I second that. The FFDshow+Limited Sharpen combo seems to have the best balance between cleaning the image and maintaining the film look of the source.

Also, take a look at the background, noteabley the tree behind the saloon in the upper left hand corner. The 2nd image is clearly crisper.

Kai

I don't get it...

I have a AMD-64 3000 with a gig of memory, 6600GT, all pretty good stuff.

Whenever I run the limited sharpen I get all these small stutters and my CPU (through ffdshow OSD) is at 100%.

I d/led the scripts and added the calls in ffdshow. ALL I have turned on in ffdshow is limited sharpen!

ZP, nvidia 185, avisynth 2.5 and this script:

# LimitedSharpen() ( a modded version, 29 Oct 2005 )
#
# A multi-purpose sharpener by Didée
#

I've tried it with dscaler 5 too, still stutters.

The only way I can stop the stutter is to drop limited sharpen down below 1.0 in ffdshow. I don't like that because it's too soft.

Any ideas? I REALLY like the way it looks at 1.25 but I really HATE the stutter.

BTW, I have WinDVD7 and DNM looks awesome. But as BBQ said, sometimes movies look "fake" like watching a video and not film.

Thanks...

I don't get it...

I have a AMD-64 3000 with a gig of memory, 6600GT, all pretty good stuff.

Whenever I run the limited sharpen I get all these small stutters and my CPU (through ffdshow OSD) is at 100%.

I d/led the scripts and added the calls in ffdshow. ALL I have turned on in ffdshow is limited sharpen!

ZP, nvidia 185, avisynth 2.5 and this script:

# LimitedSharpen() ( a modded version, 29 Oct 2005 )
#
# A multi-purpose sharpener by Didée
#

I've tried it with dscaler 5 too, still stutters.

The only way I can stop the stutter is to drop limited sharpen down below 1.0 in ffdshow. I don't like that because it's too soft.

Any ideas? I REALLY like the way it looks at 1.25 but I really HATE the stutter.

BTW, I have WinDVD7 and DNM looks awesome. But as BBQ said, sometimes movies look "fake" like watching a video and not film.

Thanks...
I've had just about every problem known to man with this crap since I started it, but I think I've finally gotten a pretty good handle on it. I'll assume that you are using Zoomplayer, and the Nvidia Purevideo Decoder.

With my particular system (AMD64 4000+), overlay mode is the performance king. I can do up to 1.2 SS in Limited sharpen, 3dDenoising, and resizing up to 720p at a max of 80% CPU usage. Knocking off the SS brings it down to an average of 40% usage. To me, the picture quality is the worst, though. It still looks good, but once you use one of the VMR modes, you won't want to use it.

The VMR modes are a mixed bag. Renderless performs the worst, regardless of whether I'm using the "exclusive" mode or not. Perhaps future optimizations within Zoomplayer will fix this, but for now they should not be considered options if you want to do anything other than resizing.

The VMR9 Windowed and Windowless options, as well as VMR7, will handle denoising, limitedsharpen with no SS, and resizing with about 80% CPU usage. This is acceptable to me because I feel the image quality you get from using the VMR9 modes makes up for the loss in SS. After tweaking the colors a bit from the Nvidia decoder, the image quality is in my opinion stunning.

Using Nvidia Post Processing in ANY mode results in a pretty nasty speed hit, so avoid using it.

Kai

I don't get it...

I have a AMD-64 3000 with a gig of memory, 6600GT, all pretty good stuff.

Whenever I run the limited sharpen I get all these small stutters and my CPU (through ffdshow OSD) is at 100%.

I d/led the scripts and added the calls in ffdshow. ALL I have turned on in ffdshow is limited sharpen!

ZP, nvidia 185, avisynth 2.5 and this script:

# LimitedSharpen() ( a modded version, 29 Oct 2005 )
#
# A multi-purpose sharpener by Didée
#



BTW, I have WinDVD7 and DNM looks awesome. But as BBQ said, sometimes movies look "fake" like watching a video and not film.

Thanks...


Make sure you are using the versions on the ffdshowGuides (http://free.hostdepartment.com/f/ffdshowGuides/index.html)

Avisynth 2.56 and fast limitedsharpen and the lastest masktoools 2.0 should give you a bump in speed.

On Trimension;

In WinDVD 7 like WinDVD 6 that "fake" look is due in part to WinDVD's Overlay settings thus that"fake" look is not nearly as prevalent when used in ZP or GAM.

The third image does appear a little soft but more natural, I noticed that sometimes the sharpening can make images kind of 3dish almost pop off the screen but the TrimensionDNM seem to keep most of the sharpening yet de-3dishes the image.

You can always crank up the strength parameter or try supersampling in Limitedsharpen to make it more crisp. However you need to see it in ZP or GAM first to get a real visual judgement of what it does.

Now if we can only find a way to use WinDVD 7 "enhanced" version of TrimensionDNM that uses less processing power to work so we can re-size or kick in Supersamplging easier it would even be better. :D

Make sure you are using the versions on the ffdshowGuides (http://free.hostdepartment.com/f/ffdshowGuides/index.html)

Avisynth 2.56 and fast limitedsharpen and the lastest masktoools 2.0 should give you a bump in speed.


Ah, that reminds me. For joeuser2, these are the avisynth DLL versions I'm using:

Warpsharp YV12 - only difference from regular Warpsharp is that it is YV12 native, and it adds the Xsharpen function (which I don't use). Saves you a manual colorspace conversion step if you notice the colors are a little off with the standard Warpsharp.

Masktools 2.0 alpha 11 - Latest version

RemoveGrain package - SSE2 versions. Definitely a bit faster than the standard builds, as long as your AMD64 core revision supports the SSE2 extensions. The SSE3 version crashes, even though my processor allegedly supports the extensions.

For FFDshow, I'm using the latest daily build from Milan's site, which seem to be pretty quick and stable. I plan on giving the now ancient "Andy" SSE2 builds a try to see if they will help reduce the Denoising and Lacznos resizing performance penalites, but have my doubts that they will be of any benefit to my CPU due to the Intel compiler he uses.

Kai

Socio,

I went back to your web site (BTW, thanks for making it) and used your links and got fresh versions of a few things.

BBQ,

I used SSE3 and it works. I'll grab Warpsharp YV12 and see if that makes things better.

BTW, you Hawaiian? I believe Kai means sea/ocean if my memory is correct. Lived there for 4 years.

Well, I reloaded everything from a fresh d/l and at least limitedsharpen is working. I have it on 1.3, 1.3 and it looks good. I still get some stutters when I turn on resize, lancoz 2 and 3 stutters at 1280x720. But the picture looks pretty good with just LS so I'm pretty happy.

Thanks for the tips. And thanks to those who made, keep current, and contributed to that limitedsharpen web site. I couldn't have done it without it.

Oh My Freaking Goodness!! Tonight has been the biggest PITA trying to update to all the new plug-ins, etc. And 99% of the time, it was my dumb ass that kept messing things up - and it would take me 30 mins per mistake just to figure out what I did wrong. My wife is making me sleep in the guest room tonight because, well, I got a "little" angry.

The good news is that everything works now, but I am too tired to hook it up to my PJ. Tomorrow night will be fun.

Anyway, the new updates have put me just under the ceiling - I can Resize now @ 1280x720 without any stutters - before I would get an occassional stutter that drove me through the roof.

I was curious as to the strength most people use. 40 seems good to my eyes, but I will have put it up against Avia's sharpness patterns to see what is going on - I don't see any visible ringing or halos.

Excellent thread, Socio - and others who have helped.

I was curious as to the strength most people use. 40 seems good to my eyes, but I will have put it up against Avia's sharpness patterns to see what is going on - I don't see any visible ringing or halos.
How much strength to use depend on what projector you use and how 3D-like you prefer the movie to look.
I use a Sony 1271 CRT projector and use strength 100-160 on it. That gives me almost as much 3D as HiDef 1080i material do without to much artifacts. I suspect though that the CRT is getting soft on me... because of the many hours the unit has on it now.
A month ago I tested a JVC HX2 and a Hitachi TX200 and with them I used much lower values (lowest on the JVC) around 30-50.
I've never needed any noise reduction or grain removal of any kind though.

Henrik

Socio,

Great work on the entire Avisynth LimitedSharpen pre-processing thread :)

Before I go any further here are my HTPC specs:

Claritas 390M Stands Unique HTPC Case
P4 3Ghz Northwood 800FSB
Zalman CNPS7000B CPU Cooler
Asus P4800-VM with onboard SPDIF out
2x 512mb 400mhz Crucial DDR
ATI 9600XT Passive gfx card
Samsung Spinpoint 2x 250GB SATA HDD
Seasonic 350 PSU
MCE 2005 SP2
MCE Catalyst 5.4 video driver
TheaterTek 2.2

I have been using ffdshow with Blur & NR, Resize 1280 x 720 Lanczos2, Luma sharpen .60 for playback on all movie materials for over a year now and have been happy so far.

I recently stumbled across your thread about a week ago have tried using limitedsharpen with ffdshow on all movie playback using the exact settings on your recent guide and following the installing as instructed.

Using your reccomended limitedsharpen settings (but reducing the strength from 40 to 30), I have found that limitedsharpen gives the sharpest picure with the least about of ringing/halos and artifacts on my Hitachi TX100 1280 x 720 PJ but there is one very slight annoying drawback though and that is I can see a slightly more shimmering effect (aliasing?) on diagnals that wasn't present before.

Cpu usage with the above settings is around 50-60%. If I up the super sampling on both y & X from 1.0 to even 1.1, my cpu just maxes out. However that does seem to reduce the aliasing effect.

Is there any way I can optimize my HTPC further so that I can achieve a higher super sampling than 1.0 without my cpu maxing out?

How can I get hold of LimitedSharpenFaster as there doesn't appear to be a link for it on your guide?

Thanks

For those who has enough CPU juice....

Try running Resize before Avisynth.

I do a 2xlanczos2 resize before my avisynth limitedsharpen script. It takes alot more CPU power but produces a cleaner image. At least for me.... ;)

N3W813,

I've just seen you latest ffdshow settings on the 'official ffdshow settings' thread and your HTPC specs seems very similar to mine (3ghz) with your gfx card being much better.

Base on our similar HTPC specs, how on earth are you able to set your super sampling to 1.3 on your limited sharpen settings and what CPU usage do you get from those settings?

My CPU maxes out at just super sampling 1.1 :confused:

Socio, I'm getting an error when starting up a dvd. It says "import couldn't open 'C:\Program Files\Avisynth 2.5\Plugins\avsfilter.vdf.avs"

I have all the settings set up as per the instructions on you webste, I'm using the 11/29 ffdshow build with the preset "dvd" call script you provide too. Any ideas what the problem is???

Vidcan,

It can't find the file. Double check that you named it the same as what you have in ffdshow. BTW, why did you add that .vdf? Just name it "avsfilter.avs"

Come to think of it...I use limited sharpen and that is ALL I call from ffdshow. I have a "limitedsharpen.avs" file in the avisynth directory and I call it in ffdshow.

tk2001,

I have the same problem. I have no idea HOW these people are doing all this stuff! I can't go any higher than 1.1 without 100% CPU and forget about adding resize too! I'm running limitedsharpenfast too.

N3W813,

I've just seen you latest ffdshow settings on the 'official ffdshow settings' thread and your HTPC specs seems very similar to mine (3ghz) with your gfx card being much better.

Base on our similar HTPC specs, how on earth are you able to set your super sampling to 1.3 on your limited sharpen settings and what CPU usage do you get from those settings?

My CPU maxes out at just super sampling 1.1 :confused:

Sorry, forgot to state....when I do resize before avisynth, my limitedsharpend is at 1.0 and 1.0.

When I use LS before resize, I then use 1.3 1.3 for the settings.

Sorry for the confusion. :P

Has anyone encountered terminal errors with TT/ffdshow/avisynth? I'm a long time ffdsow and TT user but I admit avisyth is another magnitude of complication. I have scoured the treads and have seen lots of issues but no one reporting "TT has encountered a serious error and needs to close". Once this happens, the only fix I have found so far is to uninstall everything, including TT and start from scratch. I REALLY want to give that thing a try and any help is greatly appreciated.

Vidcan,

It can't find the file. Double check that you named it the same as what you have in ffdshow. BTW, why did you add that .vdf? Just name it "avsfilter.avs"

Come to think of it...I use limited sharpen and that is ALL I call from ffdshow. I have a "limitedsharpen.avs" file in the avisynth directory and I call it in ffdshow.

tk2001,

I have the same problem. I have no idea HOW these people are doing all this stuff! I can't go any higher than 1.1 without 100% CPU and forget about adding resize too! I'm running limitedsharpenfast too.

I just have these settings in ffdshow:

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

LimitedSharpen(ss_x=1.3,ss_y=1.3,Smode=3,strength=40)

I'm not calling avsfilter.vdf.avs anywhere.
After playing around with ZP a little bit I was able to get it working only after I turned off VMR renderless mode. Even then it gives me the avsfilter error twice which I press ok to and then it starts to play. Also, when I set smode to 4 it crashes ZP, smode 3 and 2 worked fine and I think 1 would too.

The only way the avisynth route would work for me is with VMR renderless mode because on Nvidia 6800 graphics cards when you go to 1080i resolutions you will get stutter even with like 50% cpu useage. BTW this is on Andy's 08 2004 ffdshow build.

Any ideas???

