorsovershoot2ak.jpg
http://img224.imageshack.us/img224/7448/warriorslsmode41xi.jpg
http://img204.imageshack.us/img204/797/warriorsnewsmode44ys.jpg

Great.
Allways good if processing load is reduced.
The "old" LimitedSharpen is so cpu-heavy that I mostly just use it on NTSC material (480 instead of 576 lines).

Henrik

Hey Socio,

Thanks for all the great information thus far. I just had a couple questions/problems...

1) I'm gettin' an error with the latest LimitedSharpen.avs, it loads, but then the ffdshow OSD states
Script error: there is no function named "removegrain"
(C:\Program Files\AviSynth 2.5\plugins\LimitedSharpen.avs, line 69)
(ffdshow_filter_avisynth_script, line 3)

My calling lines in ffdshow->AviSynth are the same as yours above.

2) With the latest version of MaskTools (1.5.8), can we just use that .dll if we have a P4 (it seems to work) or should we still be using the MaskTools-P4-5.dll? Neither fix the above problem...

3) Curious as to what your luna/chroma sharpen settings were in your lanczos resize...

Should have searched the post first...I got the RemoveGrainSSE2.dll in the AviSynth plugin directory, but now the error states

RemoveGrain: Invalid mode 19

What's the deal here? lol...

Oops sorry about that!

You need the latest version of RemoveGrain here (http://home.arcor.de/kassandro/RemoveGrain/RemoveGrain.rar) older versions do not have the mode 19 function.

Also Masktools1.58 works with P4 and AMD and has the latest tweaks so I would use it.

Thanks for the help...all is working gloriously now...except for a new problem

For some reason Andy's latest SSE2 ffdshow build (08012004a_preview) works great for DVD playback, but won't allow any other video content to be played either in ZP or WMPlayer. XViD files, DiVX, anything just starts to open up and then the program closes down. I uninstalled the ffdshow and installed a previous version and it appears to work, however the older versions lack the "FAST" Denoise3D, which was a noticeable improvement in my eyes.

Has anyone else experienced this?

There is a new completely re-written version of masktools available here (http://manao4.free.fr/masktools-v2.0a7.zip) ,you will need to delete the old version of masktools in your plugin folder and replace it with this new one.

This version seems to cut CPU usage by as much as 30% at least for me anyway.

However it will not work with that new Limitedsharpen I posted or any other version. I had to re-write some of the code in that new version to get it to work, you can download my reworked version of the new Limitedsharpen

here "fixed again" (http://www.savefile.com/files/3872085)

NOTE: For some reason it takes several seconds for this new Masktools/Limitedsharpen combo to initialize, so if you start up Zoomplayer hit the playbutton and it looks like it is locked up its not just give it time it may take up to 20 seconds.


Edit: I just uploaded a new version of LS I had to add two parameters, sorry about that.

Edit again, I just uploaded a newer version with an Smode 4 tweak that Didee the script author came up with that works with the new version of Masktools.

Also I am going to post another pic in my post on the previous page so you can see what the tweaked Smode 4 looks like using a "soft" parameter instead of "overshoot".

Thanks for the updates...however they aren't fixing my problem with the crashing due to ffdshow, but I at least found a temporary work around by putting the Levels filter before Blur & NR, Avisynth, and Resize just to get it workin'...silly ffdshow...

And you weren't kidding, this does take substantially longer to initialize and get going...hehe

Socio...how do you get your screen grabs? I would like to share my results as well for this awesome procedure...

Socio...how do you get your screen grabs? I would like to share my results as well for this awesome procedure...

If you use Zoomplayer you have to use VMR9 windowless as your renderer and use alt + f to take a screen capture.

Thanks for the updates...however they aren't fixing my problem with the crashing due to ffdshow, but I at least found a temporary work around by putting the Levels filter before Blur & NR, Avisynth, and Resize just to get it workin'...silly ffdshow...

And you weren't kidding, this does take substantially longer to initialize and get going...hehe

Exactly;

You want any noise reduction or other filtering to be done before avisynth and any resizing to be done after avisynth.

The slow starting has to do with the new masktools plugin, the masktools author is aware of the problem and looking into it, hopefully we will have a fix soon.

I know this looks rookieish, so BACK OFF...hehe

These were taken using ZP 4.51 and the latest NVidia DVD Decoders (1.02-185)

First: Normal
http://img87.imageshack.us/img87/593/grab21qu.jpg

Second: Denoise3D FAST, Avisynth, Resize 2x
http://img383.imageshack.us/img383/2245/grab13zz.jpg

I think the most impressive things to note are the definition in the eyebrows/eyelashes and the new accentuation (word?) of the pores around the eyes/nose area...

Another screenie for your viewing pleasure...this one is a HUGELY noticeable difference. I wanted something textured with a lot of contrast to try and show this thing off and I think the results speak for themselves...

First: Normal
http://img453.imageshack.us/img453/1982/grab30iz.jpg

Second: Denoise3D FAST, Avisynth, Resize 2x
http://img453.imageshack.us/img453/4333/grab45we.jpg

A final sample...note the hair on Sulley...

First: Normal
http://img353.imageshack.us/img353/7980/grab57um.jpg

Second: Denoise3D FAST, Avisynth, Resize 2x
http://img353.imageshack.us/img353/1015/grab63kq.jpg