I'm trying to get ahold of the WarpSharp filter but the AviSynth Site is down, can anyone hook me up? Email is all one word: snowaddict at gmail dot com
Thanks

Sorry, forgot to state....when I do resize before avisynth, my limitedsharpend is at 1.0 and 1.0.

When I use LS before resize, I then use 1.3 1.3 for the settings.

Sorry for the confusion. :P

N3W813,

What is your CPU usage when you have your super-sampling set @ 1.3 1.3 and do you have HT enabled?

Also are you using LimitedSharpenFaster and if so can you post a link to it?

Thanks

Btw any tips on how you are able to acheive ss 1.3 1.3 using ur 3ghz P4 without maxing out the CPU would be much appreciated among those trying to trying to acheive the same goal using similar set-ups. :)

N3W813,

What is your CPU usage when you have your super-sampling set @ 1.3 1.3 and do you have HT enabled?

Also are you using LimitedSharpenFaster and if so can you post a link to it?

Thanks

Btw any tips on how you are able to acheive ss 1.3 1.3 using ur 3ghz P4 without maxing out the CPU would be much appreciated among those trying to trying to acheive the same goal using similar set-ups. :)

I'm not quite in your CPU class - all I have is a 2.8 Northwood and no OC - but that sounds pretty outlandish to me - both the 1.3 stuff before and even more with 1.1 stuff AFTER resize. (?!) Maybe without any kind of denoiser running? But even then, it sounds more like what I would expect from a FAR more powerful CPU. Getting up around the 4G level, I could imagine running something like that. Getting that kind of performance underneath that CPU level sounds more like an urban myth to me.

With a good denoiser running and a Lanczos4 resize to 1280X720 I have to leave my settings at 1.0. Even doing that I have to turn up my CPU and system fans - otherwise the CPU goes into overheat protection after about an hour of DVD - something my son can't even do playing running Madden 2005 at SXGA for two hours straight.

(I've also got to say that I can't notice the difference between a 1.0 and a 1.3 on my Benq 6200 pj, though that's only native XGA.)

Your mileage may vary, of course ... I'm certainly not trying to accuse anybody of faking it.

Mike

I'm not quite in your CPU class - all I have is a 2.8 Northwood and no OC - but that sounds pretty outlandish to me - both the 1.3 stuff before and even more with 1.1 stuff AFTER resize. (?!) Maybe without any kind of denoiser running? But even then, it sounds more like what I would expect from a FAR more powerful CPU. Getting up around the 4G level, I could imagine running something like that. Getting that kind of performance underneath that CPU level sounds more like an urban myth to me.

With a good denoiser running and a Lanczos4 resize to 1280X720 I have to leave my settings at 1.0. Even doing that I have to turn up my CPU and system fans - otherwise the CPU goes into overheat protection after about an hour of DVD - something my son can't even do playing running Madden 2005 at SXGA for two hours straight.

(I've also got to say that I can't notice the difference between a 1.0 and a 1.3 on my Benq 6200 pj, though that's only native XGA.)

Your mileage may vary, of course ... I'm certainly not trying to accuse anybody of faking it.

Mike

No faking here! Got my P4 530J (3.0Gig) processor and I turn Hyperthread on (in fact, with my HTPC, everthing runs slow without it on).
-Video Card: PCI-E 6600 128Mb (77.72 Driver) Desktop @ 1080i
-Display is a NEC 6pg Xtra Crt
-DVD Player is TheaterTek 2.2.1 using VMR9 Fullscreen Exclusive mode

Now using FFDshow-20051129 (Daily Build) as followed (only need a good sharpener for my CRT so no denoise used):
-Avisynth
Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpen_Faster.avs")
SetMTMode(2,0)
LimitedSharpen(ss_x=1.3,ss_y=1.3,Smode=4,strength=70)
-Resize @ 1280x720
Settings = Lanczos @ 2.00 / Luma Sharpen @ .30

Currently, my CPU is running about 96% (48% w/Hyperthread) and I see no stuttering. About a month ago, I was using FFDshow-20050303 and could only use LimitedSharpen(ss_x=1.2,ss_y=1.2,Smode=4,strength=45) but switched to the 20051129 build and to my surprise, got about an extra 5% speed boost to allow me to get to "ss_x=1.3,ss_y=1.3". By increasing the sampling size I was able to increase the strength without seeing any adverse affects you would get with over-sharpening (ringing, jaggies, moire etc).

Also found (for me) that using SetMTMode(2,0) instead of MT("LimitedSharpen(ss_x=1.3,ss_y=1.3,Smode=4,strength=70)", avisynthTS=true) was faster (about 3%) but others say MT() works better - the only way to know is to just try it...

jidelite

Sure it's possible...Avoid using Denoise give you some extra CPU to use SS:1.3...
Jidelite, is there a huge difference between SS:1.0 & SS:1.3...?

Sure it's possible...Avoid using Denoise give you some extra CPU to use SS:1.3...
Jidelite, is there a huge difference between SS:1.0 & SS:1.3...?

khordo,
Huge - such a relative word ;)... On my analog display it has added some noticable sharpness but I would not say huge...

jidelite

I'm trying to get ahold of the WarpSharp filter but the AviSynth Site is down, can anyone hook me up? Email is all one word: snowaddict at gmail dot com
Thanks

I am not able to get it either.....any help is appreciated. Would someone mind posting it to an alternate hosting location?

Although I don't have it, LimitedSharpen seems to be working without error. Is this expected?

Thanks,
Kevin

For those who has enough CPU juice....

Try running Resize before Avisynth.

I do a 2xlanczos2 resize before my avisynth limitedsharpen script. It takes alot more CPU power but produces a cleaner image. At least for me.... ;)

Thanks for the tip. I changed from blur and NR , then limitedsharpen @ 40, then lanczos 4 resize to 1920 X 1080 to moving the resize before the limitedsharpen. Looks a lot better. Difficult to put into words, but the picture jumps off the screen. I am able to push the ss to 1.7 with an AMD x2 4.4 without problems.

I'm trying to get ahold of the WarpSharp filter but the AviSynth Site is down, can anyone hook me up? Email is all one word: snowaddict at gmail dot com
Thanks

I just sent you an email with the Warpsharp zip file that I have in My computer.

Good Luck,

m


EDIT: It got sent back to me.... I'll try again
EDIT#2- I tried it again, and it got sent back again.
I sent it to snowaddict@gmail.com
Did I send it to the wrong place?

I am not able to get it either.....any help is appreciated. Would someone mind posting it to an alternate hosting location?

Although I don't have it, LimitedSharpen seems to be working without error. Is this expected?

Thanks,
Kevin

I just sent you a PM

m

Thanks a lot Socio for the guide and everything!

I am having some issues with getting good DVD playback. I have a brand new build:

AMD Opteron dual core running at 2820mzh each
2 gig DDR524 ram
eVGA 6600GT

Here are my settings:

multithread version of AviSynth
latest version of zoomplayerpro
Dscaler5 audio/video decoding
ffdshow with:

Blur and NR (seems to have little effect turned on or off):
Gradual Dnoise 30
Dnoise3d Luma 0.39 Chroma 1.55 timing 3.88

AviSynth:
MT("LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=4,strength=50,oversho ot=7)",avisynthTS=true)

Resize with Lanczos2
If I push the resolution beyond 1.5x, it kills the frame-rate. Shouldn't this CPU crunch filters like gangbusters?? Maybe it's my 6600gt? Oh well. Any advice to get a better picture? Thank-you!!

EDIT - Task Manager shows one core maxing out and one core idling.... hmm
Why won't it use both?

Thanks for the tip. I changed from blur and NR , then limitedsharpen @ 40, then lanczos 4 resize to 1920 X 1080 to moving the resize before the limitedsharpen. Looks a lot better. Difficult to put into words, but the picture jumps off the screen. I am able to push the ss to 1.7 with an AMD x2 4.4 without problems.

This is strange... I did this same thing yesterday, and had stuttering problems running AVISynth BEFORE resize. But when I switched the order around and run AVISynth AFTER resize (which should be even more CPU intensive) I get no stutters and I am also able to max out settings even further.

This seems wrong... I wonder what gives? Not that I"m complaining!

I tried Avi AFTER resize, and it maxes out my cpu.

I remember when I used Blur(I dont normally use blur, as i have a CRT) in front of Resize it actually lowered cpu usage before.....
I may try that again.

Im getting ringing around harse things like windows and neck ties and stuff.

hqdn3d(2)
SetMTmode(2)
MT("LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=3,strength=40,oversho ot=7).converttoyv12()",avisynthTS=true)


Any ideas on how to tone it down a bit? Im not sure what each value does.

lower overshot to 0 or leave it out (=1)

Henrik

Resize with Lanczos2
If I push the resolution beyond 1.5x, it kills the frame-rate. Shouldn't this CPU crunch filters like gangbusters?? Maybe it's my 6600gt? Oh well. Any advice to get a better picture? Thank-you!!

EDIT - Task Manager shows one core maxing out and one core idling.... hmm
Why won't it use both?


Try adding SetMTMode(2,0) or SetMTMode(2). My setup is identical to yours other then i have a 7800gt, my load is pretty even with this setup.

okay, I tried the different MT commands, and my second core is still doing NOTHING. It's sitting still and I only get 50% CPU usage no matter what I try. Here're my settings:

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

SetMTmode (2)

LimitedSharpen(ss_x=1.3,ss_y=1.3,Smode=3,strength=60)
ColorMatrix()
Lanczos4Resize(2596, 1104)

It works, but I need a little more speed to pull off that resize. I'm sure I'd be fine if only my other CPU would kick in. What's going on?

I'm using the same settings as Socio listed on post #316

Actually without the converttoyv12():
MT("hqdn3d(2)
LimitedSharpen(ss_x=1.0,ss_y=1.0,Smode=3,strength=40,oversho ot=7)",avisynthTS=true)

I follow this up with a ffdshow/lanczos4 resize X 2 and use approx 25% of 1 core and 80% of the other (athlon64 3800 X2)

I'm very happy with the results. I find the PQ much better than that from just ffdshow with the settings from the htpcnew.com tutorial. I just wish now I had a few more Mhz to tweak the settings.

What exact version of avisynth.dll are you using? ...and from where?

This is what I have in ffdshow:

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

SetMTMode(2,0)
LimitedSharpen(ss_x=1.3,ss_y=1.3,Smode=3,strength=60)

And I got an error message when I played the DVD (in Zoomplayer):

Unexpected character "y"

(ffdshow_filter_avisynth_script, line 2)


There is no such character "y" in line 2.

What did I do wrong? Please advice.

Thank you very much in advance.

What exact version of avisynth.dll are you using? ...and from where?

Followed the guide from Socio:

http://free.hostdepartment.com/f/ffdshowGuides/index.html

I just tried that Trimension Dnm/Limitedsharpen/ffdshow and........WOW
I've never seen anything like this, really gives it that 3d look. Im using WinDVd 6 v6.0 B06.056 and it works fine. Its a must see..

http://img223.imageshack.us/img223/6926/a5py.th.jpg (http://img223.imageshack.us/my.php?image=a5py.jpg)http://img223.imageshack.us/img223/4965/b9xv.th.jpg (http://img223.imageshack.us/my.php?image=b9xv.jpg)

Gents, it looks like Socio is away for a while, here are some updates


Masktools V2.0 A 15 : http://manao4.free.fr/masktools-v2.0a15.zip

This limitsupport DLL seems newer then the one Socio posted : http://people.pwf.cam.ac.uk/mg262/posts/LimitedSupport_03Dec05B.dll



Here is a special version of Warpsharp for YV12

http://www.avisynth.org/warpenterprises/files/warpsharp_25_dll_20030103.zip

However this version seems older than the one Socio posted


Here is the latest official version of Dscaler mpeg decoder

http://prdownloads.sourceforge.net/deinterlace/DScaler5007.exe?download


The k-lite mega codepack has a new version of FFdshow

-Updated ffdshow to build 2005-12-22-

http://www.codecguide.com/download_mega.htm

okay, I tried the different MT commands, and my second core is still doing NOTHING. It's sitting still and I only get 50% CPU usage no matter what I try. Here're my settings:

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

SetMTmode (2)

LimitedSharpen(ss_x=1.3,ss_y=1.3,Smode=3,strength=60)
ColorMatrix()
Lanczos4Resize(2596, 1104)

It works, but I need a little more speed to pull off that resize. I'm sure I'd be fine if only my other CPU would kick in. What's going on?


Dude your not using the mt mode. You have to use this:

MT("LimitedSharpen(ss_x=1.3,ss_y=1.3,Smode=3,strength=60).ColorM atrix().Lanczos4Resize(2596, 1104)",avisynthTS=true)

Also make sure you have the mt.dll loaded.

NeezyDeezy,
If you need alittle more CPU try using FFdshow RESIZE instead of Lanczos4Resize() within Avisynth.

jidelite

I just have these settings in ffdshow:

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

LimitedSharpen(ss_x=1.3,ss_y=1.3,Smode=3,strength=40)

I'm not calling avsfilter.vdf.avs anywhere.
After playing around with ZP a little bit I was able to get it working only after I turned off VMR renderless mode. Even then it gives me the avsfilter error twice which I press ok to and then it starts to play. Also, when I set smode to 4 it crashes ZP, smode 3 and 2 worked fine and I think 1 would too.

The only way the avisynth route would work for me is with VMR renderless mode because on Nvidia 6800 graphics cards when you go to 1080i resolutions you will get stutter even with like 50% cpu useage. BTW this is on Andy's 08 2004 ffdshow build.

Any ideas???

try to remove warsharp plugin
because warsharp has a avsfilter.vdf.avs
i don't know what does this mean
but i'm having this problem too.
but when i'm removing warsharp plugin
it's work!!

In addition to the updates tetsuo55 mentioned there is a newer version of the LimitedSharpen script that makes better use of the Limitedsuport.dll . It does require the latest limitedsupport.dll that tetsuo55 posted which I also included in the .rar

Get it here (http://rapidshare.de/files/10588567/NewLimitedSharpenn.rar.html)

try to remove warsharp plugin
because warsharp has a avsfilter.vdf.avs
i don't know what does this mean
but i'm having this problem too.
but when i'm removing warsharp plugin
it's work!!

FYI: WarpSharp is only required if you try/use Smode=2 if you so do not use Smode=2 and have no intention to try it then you can just leave Warpsharp out of the mix.

Dude your not using the mt mode. You have to use this:

MT("LimitedSharpen(ss_x=1.3,ss_y=1.3,Smode=3,strength=60).ColorM atrix().Lanczos4Resize(2596, 1104)",avisynthTS=true)

Also make sure you have the mt.dll loaded.

I get MT function changed the height! Try splitvertical = true. But using split vertical=true gives me another error.

Besides, the setMTmode command should work anyway, why do you say that it shouldn't??

I get MT function changed the height! Try splitvertical = true. But using split vertical=true gives me another error.

Besides, the setMTmode command should work anyway, why do you say that it shouldn't??

SetMTmode will work alone as long as you have the multithreaded version of the Avisynth.dll installed.

This is what I have in ffdshow:

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

SetMTMode(2,0)
LimitedSharpenFaster(ss_x=1.3,ss_y=1.3,Smode=3,strength=60)

And I got an error message when I played the DVD (in Zoomplayer):

Unexpected character "y"

(ffdshow_filter_avisynth_script, line 2)


There is no such character "y" in line 2.

What did I do wrong? Please advice.

Thank you very much in advance.

I added what I think needs to be changed to get ride of the message

I get MT function changed the height! Try splitvertical = true. But using split vertical=true gives me another error.

Besides, the setMTmode command should work anyway, why do you say that it shouldn't??


Here is me:

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

hqdn3d(2)
SetMTmode(2)
MT("LimitedSharpen(ss_x=1.5,ss_y=1.5,Smode=3,strength=30,oversho ot=5).converttoyv12()",avisynthTS=true)

I have no other goodies in ffdshow running, and i have a 60/40 split on cpu usage with avi's and dvd's on my opteron 165. it was my understanding that anything not in the MT brackets would not be smp. Hmm i should try popping my denoise in the mt command too, might get a 50/50 split.

Try it like this:

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

SetMTmode(2)
MT("hqdn3d(2)
LimitedSharpen(ss_x=1.5,ss_y=1.5,Smode=3,strength=30,oversho ot=5).converttoyv12()",avisynthTS=true)

"I'm not calling avsfilter.vdf.avs anywhere."

I keep getting the same error message; this makes Theatertek not respond. I have searched and deleted all other avs files except the LimitedSharpen one. Any suggestions?

I'm certain I'm using the multithreaded version of avisynth. I've tried several, actually, even the 2.60 beta. CPU core 1 gets to 100%, CPU core 0 stays at 1 to 2%. It just won't work. I've tried setmtmode in every configuration too... from (0) to (5, 5). When I use the MT(" command it keeps giving me a string of errors having to do with width and height changing that refuses to be resolved. I guess I'll just have to wait the 3 more days till I can post on Doom9 forums... that 5 day rule is such garbage. Were they having any real problem with spammers even???

NeezyDeezy, just a guess, you use super-sampling at "ss_x=1.3,ss_y=1.3" and LimitedSharpen will do some Resize during the sharpen process. That may cause the MT function to give the error as it's doc stated it won't work with process that change BOTH height/width.

You may try to use the default "ss_x=1.0,ss_y=1.0" to test first.

regards,

Li On

I tried it just as you said, and I get "MT: function changed the height! Try vertical split=true"

when I try verticalsplit=true, it says MT does not have a named argument verticalsplit

When I can get the HT(" command to work, the second core does work ie it is using both, but it's not worth it if I can't do anything else in avisynth.

You also need to drop that "Lanczos4Resize(2596, 1104)". Resize don't work in MT.

regards,

Li On

Yeah, that's the problem. I didn't know resize didn't work in MT. Yeah, now that I think about it that would be very hard if not impossible to program a resize in a way that has 2 separate threads. Maybe sometime in the future.

In the meantime this gives the best performance:

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

MT("LimitedSharpen(ss_x=1.3,ss_y=1.3,Smode=3,strength=60).ColorM atrix()")

Lanczos4Resize(2000, 848)

or close

CPU usage still never goes above 57% no matter what the settings... which is discouraging.

I found an update of limitedsupport.dll :
http://people.pwf.cam.ac.uk/mg262/posts/LimitedSupport_08Jan06.dll

It works with my A64...like the old limitedsupport...

I don´t get this to work! my GAM says " script error; theres no function named "MT" whats wrong?

Wiker

I can get the old limited sharpen to work fine. But when I try the the "new" one from Socio's webpage/tutorial, I get the same error.

I don´t get this to work! my GAM says " script error; theres no function named "MT" whats wrong?

Sounds like you are using a ffdshow/avisynth call for multithreaded and do not have the MT plugin installed.

I have a dual core CPU opteron 170 overclocked to 2.75 ghz so multithred is not a problem? anyway how do i get this MT plugin?

I have a dual core CPU opteron 170 overclocked to 2.75 ghz so multithred is not a problem? anyway how do i get this MT plugin?
im at 2.7ghz and i need it.

http://forum.doom9.org/showthread.php?t=94996

an even newer version of the limitsharpen dll is out

http://people.pwf.cam.ac.uk/mg262/posts/LimitedSupport_09Jan06B.dll

And also a new masktools is out


http://manao4.free.fr/masktools-v2.0a17.zip

Installed MT plugin and now it complains like this: There is no function namned removegrain.

Im starting to go crazy!

I have the SSE3 removegrain dll´s but i tried everyone with same result :(

Installed MT plugin and now it complains like this: There is no function namned removegrain.

Im starting to go crazy!

I have the SSE3 removegrain dll´s but i tried everyone with same result :(

Delete the SSE3 removegrain dll from your plugin folder it does not work correctly, use the RemovegrainS dll that seems to work for most everyone. Also the RemoveGrain SSE2 should work as well but only if your processor supports SSE2.

I have been trying to no avail to get this working.I am using the regular script .
when i apply avisynth i get there is mt edge in limited sharp line 68.What is mt edge and did i not load it .I loaded all plugins per instructions.I am resizing x 2 and i have a pentium 4 2.8 processor.

jlyon

Would these tools listed here help on a Dell XPS Gen2 laptop? The screen is a full 1920x1440 resolution natively, and I was thinking if I can sharpen up this at all (those familirar with these new Samsung screens already know how stunning the pic is already...) it would make this absolutely amazing to see.

I have been trying to no avail to get this working.I am using the regular script .
when i apply avisynth i get there is mt edge in limited sharp line 68.What is mt edge and did i not load it .I loaded all plugins per instructions.I am resizing x 2 and i have a pentium 4 2.8 processor.

jlyon

mt_edge is a function of masktools make sure you have mt_masktools.dll in your Avisynth plugin folder the newest version can be found a couple posts up.

Socio,
what do you think about using Avisynth after resize?

Hello Socio,

I've been trying the LimitedSharpenfaster() and got new dlls for Removegrain(), Warpsharpen(), MaskTools 1.5.8 but when I start zoomplayer I see a message saying Script error no function named LimitedSharpen then it says ffdshow script error line 3.

I copied the LimitedSharpenfaster as a file called LimitedSharpenfaster.avs in the avisynth plug-in directory. Is it depended on the original script, or should I have called it .avsi?

EDIT: oops I forgot to change the call to LimitedSharpenFaster instead of LimitedSharpen.

You need at least MaskTools 2.0 alpha !!!
http://forum.doom9.org/showthread.php?t=98985

Well, specifically, I didnt call the procedure right in ffdshow. but I did run into the MT_edge issue mentioned above and fixed it with the 2.0a version of MT.

Delete the SSE3 removegrain dll from your plugin folder it does not work correctly, use the RemovegrainS dll that seems to work for most everyone. Also the RemoveGrain SSE2 should work as well but only if your processor supports SSE2.

FYI. RemoveGrainSSE3.dll works fine on my Intel Prescott P4. :)

Where can I get a copy of the LimitedSharpenfaster script. My LimitedSharpen works fine but want to try the faster version as well.

socio
thanks for all the help.i took my time and went through the plugins and had the wrong dll file.I got the regular thread to work now.I am resizing to 1280x720 and my cpu usage is way up there.Do i need the multithread dvd script to resize in avisynth .I saw in one of the posts to use lanczos4 (1280,720)

What other dll files will i need if i just do limited sharpening and resize?will i have to remove the other dll files?

jlyon

A new version of the multithreading plugin and multithreading Avisynth.dll has been released, get it here (http://www.avisynth.org/tsp/MT_05.zip)

socio
thanks for all the help.i took my time and went through the plugins and had the wrong dll file.I got the regular thread to work now.I am resizing to 1280x720 and my cpu usage is way up there.Do i need the multithread dvd script to resize in avisynth .I saw in one of the posts to use lanczos4 (1280,720)

What other dll files will i need if i just do limited sharpening and resize?will i have to remove the other dll files?

jlyon

Have a look here it should answer most of your questions if not ask: ffdshow/avisynth guides (http://free.hostdepartment.com/f/ffdshowGuides/index.html)

Does anybody here both a dual core and a regular HT system that can do some comparison with MT version of avisynth/limitedsharpen?

I'm thinking about upgrading to dual core....and want to know if the $$ will be worth the performance gain.

Thanks :)

BTW, I'm getting a MT does not have a named argument "avisynthTS" error using the new MT_05 dlls.

**nevermind, fixed by removing avisynthTS from my script. Got ~10% increase in performance

Here is my script:

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

SetMTmode(2)
MT("mergeluma( removegrain(2,-1), 0.25 )
LimitedSharpenFaster(ss_x=1.0,ss_y=1.0,Smode=4,strength=70,o vershoot=3)")

Guys, there is a new mask tools 2.0 beta 18. On the limited sharpen page it says the new mask tools incorporates limited sharpen now and there is no need for limited support??? Socio do you know how to set up the new version of mask tools correctly? The new mask tools also comes with limitedsharpenfaster.dll but I didn't see the limited support file there.
Thanks

Guys, there is a new mask tools 2.0 beta 18. On the limited sharpen page it says the new mask tools incorporates limited sharpen now and there is no need for limited support??? Socio do you know how to set up the new version of mask tools correctly? The new mask tools also comes with limitedsharpenfaster.dll but I didn't see the limited support file there.
Thanks

Vidcan,
Just replace the mt_masktools.dll and limitedsharpenfaster.avs in your plugins directory with the new ones, and you can delete the limitedsupport.dll if you'd like. Since the function is included in masktools.

N3W813 I removed limitedsupport.dll and the old limitedsharpen.avs files. The new mask tools comes with a file called limitedsharpenfaster.avsi which I renamed limitedsharpen.avs because that was what it was named from Socio's website and what my calls are listed as in ffdshow. I now get an error message saying that limitedsharpen is not found in line 4.

My ffdshow settings are:
Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpen.avs")

LimitedSharpen(ss_x=1.3,ss_y=1.3,Smode=4,strength=70)
converttoyv12()

Any ideas what I'm doing wrong?

N3W813 I removed limitedsupport.dll and the old limitedsharpen.avs files. The new mask tools comes with a file called limitedsharpenfaster.avsi which I renamed limitedsharpen.avs because that was what it was named from Socio's website and what my calls are listed as in ffdshow. I now get an error message saying that limitedsharpen is not found in line 4.

It's OK, if you explicitely call import:

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


In case of using original script's extention (*.avsi), the script will be loaded automatically so you would not need import and string

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

may be safely removed from ffdshow.

The new function in mask tools is called "LimitedSharpenFaster". Instead

LimitedSharpen(ss_x=1.3,ss_y=1.3,Smode=4,strength=70)

you should use

LimitedSharpenFaster(ss_x=1.3,ss_y=1.3,Smode=4,strength=70)

in ffdshow settings.

As the last remark I would suggest to remove string

converttoyv12()

as redundant. For faster process you need to feed YV12 format to LimitedSharpen by setting input format for AviSynth filter in ffdshow. LimitedSharpen outputs the same format you feed in (at least in earlier versions, I have not taken closer look on LimitedSharpenFaster), so extra conversion call is just useless burning of CPU cycles.

Hmm that new version of masktools is very interesting

do we need new commands?

Nobody running LimitedSharpen MT has a dual core system???

:D

Nobody running LimitedSharpen MT has a dual core system???

:D

I was running Limited Sharpen MT and noticed that I have slightly visible horizontal faint lines in the middle. When i revert back to regular Limited Sharpen no faint lines that is showing, any ideas what causes this faint lines to show up using MT.

Thanks,
Jerry

When i revert back to regular Limited Sharpen no faint lines that is showing, any ideas what causes this faint lines to show up using MT.
Thanks,
Jerry

Not sure - but i saw it, pretty sure using a15 of masktools and the 06janb limitedsupport (sorry cant be more specific, i changed way to many things that night!)

Cheers,
Bitey

Is there any way to run ffdshow with LimitedSharpen, yet still take advantage of DXVA MPEG-2 decoding in my Nvidia PureVideo decoder? The problem I'm having is that the moment I insert ffdshow into the graph by assigning it to handle all supported uncompressed formats, the entire DVD playback graph becomes software-only, no DXVA is used and the peformance suffers greatly.

zambelli,

Nope. ffdshow runs in software mode.

Cheers,
Bitey

Say, I'm just curious:

When I use Avisynth in my ffdshow chain, it takes DVDs about 30 seconds to load. When I take it out, they load in seconds. This isn't the end of the world, but it isn't exactly fantastic either. IS this normal? I don't think I've seen anyone else commenting on this.

Partly, I'm wonderin whether my load times are so slow because of all the filters I have sitting in my Avisynth plugin directory. They just sort of accumulate and I can hardly remember which ones are for what, so I just leave them there ...

Or are there any other tips to improve load times?

Peace,

Mike

If I remember correctly you'll find the answer somewhere earlier in this long thread.
Something about a older/newer version of something....
Sorry, can't remember more.

Henrik

Say, I'm just curious:

When I use Avisynth in my ffdshow chain, it takes DVDs about 30 seconds to load. When I take it out, they load in seconds. This isn't the end of the world, but it isn't exactly fantastic either. IS this normal? I don't think I've seen anyone else commenting on this.

Partly, I'm wonderin whether my load times are so slow because of all the filters I have sitting in my Avisynth plugin directory. They just sort of accumulate and I can hardly remember which ones are for what, so I just leave them there ...

Or are there any other tips to improve load times?

Peace,

Mike

It takes me an extra 5 sec to load an avi, proc is peged for that 5 sec too. Dvds load alittle slower but no where near 30 sec.

I think the slowness is to do with masktools initialising.... the answer is indeed back in this thread if it really botheres you - its only a few seconds on my system for dvd's - the older versions were slower to initialise..

Cheers,
Bitey

Howdy All,

Im noticing a lot of people using calls like the following:

Setmtmode(2)

MT("LimitedSharpen(ss_x=1.5,ss_y=1.5,Smode=3,strength=30",avisynthTS=true)

Quoting from Socio's site:

There are two good ways to use this in your call, the first just utilizes the multithreading Avisynth.dll and you would use it like this keeping the "SetMTMode(2)" above any other call functions :


and


The other way which utilizes both the plugin and the dll and is my preferred method would be used like this keeping the MT(" portion infront of other call functions and keeping the ",avisynthTS=true) at the end of all the call functions :


So when using the avisynthts=true at the end, does the setmtmode(x,x) actually do anything, or any harm???

Cheers,
Bitey

I tried the new MaskTool 2a18 & LimitedSharpenFaster.avsi (autoloading now from plugin - ie I no longer Import the LSF script) and things just keep getting better. Reading my taskmgr, it looks like I'm getting about 5% more CPU than I did with Masktool 2a17 on my PC (3.0Gig w/HT on). I'm happy to report I do not see any visible horizontal faint lines in the middle so I'm not sure why it's happening to others; maybe its caused by diff versions of FFdshow (I'm using Daily build 20051129 from Sourceforge), a combo of filters used (I only RESIZE 2x then use Avisynth w/LimitedSharpenfaster) or MT() itself (try using SetMTMode() instead of MT()).

Mr Bitey,
I've experimented with MT() & SetMTMode() and for some reason the SetMTMode() parm give me better CPU utilization on my machine - others have reported just the opposite, of course YMMV. I did remember reading the developer of MT saying since MT does it's own multithreading it wouldn't take advantage of SetMTMode(). Everytime a new version of these things come out, I usually check the CPU % to see which is better. By the way, with MT ver05, I did find I had to remove the 'avisynthTS=true' parm as I believe N3W813 reported.

jidelite

OK, I decided to try updating to a newer MaskTools set and the new MT - partly in hope of speeding up load times a little, mostly just to keep up with things. Since I updated to the v.2a18 I've started getting an error, however:

There is no function named "mt_average."

It seems to crop up in 76 or 77 of the new LimitedSharpenFaster.avs.

I'm guessing that I haven't rounded up a filter or two that I need - or else that I need a different version of some limitedsupport (though I thought I didn't need this any more).

Does anyone have an idea how to get this sorted out again?

Mike

OK, I decided to try updating to a newer MaskTools set and the new MT - partly in hope of speeding up load times a little, mostly just to keep up with things. Since I updated to the v.2a18 I've started getting an error, however:

There is no function named "mt_average."

It seems to crop up in 76 or 77 of the new LimitedSharpenFaster.avs.

I'm guessing that I haven't rounded up a filter or two that I need - or else that I need a different version of some limitedsupport (though I thought I didn't need this any more).

Does anyone have an idea how to get this sorted out again?

Mike

Mike,
Check Here (http://forum.doom9.org/showthread.php?p=771568#post771568) ...

EDIT..I see you did already ;)

jidelite

Masktools has been updated again

http://manao4.free.fr/masktools-v2.0a21.zip

Changes from alpha 18 to alpha 19 :
* code refactoring
* fixed : bug with asm and width lower than 64
* fixed : doesn't prepend (0, 0) pixel to the forms helpers
* added : "min/max" mode to mt_edge. The edge value is local max - local min ( taken on a
3x3 square ).
* added : mt_lutf : a frame lut, see the description above
* added : mt_luts : a spatial lut, see the description above

Changed from alpha 19 to alpha 20 :
* fixed : huge bug preventing most filters from working

Changed from alpha 20 to alpha 21 :
* fixed : two & three input clips filters where requesting wrong frames
leading to ghost artefacts.

Is the ffdshow guide site down.i tryed to go there today and keeps coming up not available.


jlyon

I got the same thing. Timed out.

working now.

jlyon

With the latest mask tools, except a18 which works fine, the image is really blurred. I tried mask tools a21 and a22 and both blur the image very noticeably. There is some kind of bug there.

I think I had the same problem, but thought it was due to moving from LimitedSharpen to LimitedSharpenFaster. I updated Mask Tools at the same time, so that might have been the real cause of my problem.

well i did some upgrading. Ant sure if there is a new limitedsharpen dll or not, but i did go with limited sharpen faster. I cant run mask tools 2.0.21 or 22, i have to stick to 18, else its an instant death when i play anything. How ever my upgrade did get rid of the ringing ( looks like the contrast was set way to high) that my previous setup did.

I can verify that MaskTools 21 and 22 does not work, only 18. I think it has to do with the MT version of avisynth.dll that we use. I registered at doom9 but cannot post for 5 days. Can someone please post this information in the LimitedSharpen or the MT thread?

Thanks

To answer my own question......here are my results comparing a P4 3.0GHz HT Prescott and a Pentium D 2.8GHz.

Hardware
1.0GHz DDR2 4-4-4-12
Nvidia 7800 GT 256MB

Software
Windows MCE 2005
Nvidia Forceware [81.98]
ZoomPlayer [4.51]
Nvidia PureVideo DVD Decoder [188]
Avisynth [2.5.6 MT version]
MT.dll [5.0]
RemoveGrainSSE3.dll [pre1.0]
MT_Masktools.dll [2.0a18]
LimitedSharpenFaster.avs [from www.avisynth.org/LimitedSharpen]

Avisyth Script
Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpenFaster.avs")
SetMTmode(2)
MT("mergeluma( removegrain(2,-1), 0.25 )
LimitedSharpenFaster(ss_x=1.0,ss_y=1.0,Smode=4,strength=70,o vershoot=2)")

**Contrary to my previous posts, I cannot go higher than 1.0 for ss_x and ss_y without getting dropped frames with the P4-3.0 HT. Even though task manager shows usage below 60%, one side of the cpu graphs is always maxed.**

Ffdshow
Avisynth [Add ffdshow video source, YV12]
Resize [2x, Lanczos2, no luma or chroma sharpen]

Results
With P4 3.0GHz HT
38% (720x480, 30fps, 1.78:1)

With Dual Core 2.8GHz
30% (720x480, 30fps, 1.78:1)

Here comes the real performance increase...with no dropped frames
ss_x and ss_y set at 1.4
67% (720x480, 30fps, 1.78:1) **CPU graph shows 1 at 50% and 1 at 90%**
56% (720x480, 24fps, 1.78:1)
52% (720x480, 24fps, 2.35:1)

Removing resize in ffdshow and increasing ss_x and ss_y to 1.7
73% (720x480, 30fps, 1.78:1) **CPU graph shows 1 at 70% and 1 at 80%**


It looks like resize, that is not multi-threaded, takes up too many cycles on one of the cpu cores. If someone can develop a resize filter for avisynth that can be used in MT, or better yet, a MT version of ffdshow, then we can really use limitedsharpen to its full potential.

Hope this will help the people that are on the fence deciding to jump to the dual core camp. :D

P.S. May test PAL (720x576) sources in the near future, will post results if I do. :)

Now if i could only get rid of my "dual core line" id be set. Seems to happen even if i go from 11/28 drivers to 1/22 ffdshow drivers, i cant get rid of the bugger.

N3W813,
Just out of curiousity, what kind of CPU % do you get with your P4 3.0GHz HT if you did:
Ffdshow -
Resize [2x, Lanczos2, no luma or chroma sharpen]
Avisynth [Add ffdshow video source, YV12]

jidelite

With the latest mask tools, except a18 which works fine, the image is really blurred. I tried mask tools a21 and a22 and both blur the image very noticeably. There is some kind of bug there.

Im pretty sure I saw it was fixed with a23 - give it a shot... there are some other issues with MT05 avisynth.dll and any version past a18 of masktools..

Cheers,
Bitey

N3W813,
Just out of curiousity, what kind of CPU % do you get with your P4 3.0GHz HT if you did:
Ffdshow -
Resize [2x, Lanczos2, no luma or chroma sharpen]
Avisynth [Add ffdshow video source, YV12]

jidelite


I don't quite remember....I think it was ~60% with no dropped frames. :o I didn't really like how the image looked if avisynth ran after resize.

FYI for everybody...All of this is subjective....my display (GWIII) is probably different from yours. I like my image a little over sharpened (more of a HD look). Some people like it sharpened a little to keep that soft film look. There isn't one configuration for everybody, adjust and test settings to your liking.

Hi Socio, the guide you posted on the website doesn't seem to work now :-(

Is there any other way to find the guides written by you on this subject?

Thanks.

Well I tried the WinDVD DNM tweak on Zoomplayer. That is the freakiest thing I have ever seen! It looks 3D. I don't know if that is a good thing or not yet, but I was getting some pixelation around the edges during movement. Anyone else that has tried this seen that as well? If you do not have the pixelation with the DNM tweak, how about posting your versions and settings.

Thanks!

Hi Socio, the guide you posted on the website doesn't seem to work now :-(

Is there any other way to find the guides written by you on this subject?

Thanks.

WOW!

They deleted my site and I can no longer log on to my account and it won't allow me to sign back up! :confused:

There is an outdated mirror on www.hardwarexl.com

Socio, i suggest you contact the hardwarexl admin, he will give you full edit rights to your guide's mirror

another new masktools release

http://manao4.free.fr/masktools-v2.0a24.zip

this one should have a fix for the problems people have been reporting, also this version should be a little bit faster

fingers crossed

changelog:

Changed from alpha 21 to alpha 22 :
* added : "med"/"median" mode to luts/lutf
* changed : luts doesn't necessarily consider the center pixel
* changed back : forms helpers prepends (0, 0).
* changed : forms helpers now have a bool "zero" parameter, defaulted to true
* added : bool "luma" parameter to mt_merge, which makes it use the luma mask for
all three planes, and which forces chroma modes to "process" ( u=v=3 )

Changed from alpha 22 to alpha 23 :
* fix & speed up : median mode, thanks to tsp's insightfull remark. Note to self : think
less like a mathematician, and more like a programmer. Simpler, faster & not bugged.

Changed from alpha 23 to alpha 24 :
* fixed : issues with MT.dll ( thanks tsp, Boulder, vanessam and all those who suffered the bug )
* fixed : check for YV12 colorspace, and report an error if it isn't.
* speed up : median mode for luts ( once again, thanks to tsp )

I get really bad rainbowing on flesh tones. Anyone know an easy fix or some masktools settings that may work? And how I might go about implementing it in Avisynth? Thanks a lot, everything else looks great!

another new masktools release

I get a wide black line through the centre of the screen and the area above the line is scrambled and breaking up, its back to a18 for me although I havn't tried a23 yet.

release 24 works perfectly for me, on all media types, but i dont use dualcore/multithread

New 26 build fix the problem with MT Cpu´s!!

Volley,
I used the limitedsharpen (not faster) and very satisfied with the result, using zoomplayer, DVD motion is smooth on a intel 3.3GHz (HT enabled with MT) with resize to 1360x768. Tried the WinDVD DNM tweak on Zoomplayer, the picture is more even more 3D, but the motion becomes not smooth, very obovious during motion an dlooks like the frame per sec is not able to match the 24. Strangely enough, when turned on the ffdshow OSD, the frame per per sec looks above 24 most of the time (sometimes 30+) and cpu most of the time < 50% but once a while 100%. I like the DNM picture and but the motion impact makes the film unwatchable. Any one has idea too? ffdshow on 2006Dec version.

Just a side note, Socio's site had said that you create a regkey of zplayer under videodec and then useDNM=1. This did not work for me and I just got the normal WinDVD codec. I found another guide that said to put the useDNM in the videodec and set it to 1. That did work. Like I said, very freaky effect. But even with ffdshow off on my dual core 3.2GHz I get stuttering. The pasteurization change helped, but I still have stuttering and some odd artifacts around edges. Like I said, I am still not sure I like the effect. It is kind of like looking at the film through your kids viewmaster! Very 3d, but almost disturbingly so. Try out Mr. and Mrs. Smith with it. Will probably go back to denoise, resize, then sharpen on NVIDIA codecs.

thanks Volley. the UseDMN seems work for me by updating the zplayer section, and I got a 3d picture but stuttering. This is strange as cpu% and frame rate seem to be fine, just stuttering is obvious. I give up (you have 3.2 dual core) and turn it back to the my original windvd 7 codecs.

is it necessary to insert some keys in the system registry to make the last versions of avisynth / dscaler / ffdshow / windvd to work correctly in zplayer ? , cause i instaled the last version of each one (with masktools v2.0a26, warpsharp, removegrain) and the image is very bad ... help

Is a Pentium 630 at 3.0GHz too lowly to process LimitedSharpenFaster at x=1.5 and y=1.5? I tried the settings and my laptop was at 100% with a DVD. Without 1.5, at 1.0 I see jaggys.

is it necessary to insert some keys in the system registry to make the last versions of avisynth / dscaler / ffdshow / windvd to work correctly in zplayer ? , cause i instaled the last version of each one (with masktools v2.0a26, warpsharp, removegrain) and the image is very bad ... help

No - it shouldn't be necessary.

What are your system specs and what do your ffdshow and avisynth settings look like?

In my experience, the whole thing is just VERY finicky. It tends to take a long time and a lot of grief before everything works correctly together. Keep on testing - and try to explain your situation as clearly as possible. Who knows, it might even all work out ...

Mike

Aaargh.

Becuase of a number of problems (none related to this thread) I had to reformat and setup XP sp2 again from scratch.

The system is a 2.8 Northwood with HT, not overclocked, with 512 MB Ram. I was and am still using ZP 4.51 (VMR9 exclusive), the latest avisynth and MT version 5.

Before I had LimitedSharpenFaster (the "18a" version) running with these settings:

SetMTmode(2)
MT("hqdn3d(2)
LimitedSharpenFaster(ss_x=1,ss_y=1,Smode=4,strength=40)")

This was followed by a Lanczos resize in ffdshow - 2 tap, to 1280/720. I had high CPU usage - maybe 65% overall, - one side nearly topped, the other about halfway. The picture was terrific.

Now that I've rebuilt, I'm getting a weird phenomenon. Either I can do a ffdshow resize OR I can run a limitedsharpenfaster + denoise combo - but as soon as I try to use both the CPU bonks. No matter how much I tweak the resize - including a drop to lanczos 1 or even bicubic - the CPU hits the roof. Even with no denoise at all, I still can't make it work. Putting the resize in the avisynth box instead of doing int through ffdshow makes no difference.

I've checked my versions of MT and MaskTools till I was cross-eyed, I've tried several recent version of ffdshow (settling on Milan Vutka's Nov 3, 2005 version) - I just can't solve it.

But here's the Weirdest Thing of All: every once and a while - usually after extensive testing - the thing just seems to settle in and work fine - but, at the very latest, as soon as I reboot the computer, it starts all over again.*

Can anyone give me a hint about how to troubleshoot here? I've prett much run into a wall.

P.S. BREAKING NEWS: FOR EXAMPLE, RIGHT NOW, THE *#% THING HAS SUDDENLY STARTED WORKING AGAIN. I was testing the situation as I wrote this post and getting the exact problems I described above. Just before hitting send, I added the line "hqdn3d(2)" to my avisynth box and tried again. All of a sudden, CPU usage in Task Manager dropped back a bit and ffdshow reported an FPS up around 31 - up from about 12 before.

So, I think, is it an issue of denoising before I sharpen? Seems plausible. I remove the denoise to check. Nope: CPU usage drops even further and fps is even more stable. I try PAL and NTSC material. I try SageTV and Zoom Player. It all works now. But WHY?!

What on earth is going on here??

Peace,

Mike

Maybe hqdn3d(2) is doing a color conversion that magically makes all the other filters work more efficiently? (Just a guess).

Maybe hqdn3d(2) is doing a color conversion that magically makes all the other filters work more efficiently? (Just a guess).

I dunno. I don't think so, because the whole thing is working nicely this morning without it ... (I am aware that hqdn3d is a YV12 filter - so maybe something weird did happen there.)

I'll keep fiddling and updating - if I can isolate what is happening, this may be of help to some other poor saps out there ...

BTW, I'm using MaskTools a26 now. Before I reformatted Iwas getting that horizontal streak; now it's gone. Why do computers have to be as incomprehensible as everything else in life??

Mike

Is a Pentium 630 at 3.0GHz too lowly to process LimitedSharpenFaster at x=1.5 and y=1.5? I tried the settings and my laptop was at 100% with a DVD. Without 1.5, at 1.0 I see jaggys.

I have a 530J @ 3.0GHz and LimitedSharpenFaster at x=1.5 and y=1.5 is definitely to much. Did you try x=1.2 and y=1.2 (or x=1.3 and y=1.3) and what is your STRENGTH= (the higher the more artifacts you may see)?
I have actually gone back to LimitedSharpenFaster at x=1.0 and y=1.0, Strength=70 (RESIZE @ 2x/Lanczos 2 before AVISynth) and I don't really see any jaggies. Of course YMMV...

jidelite

mikesusangray,

Im with you on the 'fiddly' nature of this beast :-)

Im running a P4 3.0 at ~3.5 and in order to have smooth pans everytime im running really low settings:

hqnd3d(1), no supersampling, smode4 and shapening 30.. I then lan2 or lan4 to 1776x1000.. I get very very smooth motion all the time using this method.. If I up hqdn3d to 2, or increase sharpening to even 40 then i get occasional stutters that are repeatable on specific scenes - ie. it stutters in the same place in the same scene but is smooth most of the time..

This is using mt5 and masktools a26.. I thaught I ran a higher setting using a18 and mt4 and a lower overclock - but I might be mistaken (ive been tuning my overclock so its stable for encoding too)...

Cheers,
Bitey

I have a 530J @ 3.0GHz and LimitedSharpenFaster at x=1.5 and y=1.5 is definitely to much. Did you try x=1.2 and y=1.2 (or x=1.3 and y=1.3) and what is your STRENGTH= (the higher the more artifacts you may see)?
I have actually gone back to LimitedSharpenFaster at x=1.0 and y=1.0, Strength=70 (RESIZE @ 2x/Lanczos 2 before AVISynth) and I don't really see any jaggies. Of course YMMV...

jidelite

I was trying to go for eliminating the jaggies with 1.0 and strength of 100.
The reason I did this was because I recently popped a DVD into my RP82 connected through component to watch Temple of Doom. Granted the DVD is old and the video wasn't that great, it was very sharp on my RP-82 compared to LimitedSharpenFaster(). I can only run the S versions of the RemoveGrain().

So I decided to up the strength but then I saw jaggies and tried 1.5 supersampling. It brought the P4 down to it's knees. Forget about playing DVDs, it was having a hard time with an AVI file.

So I wondered what improvement was there between the original LimitedSharpen and LimitedSharpenFaster. It seems to me LimitedSharpenFaster lost it's edge, or is the edge only found in SSE2/SSE3 P4's with dualcore?

Seems like you need a pretty hefty machine to do 1.5 or 2.0 SS, Lancos4Resize, etc with the MT version of masktools running on a 3.4GHz dual core system with a high end video card like a 6600GT+...sheesh.

I was playing with Lancos4Resize and read the thread on it here on AVSforum linked from doom9. I was surprised to find a cleaner video by itself compared to LimitedSharpenFaster by itself.

I cannot run both as my system bombs. I know that would get the best output.

SpeedyHTPC,

as fas as I know....

You'd need supersampling to remove/reduce jaggies - I cant see how sharpening alone would help - it'd only make it worse... (on a side note, I find mergeluma to highlight jaggies whereas hqdn3d does not)..

Limitedsharpenfaster just has some of the code written into a DLL file rather than within the script - DLL's run faster

You should see a speed improovement with LSF compared to straight LS - without dual-core/hyperthreading etc..

The advantage of dual-core etc comes from running LSF with MT and or the multithread version of avisynth..

You can get some small speed improovements running the SSE2 or SSE3 optimized versions of some of the functions (eg removegrain)..


Seems like you need a pretty hefty machine to do 1.5 or 2.0 SS, Lancos4Resize, etc with the MT version of masktools running on a 3.4GHz dual core system with a high end video card like a 6600GT+...sheesh.


SS yes - but resize (lanczos4) is a single threaded function - you get no benefit running dualcore...


I was playing with Lancos4Resize and read the thread on it here on AVSforum linked from doom9. I was surprised to find a cleaner video by itself compared to LimitedSharpenFaster by itself.


I used to run lan4resize (1776x1000) with luma sharpen 1.20 (and denoise3d fast) - but I find LSF, then lan4resize (1776x1000) to produce a nicer image - albeit with only a small sharpen (30) and no SS...


I cannot run both as my system bombs. I know that would get the best output.

I would imagine a LSF with SS1.7 and Sharpening around 60-80 then lan4resize to 1920x1080 would be quite nice indeed :-)

Cheers,
Bitey

I would imagine a LSF with SS1.7 and Sharpening around 60-80 then lan4resize to 1920x1080 would be quite nice indeed :-)

Cheers,
Bitey

This is what I dont understand..

I would think a mere SS of 1.3, Lancoz4Resize of just 1024,576 to match my BenQ 7800 PJ and LSF would do..but no..it runs my P4 CPU at 80% already with either Lancoz4Resize or LSF alone..

I guess my P4 is too old even though its a 3.0GHz version. The only version of RemoveGrain I can use are the S versions. I cannot run SSE2 or SSE3 on this P4.

BTW I cannot Lancoz4Resize to anything like 1920,1080..it would be at 100% CPU.

AviSynth 2.56a
ffdshow latest build
dscaler 5
Zoomplayer Pro 4.51
LimitedSharpenFaster
Lanczos4Resize
mt_masktools 2.0a26

Using the S versions of RemoveGrain. This is running on my Pentium 630 3.0GHz HT.

Same thing happened on my HTPC which is an older P4 at 3.0GHz.

Does anyone know why Zoomplayer bombs out just before it starts playing the video? Doesnt play until I remove either LimitedSharpenFaster or Lanczos4Resize .. cannot have both.

Is Smode=4 generally considered to be better than Smode=3? I honestly can't see a big difference, but perhaps I'm looking at the wrong details. How does the performance measure up?

Is Smode=4 generally considered to be better than Smode=3? I honestly can't see a big difference, but perhaps I'm looking at the wrong details. How does the performance measure up?

I believe it's the other way around - Smode 3 ought to give slightly higher quality than Smode 4; Smode 4 should be a bit faster. I can't tell the difference either way.

Mike

Does anyone know why Zoomplayer bombs out just before it starts playing the video? Doesnt play until I remove either LimitedSharpenFaster or Lanczos4Resize .. cannot have both.

Does anyone know why Zoomplayer bombs out just before it starts playing the video? Doesn't play until I remove either LimitedSharpenFaster or Lanczos4Resize .. cannot have both.

Hmm. Wha are your system specs and what kind of settings have you got? Lanczos4 putts some pretty heavy stress on a system - will it work with Lanczos2?

Anyway, I have had some similar problems - in fact, I couldn't get any ffdshow or avisynth resize at all. For some reason, the problem seemed to disappear after a fair amount of twiddling, but I never did find out exactly which twiddle did the job.

I finally got my problem banished by (1) tightening my RAM timings (for 3,4,4,8 to 2,3,3,7), (2) enabling dual channel, and (3) adding a modest OC (from 2.8 to 3.05). The first two improvements were actually areas where my bios wasn't taking advantage of what my system should have been doing all along; the third was an experiment that has worked out very well: temperature increase has been low enough that I can still let my system run in silent mode.

These moves have opened up enough extra power to get ovre the hump. My CPU levels still tend to go extra high for a minute or two at the beginning then drop back - by I have just enough juice to push through.

Best,

Mike

SpeedyHTPC,

Overclock it!!

All,

If you have a copy of oceans 12 (PAL R4 in my case), the openining scene is what im using to test for stutters - ive found this to be one of the most demanding.

The area where stutter occurs on my system is:
There is a single mild stutter immediatly as the camera zooms out back from that glass-bowl, then major stuttering when brad pitt walks to the right (on screen) behind a pillar then towards the camera, around the pillar and walks left (on screen) the camera pans to follow him, and its just as he walks around that pillar is where the major stutter is.. Another major stutter follows a couple of seconds later as he walks into the next part of that room (walking left on screen) there is a tall-light (with white shade) in the foreground I think as the camera pans left to follow brad, there is a stutter just prior to that light going off-screen on the right.

Ive really had to wind back my settings to get that scene smooth... I would be very interested to know if anyone else has the same issues. If i get that scene smooth then there are no more stutters in that title, or other titles I have viewed..

Cheers.
Bitey

well I am using a laptop..a Dell 5160 laptop with an older Pentium 4 mobile at 3.06GHz.

I tried the same settings on my Dell 8400 desktop with a Pentium 630 HT at 3.0GHz and it bombs out and does the same thing..

Its like a dll issue..it quits right away and wont even attempt to play.

SpeedyHTPC,

My bad - I didnt realise it was a laptop (I didnt read properly :-)

Ive had ZP complain like that (just quit) when I put ffdshow denoise3d denoising prior to avisynth - I then added levels prior to it and that worked (but was too slow).. Levels (just adjust by 1 somewhere) converts the colourspace - perhaps it was this sort of problem..

It didnt really make sense why it worked with levels and not if it wasnt there - as dscaler5007 was set to use YV12...

Cheers,
Bitey

Well..im quit disappointed in my findings..

Zoomplayer is not a hog..appears to be still a problem with using Lanczos4Resize and LimitedSharpenFaster..just can't do them both.

I'd like to hear from those that are using Lanczos4Resize and LimitedSharpenFaster together. could you tell me your hardware?

I am using an AMD 64 Opteron 146 clocked at 2.8ghz and am able to use Denoise3d, Lanczos4resize and LimitedSharpenFaster in that order, with smooth, stutter free playback in Zoomplayer.

Doesn't using LanczosResize defeat the whole purpose of LimitedSharpen? AFAIK, Lanczos algorithm does some sharpening of its own.

Isn't it better to use the dest_x and dest_y parameters to resize while using LimitedSharpen?

metallic,

Would you mind posting your settings for each filter/resizer/denoiser/decoder ? along with which version of ffdshow your using?

SpeedyHTPC,
Im running a P4 (northwood) 3.0 HT overclocked to ~3.5 (3.49) (memory at 466mhz, fsb 233), radeon9800pro in VMR9 renderless mode, dscaler5007, ac3filter 0.70b, reclock 1.4. hqdn3d(1), LSF (ss1.0, overshoot 7, sharpening 30) then Lan2 resize 1776x1000 and is consistently smooth.

Cheers,
Bitey

Please post your scripts? I'm running everything the latest..but so far the HW mentioned above seems to be able to run SSE2 at least. Mine cannot load those DLL for RemoveGrain. Only S versions.

metallic,

Would you mind posting your settings for each filter/resizer/denoiser/decoder ? along with which version of ffdshow your using?

SpeedyHTPC,
Im running a P4 (northwood) 3.0 HT overclocked to ~3.5 (3.49) (memory at 466mhz, fsb 233), radeon9800pro in VMR9 renderless mode, dscaler5007, ac3filter 0.70b, reclock 1.4. hqdn3d(1), LSF (ss1.0, overshoot 7, sharpening 30) then Lan2 resize 1776x1000 and is consistently smooth.

Cheers,
Bitey

Thanks Mr Bitey..

Of the HW you spec'ed I'm not too far behind. No matter what I set for LimitedSharpenFaster, SS at 1.0 or what, Zoomplayer bombs.

zambelli,

Yes thats true. Lanczos4Resize does do a bit of sharpening and Lanczos4 does it best. Check the thread on the comparisons of screenshots. I'm almost willing to do just Lanczos4 by itself to 2* the size of my PJ's native output. It is very sharp.

http://www.avsforum.com/avs-vb/showthread.php?s=&postid=4760581#post4760581

Check Charles Black's comparisons of the various modes of LanczosResize with the pics on post #64, page 3 of that thread. Best way is to load it with Firefox in tabs. Lanzcos4Resize with LimitedSharpenFaster would give you a nice 3D look.

I am using an AMD 64 Opteron 146 clocked at 2.8ghz and am able to use Denoise3d, Lanczos4resize and LimitedSharpenFaster in that order, with smooth, stutter free playback in Zoomplayer.

Im pretty sure its safe to say your AMD 146 at 2.8GHz can do more work than my P4 mobile http://www.intel.com/products/processor/mobilepentium4/index.htm .

Oooh..shudder..P4 mobile only has 512KB of cache. I guess HTPCs now need to be equipped with an Opteron/Xeon..

Aiieee... Ok, is there a point I should start reading this thread at? It seems like things have changed pretty significantly ince the start, and Socio's guide is gone :(.

Thanks,
MP

More info:

dscaler5007 set to no analogue blanking (full 720x576), and YV12

ffdshow filters in order:
Avisynth (script):
SetMTmode(2)
MT("hqdn3d(1)
LimitedSharpenFaster(ss_x=1.0,ss_y=1.0,Smode=4,strength=30,o vershoot=1)
")

resize:
Lan2 at 1776x1000

Output:
YV12

using MT5.0, masktools a26 (and the limitedsharpenfaster.avsi included with a18 I think it was).. SSE2 version of removegrain

With those settings im ok.. If I up strength to 40 - I get stutters on the oceans12 opening, or if I up hqdn3d(2) I get the same.. Ive tried adding a touch of luma sharpening in ffdshow but at 0.40 it introduced stuttering...

CPU usage hovers in the high 50's low 60's (Hyperthreading).. FPS in ffdshow is too erratic - goes from low 30's to low 20's but playback is smooth..

Ive tried mergeluma or using denoise in ffdshow - but it just doesnt quite look as good as hqdn3d even at 1...

Cheers,
Bitey

Mr Bitey, thanks..
Youre able to run MT()..Im not. I dont have HT on the laptop..just a plain P4. And youre running Lanczos2..Have you tried Lanczos4 by itself to 1776x1000?

SpeedyHTPC,

Lan4 by itself (no limited sharpen) is pretty good - nice and smooth and reasonable sharpening (i used to use it with luma 1.20 sharpen)...

I'll turn Lan4 back on (in the LSF combo) and see if its stutter - free... ive gotta warm up the kit anyway for some viewing a little later..

Now this is odd... Ive just changed the settings to Lan4 resize (from lan2) (1776x1000) and im now able to up the strength to 50 and not get any stuttering on my test scene - i got some the first run, but subsequent runs have been smooth - it could have been a 'bug' some people (inc me) get when pausing/starting/ff/rr a movie.. No other changes...i'll go run some more tests and see how high I can push strength.. It must be close as I can see it hesitating slightly - not quite stuttering... but not perfectly smooth :) I could be imagining it....

Ran some more tests... Running strength 90, and upped hqdn3d to (2) and still smooth!! - I didnt think the script was running, so I added SS 1.6 - that sure put the stoppers on it.. Put it back to SS 1.0 and smooth again.. I think its time to ghost the HTPC and never touch it again!!

It seems Lan2 was my stumbling block.. Im using andy's SSE (I think its sse2 but could be wrong) optimised version of ffdshow (20040801a).. Lan4 and im now sitting pretty.. the wierd thing is, CPU usage is in the mid-high 60's now (performance monitor) wheras it was stuttering in the low 60's previously... so all I can think is that Lan4 is more optimised than lan2 (I thaught andy optimised upto tap4...)

After more testing... I upped the strength to 200 - still smooth... I then put it back to 90 and watched the rest of a movie.. I then went back to my test clip and... stuttery again - at 90... I reduced overshoot to 1 and it was pretty smooth..

Not sure whats happening here.. might reboot and see if something has turned evil on me...

zambelli,

I didnt think LSF has destx and desty.... might be better to do it in there, BUT i dont think you can resize with MT as it splits the image.... (maybe thats why we dont do it :-)

Cheers,
Bitey

Aiieee... Ok, is there a point I should start reading this thread at? It seems like things have changed pretty significantly ince the start, and Socio's guide is gone :(.

Thanks,
MP

There was a mirror to socio's site which was still alive - there are some links to it back some.. Looks like socio's site is back: http://free.hostdepartment.com/f/ffdshowGuides/adv.html

Cheers,
Bitey

Has anyone been succesful in getting MT working with a dual core Pentium? I have it working fine on a single core Pentium 3.4Ghz using hyper threading but the dual core causes the image to get screwed up, it looks like the frames are being overlayed on each other. My guess is that the frames are being processed out of order.

These are my settings

Opteron 146@2.8ghz
2Gb Ram
Geforce 6800GT clocked to Ultra
DVD Player: Zoomplayer ver5.00 preview 3
Video Codec: Dscaler 5 0.0.7 (default apart from output to yv12)
Reclock 1.7 beta 4
Video Renderer: VMR9 Renderless
Audio Decoder: Dscaler5 (output via Spdif)
Ffdshow version: Andy's ffdshow-20040801a_preview_SSE2

Fdshow filters used (in order)
1.Denoise3d 0, 0, 4
2. Resize 1360x768(Lanczos4 L.80)
3.Avisynth
Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpenFaster.avs")
colormatrix()
LimitedSharpenFaster(ss_x=1.0,ss_y=1.0,Smode=4,strength=80,o vershoot=7)
4.Output yv12

I tried doing the lanczos resize within avisynth but it resulted in major stuttering. Andys version of ffdshow is the only one that i can use without getting any stuttering. Also using the SSE2 version of Removegrain

Is it true that the Limitedsharpen script have is own denoise function (Remove grain ???)
So is it necessary to use denoise if the script denoise the image ?

Beamer,

Sounds wierd - are you using a26 version of masktools?

Some of the other versions (between a18 and a26) had a problem with frame order..

Cheers,
Bitey

I didnt think LSF has destx and desty.... might be better to do it in there, BUT i dont think you can resize with MT as it splits the image.... (maybe thats why we dont do it :-)
Yeah, dest_x and dest_y are documented and supported. You're right about not being able to use them with MT though.

I think using BicubicResize would be better than using LaczosResize with LSF because BicubicResize won't additionally sharpen the picture.

Also, if resizing is being done via BicubicResize or LanczosResize, the resize function should always be before the LSF function, otherwise it's rather useless.

Beamer,

Sounds wierd - are you using a26 version of masktools?

Some of the other versions (between a18 and a26) had a problem with frame order..

Cheers,
Bitey

Thanks for your response.

Yes, I'm using the latest masktools.dll dated 29/1/2006 but I have not tried the limitedsharpenfaster that's included in the file. I will give that a try to see if it resolves the issue.

regards
Chris

zambelli,


Also, if resizing is being done via BicubicResize or LanczosResize, the resize function should always be before the LSF function, otherwise it's rather useless.

That would be nice, if we all had superquick CPU's :-) I certainly cant resize before LSF... well, and not have it watchable anyway!

Cheers,
Bitey

Is it true that the Limitedsharpen script have is own denoise function (Remove grain ???)
So is it necessary to use denoise if the script denoise the image ?

I believe remove grain is used as part of a sharpening routine withing LSF its not used to denoise the end result so to speak.. However theres nothing stopping you using removegrain outside LSF if you prefer that denoiser..

You should run a denoiser first to avoid having LSF sharpen noise as well as detail you want to keep.. Some run medium/heavy denoising prior to LSF, then a light denoiser after resizing to reduce any noise/artifacts created by sharpening..

Cheers,
Bitey

metallic,

Thanks for sharing your settings.. Im pretty sure that your denoisnig with those settings is doing nothing... IIRC the time parameter (that youve set to 4) doesnt do anything unless you have values for luma and or chroma and with the preview version, if fast is selected - anything under 1 from luma/chroma is treated as 0... (i'd have to go read the ffdshow faq again to find it :-)

I'll give your settings a shot and see how they go on my kit..

Cheers,
Bitey

Beamer,

Sounds wierd - are you using a26 version of masktools?

Some of the other versions (between a18 and a26) had a problem with frame order..

Cheers,
Bitey


Oooop's Bitey you were right on the money. I screwed up when I copied the file to the dual processor machine. Works fine now, got a bit worried by a thin line across the centre of the screen but found removing the overshoot 7 option removed it.

Is there an advantage to using the LimitedSharpenFaster over the latest updated LimitedSharpen ?

Thanks again for your assistance,
Chris

I just tried the 02/03/2006 version of ffdshow SSE2 and it's working great. I had stuttering at anything higher than x1.15, y1.15 and this one lets me crank up to 1.3 without a hickup. Also using ZP 5 beta 3.

AMD-64 cpu.

do you have a link to the build of ffdshow you're talking about ? Thanks !

Here you go... http://www.free-codecs.com/FFDShow_download.htm

Now heres a strange thing..from joeuser2's link I can run fddshow SSE2 on my HT but I can't run RemoveGrainSSE2???

Do I need SSE2 of fddshow?

well..tried that and zoomplayer still bombs. I can see that CPU goes way high just before it dies.

Beamer,

Cool - ive found that line means your not running the right version of something (in some combination)...

Make sure youve copied the version of avisynth from MT05 into the avisynth folder and your windows/system32 folder (I think this is the culprit).. but if overshoot fixed it.... sounds a little bizzare but might indeed be the case!

Cheers
Bitey

SpeedyHTPC,

Have you tried using lanczos4 inside the avisynth script ?? - might be worth checking...

So whats your the order and settings your trying?

Cheers,
Bitey

i tried to find examples of that, can you give me one?

I'm just running LSF() then Lanczos4 after on separate lines.

I've set Lanczos4Resize at 1024,576 to match my PJ and I'm running LimitedSharpenFaster with SS at 1.0, overshoot 7, strength 90. Pretty much the basics until I can run them both.

However either one is pushing the CPU at 80-95% already by itself.

I really like to do Lanczos4Resize(2048, 1152) and LimitedSharpenFaster with SS at 2.0, strength 100.

SpeedyHTPC,

Thats it :) its just a single line eg:

Lanczos4Resize(1280,720)

The other way (and the way im doing it) is to do the resize in ffdshow and not in avisynth... give that a shot too and see if it works then.

Cheers,
Bitey

I see..well single lines work.. I had tried resizing with Avisynth but that didnt work before. I'll look again.

a27 of masktools is out - small bugfix (not a LSF related one).

Cheers,
Bitey

Bitey,

When i use the denoise3d settings it make a noticiable visual difference to the image. I far as i understood it using the fast setting disabled the Luma and chroma options, which have no effect when under 0 anyway as you said, and only the time parameter was used. It is definatley working as there is a clear denoising in the image.

cheers,
metallic

Mr Bitey,

Resizing with AviSynth and NOT within the script of Lanczos4Resize will give me a weird image. So that didnt work.

I really like to do Lanczos4Resize(2048, 1152) and LimitedSharpenFaster with SS at 2.0, strength 100.
Wow. That sounds like a really, really distorted image. I know quality is subjective, but I just can't believe that would look good. Please don't take it personally - I'm just skeptical.

I'm just running LSF() then Lanczos4 after on separate lines.
Like I mentioned before - I think you're completely canceling out any benefits of LSF by additionally sharpening the entire image with Lanczos4. Can you provide some screenshots?

zambelli,

What resizer would you recommend using?

Cheers,
Bitey

FYI...
I was doing some comparisons of Masktools a26 versus a18 last night and it seemed like the performance had actually gotten worse with a26. I purposely set LSF very high in order to peg the CPU and then monitored the average fps and dropped frames. I saw about 0.5-1.0 fps better performance with a18. This was on a single proc 2.93 GHz Intel, with SSE3 build of RemoveGrain.

It'd be nice if the Masktools devs ran some regular perf tests with every build so we could have these numbers before we install the builds.

What resizer would you recommend using?
I'll make two points:
1. I think resizing is only useful if it's done before LSF(). Otherwise you're just wasting CPU time by blowing up sharpened pixels in software. If the CPU isn't powerful enough to resize before LSF, then you might as well let the video renderer do the final resizing.

2. LSF is designed to be better than your average sharpener. From Didee's description: "LimitedSharpen() applies one out of three different sharpeners (two domain sharpeners or a windowed range sharpener) to the source, but will limit the oversharpening (either 'hard' or 'soft') IF it exceeds a defined overshoot." AFAIK, LanczosResize just sharpens the entire image equally with no consideration for oversharpening. Someone please correct me if I'm wrong. Applying LanczosResize together with LSF is then double sharpening. If that's anyone's goal, why not just tweak LSF to produce that much sharpening in the first place?

If you're going to resize before LSF, I recommedn BicubicResize or LSF's own dest_x and dest_y parameters. It'll be faster and is less likely to produce an oversharpened image. Just my 2 cents. If anyone has screenshots that prove me wrong or wants to put it up for a vote - all the better.

new version of masktools has been released

it seems to be a speed/bug fix version

http://manao4.free.fr/masktools-v2.0a27.zip

* fixed : mt_binarize asm optimizations that borked with some thresholds

I have made another discovery

in the the masktools zip file is a newer version of limitedsharpenfaster, last updated on 21-01-2006

If i understand correctly that version uses the limitsharpen.dll that is intergrated into masktools, this version is also newer the the regular DLL

this means you do not need the limitedsharpen.dll anymore

Also i managed to find some newer builds of ffdshow

http://kurosu.free.fr/ffdshow-20060127-05H49-athlon.exe

http://kurosu.free.fr/ffdshow-20060127-05H49-P4.exe

http://kurosu.free.fr/ffdshow-20060127-05H49-P3.exe

Remeber to download the correct version for your cpu

this new version should be faster/stabler/better or none of the above ;)

and a new version of dscaler
http://prdownloads.sourceforge.net/deinterlace/DScaler5008.exe?download

check this website : http://www.hardwarexl.com/index.php?option=com_content&task=view&id=1192&Itemid=41&limit=1&limitstart=3&lang=en

I think the Copy/Paste function works well at hardwareXL.com !

in the the masktools zip file is a newer version of limitedsharpenfaster, last updated on 21-01-2006
I'm pretty sure it's the same one as up on http://www.avisynth.org/LimitedSharpen. It's been in the masktools package for several versions now.

this means you do not need the limitedsharpen.dll anymore
I don't remember limitedsharpen.dll ever being a requirement. LS was originally (and still is developed as) a script. The DLL was something that was coded up as an alternative in order to speed up the operations, but it was never a requirement in any way.

Also i managed to find some newer builds of ffdshow
The reports on the new versions have been fairly mixed. The most reliable version I've found so far is the one hosted at http://www.x264.nl, and they are still on the 12-21-2005 release. Has someone done any comparisons between the older and newer ffdshow builds?

I'm seeing a curious problem when using Avisynth within ffdshow. If I do any sort of resizing in Avisynth where the output resolution is larger than the input resolution, both WMP and MCE zoom into the video so that I lose a great portion of my video image. I think what's happening is that the incorrect image size is being passed to the renderer and the renderer is calculating the wrong scaling factor in order to render the video full screen. For example, if a 640x480 video is being resized in Avisynth to 800x600, the renderer still somehow thinks the video is 640x480, so in order to fill a 1024x768 full screen it resizes by a factor of 1.6 on each axis, resulting in a picture too big (800x600 * 1.6 = 1280x960) to fit the rendering window.

Now, the question is: why is this happening? I'm seeing this behavior in WMP 10 (VMR-7) and MCE 2005 (VMR-9). Media Player Classic with VMR-9 renderless doesn't exhibit the problem, although I can see it adjust the video size every time I launch a file that plays through ffdshow. Like, using the previous example, you'd see it create a 640x480 video window first, then quickly adjust it to 800x600.

Is anyone else seeing this?

I really like to do Lanczos4Resize(2048, 1152) and LimitedSharpenFaster with SS at 2.0, strength 100.


Wow. That sounds like a really, really distorted image. I know quality is subjective, but I just can't believe that would look good. Please don't take it personally - I'm just skeptical.


Zambelli,

What do you mean?

The resolution of 2048, 1152 is just doubling the native res of my BenQ 7800 which is at 1024x576 PAL. But I'm in the NTSC North America area. I found that Lanczos4Resize to (1024,576) is good so why not double it?

The resolution of 2048, 1152 is just doubling the native res of my BenQ 7800 which is at 1024x576 PAL. But I'm in the NTSC North America area. I found that Lanczos4Resize to (1024,576) is good so why not double it?
Because, IMHO, you're just wasting CPU and oversharpening the image. You're resizing everything to 2048x1152, then running LSF with SS=2.0, right? That means that internally LSF is resizing even more to 4096x2304 (that's what the 2.0 stands for), before resizing it back down to 2048x1152 for output, only to be once again resized to 1024x576 in order to fit your projector's native res. Supersampling is useful, but I think resizing twice, especially with Lanczos, is counterproductive.

Out of the several scaling methods available in Avisynth, Lanczos4Resize sharpens the most. Don't believe me? Run this script:
video = ColorBars(320,240).ConvertToYUY2()
v1 = video.Lanczos4Resize(640,480).ColorYUV(analyze=true).Subtitl e("Lanczos4")
v2 = video.LanczosResize(640,480).ColorYUV(analyze=true).Subtitle ("Lanczos3")
v3 = video.BicubicResize(640,480).ColorYUV(analyze=true).Subtitle ("Bicubic")
v4 = video.PointResize(640,480).ColorYUV(analyze=true).Subtitle("Point")
StackVertical(StackHorizontal(v1,v2), StackHorizontal(v3, v4))
Note the sharpening around borders between red and blue color bars. Also note min and max luma and chroma values - they indicate how much each resizing method is boosting the luma and color through sharpening.

IMHO, I think you'd get optimal results through a simple:
LanczosResize(1024,576) # or maybe even BicubicResize
LimitedSharpenFaster(ss_x=1.0, ss_y=1.0, ...)

I'm pretty sure it's the same one as up on http://www.avisynth.org/LimitedSharpen. It's been in the masktools package for several versions now.


I don't remember limitedsharpen.dll ever being a requirement. LS was originally (and still is developed as) a script. The DLL was something that was coded up as an alternative in order to speed up the operations, but it was never a requirement in any way.


The reports on the new versions have been fairly mixed. The most reliable version I've found so far is the one hosted at http://www.x264.nl, and they are still on the 12-21-2005 release. Has someone done any comparisons between the older and newer ffdshow builds?

Well the DLL is there for speedups indeed, and that speedup partis now inside masktools, also the limitedsharpen in the zipfile has to be edited somehow to actually use the new location of the dll entries right?

the reason why x264 doesnt have any newer version is because Bobor(the guy who runs that site) has been unable to compile any of the new versions since that last release

the version i posted is used by the ppl who made the limitedsharpen script and thats also where i got the link, also these builds are fully optimised for the system that they where compiled for :)

Well the DLL is there for speedups indeed, and that speedup partis now inside masktools, also the limitedsharpen in the zipfile has to be edited somehow to actually use the new location of the dll entries right?
No. What are you basing that info on?

The new Masktools package (2.0a27 as of this writing) includes a LimitedSharpenFaster.avsi script, which is the same script as posted on doom9 forum and on avisynth.org/LimitedSharpen. Its time stamp might be 1-21-06, but if you actually look at the script you'll see it says the script was last modified on 10-29-05.

No. What are you basing that info on?

The new Masktools package (2.0a27 as of this writing) includes a LimitedSharpenFaster.avsi script, which is the same script as posted on doom9 forum and on avisynth.org/LimitedSharpen. Its time stamp might be 1-21-06, but if you actually look at the script you'll see it says the script was last modified on 10-29-05.


i see now, the limitsharpen website has been updated with the newest version as included in the zipfile

:)

I just started exploring the whole Avisynth/LimitedSharpen thing this weekend, and have achieved very good results.
But I am amazed at the horsepower that is required !
I am using the SSE3 dlls and multi-threading where available.

My HTPC is running a Pentium 640 ( 3.2 GHz ) with 1 Gbyte DDR2-533 MHz RAM.
My current mobo does not support overclocking so I am going to pickup an ASUS P5WD2-Premium mobo and good heatsink/fan. I am sure I can easily get stable to 3.8 GHz, maybe even 4.0 GHz.
Also, would it be beneficial to get one of the new Intel 900 series dual-core CPUs ?

Q. What is the preferred Windows operating system for absolute performance ?
I have always felt that Windows 2000 Pro was exceptionally lean & mean, compared to WinXP Pro.
Will I be missing any performance features by installing Win2K vs XP ?

OK, one last Q.
I have a nVidia 6600GT PCI-E in their right now.
I also have a 7800GTX PCI-E in another PC.
Any merit to using the 7800GTX in the HTPC over the current 6600GT ?

- Andy

Also, would it be beneficial to get one of the new Intel 900 series dual-core CPUs ?
Sure. Dual core is more powerful than a single hyperthreaded CPU.

What is the preferred Windows operating system for absolute performance ? I have always felt that Windows 2000 Pro was exceptionally lean & mean, compared to WinXP Pro. Will I be missing any performance features by installing Win2K vs XP ?
With a computer as fast as yours, any performance advantage of Win2K due to its smaller overhead will be negligible. I recommend you go with XP and here's why: Win2K's mainstream support will be retired this summer (6/30/2005), according to the Microsoft Lifecycle page (http://support.microsoft.com/lifecycle/search/). No more service packs are planned (although there is a rollup planned for later this summer) and I'm assuming hardware manufacturers will slowly phase out Win2K driver support for new hardware.
XP is a safe bet.

Any merit to using the 7800GTX in the HTPC over the current 6600GT ?
Probably not. All ffdshow post-processing is done in software, so a faster GPU won't help much and both cards are more than capable of good video playback.

AFAIK, I have the latest dll's etc.
I can get multi-threading working by using SetMTMode(2)

My CPU is a Pentium 630 is one PC and a Pentium 650 in another PC ( OC'd to 4 GHz stable ) both with HT enabled in the BIOS.

But if my script is:
Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpen.avs")
MT("LimitedSharpen(ss_x=1.3,ss_y=1.3,Smode=3,strength=50)",avisynthTS=true)

I get an error:
Scripts error: MT does not have a named argument "avisynthTS"
(ffdshow_filter_avisynth_script, line 3)

I do have the MT v0.5 downloaded and have MT.DLL in the plugins folder and the new AVISYNTH.DLL in the SYSTEM32 folder as requested.

Any ideas ? Or do I need a real dual-core CPU ?

- Andy

Andy,
With MT ver5 the avisynthTS=true parm is no longer needed...

jidelite

Thanks for the info on the proper syntax.

On the subject of scaling ...

In light of how fast the current crop of video cards are ( I have 6600GT & 7800GTX ), are they still using "crude" means of scaling ?
Is ffdshow's Lanczos scaling still preferred over anything nVidia or ATI are doing on their higher-end cards ?
( BTW, I an using the latest Theatertek )

- Andy

Andy Lammer,

You should put the new avisynth.dll into the system32 and avisynth directory (just to be sure :-)

Just remvove the following bit from your script and it will work:

,avisynthTS=true

Cheers,
Bitey

Yup, its working. - Andy

This whole Avisynth / LimitedSharpen pssibilities has me motivated to jack up the performance of my HTPC.

I am redoing my HTPC and today just bought one of the best OC'ing mobos -> an a ASUS P5WD2-Premium. I have a Pentium 660 ( 3.6 GHz stock ) and should be able to OC it stable to 4.4 GHz with a Thermaltake Big Typhoon on it, and a good power supply. For added grins, I am going to put in the 7800GTX I have on hand.

With this HTPC, I think I should be able to get LimitedSharpen running with SS settings of 2.0.
I can already get SS fine at 1.7 on my home-office PC ( 3.4 GHz OC'd to 4.0 GHz )
I think I read elsewhere that SS over 2.0 does not accomplish much more.

Q's for you guys who have experimented a lot already .....
( These are only aimed at DVD viewing as that's all I do via my HTPC )

For Strength, I have it set at 50.
Does adjusting Strength any higher do much improvement ?

Are LimitedSharpen's Overshoot or Soft parameters of any use.

What is the preferred Noise Reduction ?
GradualDenoise & DeNoise3D within the "Blur & Noise" section ?
Or HQDN3D(1) or mergeluma( removegrain(2,-1), 0.25) from within Avisynth ?

I am already doing the Lanczos scaling with multiplier of 2.0 and parameter of 4.
Any other ffdshow / avisynth / limitedsharpen goodies that I might be missing ?

- Andy

I am already doing the Lanczos scaling with multiplier of 2.0 and parameter of 4. Any other ffdshow / avisynth / limitedsharpen goodies that I might be missing ?
Please see my replies to SpeedyHTPC. If you're resizing with Lanczos4, and THEN running LimitedSharpen with ss=2.0 on top of that - you're actually resizing twice because LimitedSharpen resizes internally when ss_x or ss_y != 1.0.

Duh, now how did I not see that -> your comments are only on the previous page ! :)

- Andy

I have a Pentium 660 ( 3.6 GHz stock ) and should be able to OC it stable to 4.4 GHz with a Thermaltake Big Typhoon on it, and a good power supply.


HOT DAMN! -Thats going to be some serious grunt there (envious)... watch out for voltage droop on the asus boards - the earlier boards suffered from it and require a voltage-mod. (CPU volts drop under load - making the cpu unstable)..


For Strength, I have it set at 50.
Does adjusting Strength any higher do much improvement ?


Yep - I prefer around 80... although didnt see any super-improovements upto about 150 where it began looking bad - at 700 it was evil :)


Are LimitedSharpen's Overshoot or Soft parameters of any use.

Overshoot to taste I think... Personally I wouldnt bother with soft.


What is the preferred Noise Reduction ?
GradualDenoise & DeNoise3D within the "Blur & Noise" section ?
Or HQDN3D(1) or mergeluma( removegrain(2,-1), 0.25) from within Avisynth ?


Out of those, I prefer HQDN3D(2), I find mergeluma to highlight jaggies really badly - so much so I wont use it. If youve got grunt to spare, then run a heavy denoise first (before LSF) like HQDN3D(2))) then a light one after your resize or LSF is that runs after rezise (HQDN3D(1))... this way you should remove some of the noisey sharpening/resizing artifacts..


I am already doing the Lanczos scaling with multiplier of 2.0 and parameter of 4.
Any other ffdshow / avisynth / limitedsharpen goodies that I might be missing ?


Look into Soothe its a soother function for sharpeners - sometimes sharpened images look a little twitchy - soothe aparently helps :) (I havent used it).

If youre clocking to 4.4, you might be able to use seesaw in realtime (check it out on doom9) it looks really good, but isnt fast enough for realtime (unless youve got a stink-hot overclock!)...

Cheers,
Bitey

It is for-sure easy to OC any Pentium 600 series by 0.6 GHz - 0.8 GHz with a good heatsink/fan and stable power supply. I am not that hard-core that I would pursue a water-cooling setup or other exotic cooling solutions.

Once I get my 4.4 GHz up and running and all my software installed, I will report back on my settings and performance walls I hit.

- Andy

Since there's been a lot of talk about resizing and which order one should resize and sharpen in, I decided to get some screenshots and metrics and let everyone see for themselves.

I used a DVD I had available on my disk - the Criterion edition of "Brazil". It's a fairly grainy transfer, so I knew sharpening would be a challenge.

My script was:

#original video size: 720x396

original = last.BilinearResize(1088,592).ColorYUV(analyze=true).Subtitl e("original")

v1 = BicubicResize(1088,592).LimitedSharpenFaster(ss_x=1.0, ss_y=1.0, Smode=3, strength=60).ColorYUV(analyze=true).Subtitle("Bicubic x1.5, LSF(ss=1.0)")
v2 = LanczosResize(1088,592).LimitedSharpenFaster(ss_x=1.0, ss_y=1.0, Smode=3, strength=60).ColorYUV(analyze=true).Subtitle("Lanczos x1.5, LSF(ss=1.0)")

v3 = LimitedSharpenFaster(ss_x=1.5, ss_y=1.5, Smode=3, strength=60).BilinearResize(1088,592).ColorYUV(analyze=true) .Subtitle("LSF(ss=1.5)")
v4 = LimitedSharpenFaster(ss_x=1.5, ss_y=1.5, Smode=3, dest_x=1088, dest_y=592, strength=60).ColorYUV(analyze=true).Subtitle("LSF(ss=1.5, dest=1088x592)")

v5 = LimitedSharpenFaster(ss_x=1.5, ss_y=1.5, Smode=3, strength=60).LanczosResize(1088,592).ColorYUV(analyze=true). Subtitle("LSF(ss=1.5), Lanczos x1.5")

v6 = Lanczos4Resize(1088,592).LimitedSharpenFaster(ss_x=1.0, ss_y=1.0, Smode=3, strength=60).ColorYUV(analyze=true).Subtitle("Lanczos4 x1.5, LSF(ss=1.0)")
v7 = Lanczos4Resize(1088,592).LimitedSharpenFaster(ss_x=1.5, ss_y=1.5, Smode=3, strength=60).ColorYUV(analyze=true).Subtitle("Lanczos4 x1.5, LSF(ss=1.5)")

Interleave(original, v1, v2, v3, v4, v5, v6, v7)

Note: I used BilinearResize to simulate scalers present in most video renderers. It'd be difficult to compare these videos if they weren't all the same size, and Bilinear does virtually no sharpening so it seemed fair.

v1 and v2 represent two methods where you scale first, then LSF sharpen with ss_x and ss_y set to 1.0.

v3 and v4 represent methods where are scaling is limited to LSF or the renderer. The video is supersampled by x1.5 on each axis. In v3 its output is the same size as it came in, letting the renderer do the rest of the scaling. In v4 we're supersampling by x1.5, but outputting pretty much the same supersampled resolution.

v5 sharpens with supersampling and THEN resizes to a higher resolution.

v6 and v7 are two methods that use Lanczos4 resizing first, then use LSF to sharpen. v6 doesn't supersample, while v7 supersamples by x1.5. So technically v7 is scaling by x2.25 altogether.

All frames include a YUV analysis, with minimum and maximum luma and chroma values. These are very important because they indicate how much each method is expanding those ranges and changing the image. A little range boost is fine, but too much can hurt your image quality.

Rather than posting the screenshots, it was much easier to save the 8 frames in a sequence as HuffYUV AVI. You can download it from:

http://www.citizeninsomniac.com/video/LSF.avi

If you don't have HuffYUV codec installed, you can get it from here (http://neuron2.net/www.math.berkeley.edu/benrg/huffyuv-2.1.1.zip) . I recommend using VirtualDub to view the clip frame-by-frame.

I also measured performance of each method, using GraphEdit with clock sync disabled, maximized window, letting the VMR render as fast as it could for about 1 minute.

v1 - 41.35 fps
v2 - 39.86 fps
v3 - 34.06 fps
v4 - 29.25 fps
v5 - 26.96 fps
v6 - 37.09 fps
v7 - 16.80 fps

Now for my personal observations...

v1 and v3 appear to have the same level of sharpening. However, v3 seems to push the luma range out a little unnecessarily too far, so v1 is probably the safer bet. v1 has the best performance, so if this is the look you prefer - v1 is the way to go.

v2 sharpens more than v1 and v3, while still being pleasant to the eye. It's only a few fps slower than v1, so if you prefer this level of sharpening over v1 and don't mind the luma boost - v2 is another great option.

v4 doesn't show much difference compared to v2, yet the luma is maxed out and the perf is 25% worse. Not a good choice.

v5 is slightly softer than v4 because it actually scales 3 times - supersampling, output, additional Lanczos. Accordingly, its performance suffers at no gain. Avoid this one.

v6 is a great example of why you want to avoid Lanczos4 in scenarios like this. It's nearly 3 fps slower than v2 (Lanczos3), yet looks virtually the same. A waste of CPU.

v7 - 16.80 fps on a 2700 MHz Athlon64. Need I say more? Avoid like hell.


So, in the end it pretty much boils down to v1 or v2, although v3 is a decent competitor too. This just goes to prove what I've been saying all along - scale first, sharpen second, and try to avoid unnecessary supersampling and scaling.

One last important note I should make is that I deliberately chose 1088x592 as my final resolution because it's divisible by 16 in both directions. Some of the code in LimitedSharpen checks for this when determining whether to scale or not, so keep in mind that non-mod16 resolutions could trigger additional scaling operations. If you're scaling before LSF, just make sure you're scaling to a mod16 resolution, like this: x = Round(width/16)*16. It might slightly distort some aspect ratios, but it will probably help your performance.

I hope this helps. :)

Thanks zambelli - I'll have a look at the results when im at work (fast connection :-)

I'll have to go and see if I can scale first... last time i tried it stuttered to hell :(

Cheers,
Bitey

Ditto, my thanks zambelli for your info.
I will spend much time to evaluate and digest your findings & recommendations.

Man, I luv the Internet !
So much great info & knowledge / experience sharing - Andy

BTW: from your test samples, I like V2 as it is sharper.

Well, I just simply cannot resize first then sharpen - even with low strength (30).. it just stutters to hell :-( (bilinear, bicubic, lanczos2 or 4).. In avisynth or in ffdshow (ffdshow is heaps quicker for me BTW)..

Im running a Northwood P4 HT 3.0 at 3.49ghz, 512mb of 466mhz dual-channel memory (running at 466 1:1), 9800pro VMR9 Renderless and it just wont cut the mustard :((

I can only resize AFTER LSF.... There also seems to be a colour-space conversion to YUV2 happening - If i dont resize then i get a garbled picture unless I tell ffdshow to output YUV2 (i think its YUV2 - I had a late night :)... all my decoders etc are set to YV12... wierd heh..

Im resizing to 1776x1000, but tried 2X too - same stuttering (severe).. - PAL material 720x576

Im still half considering going back to Lanczos4 resize with luma 1.20 sharpening and denoise3d fast as this was super-smooth... (sigh)... Im very intolerant of ANY stuttering/frame drops etc - even very minor..

Is anyone able to resize before LSF with a similar hardware??

Cheers,
Bitey

Andy's version ( not me :) ) of ffdshow is available here at this excellent guide:
http://free.hostdepartment.com/f/ffdshowGuides/index.html

I have used all the software via the above link and have no issues.

- Andy

The 1st page of this thread is very old so following its instructions can yield nomn-current results.

Just follow the instructions in the guide via my link above, and follow it verbatim, as it worked for me. I use the latest ffdshow, not "Andy's" ffdshow.

PM me with your email and I will send you my avisynth\plugins folder and the required installers for ffdshow & avisynth.

- Andy

When I try to resize at any number higher than 2.0X, I Get an error in zoomplayer with quartz.dll....any ideas? I'd like to go to 3.0X, but this error is preventing me.

Thanks

Why do not not follow the previous instructions ? They work !!!
Click on this link: http://free.hostdepartment.com/f/ffdshowGuides/index.html
Then click on DVD Guide at the top and follow verbatim.

I am no ffdshow nor avisynth wiz, but I do understand it, and got mine to eeeasily work.

- Andy

Here's what's NOT necessary for LimitedSharpen (modes 3 and 4):

avsfilter.dll
DenoiseSharpen.dll
LimitedSupport_08Jan06.dll
RepairSSE2.dll
RSharpenSSE2.dll
warpsharp.dll

The latest script and list of requirements are listed on http://www.avisynth.org/LimitedSharpen.

The build of ffdshow hosted on http://www.x264.nl has been proven to work with avisynth and limitedsharpen.

Can you get limitedsharpen to work independently in a script outside of ffdshow?

The latest script and list of requirements are listed on http://www.avisynth.org/LimitedSharpen.


Jeremy, use the LimitedSharpenFaster script from the above.

regards,

Li On

You do know that you should call LimitedSharpenFaster instead of LimitedSharpen, right?

regards,

Li On

Try removing some of those unnecessary Avisynth plugins from the autoload plugin folder. There might be a conflict in namespaces that's making the script load the wrong functions. I'm particularly suspicious of the LimitedSupport*.dll.

Also, like I already said, try to get LimitedSharpenFaster to work outside of ffdshow first. Once you've got that working on its own, then move on to incrorporating it into ffdshow.

Is there any quality difference between LimitedSharpen normal vs. faster, or is it just speed?

My FFDShow OSD shows my frams jumping spaztically between 2? and 3? that I can't see what it's average is. Can't get the faster working for some reason (so many damn links to different locations and files)

SpyGuy311,

faster is just faster..

Jeremy Duncan,

Are you using the limitedsharpen.avsi that comes in masktools-v2.0a18.zip ?
Are you running the right version of avisynth?.

I'd try re-installing avisynth - other programs (that include avisynth) can change the default plugins directory in the registry (which means the plugins you think should be loaded are not being loaded). Make sure you also replace the avisynth.dll in your system32 and avisynth directory with the ones that come with MT (if your using MT) after you re-install.

In your avisynth settings in ffdshow you MUST call limitedsharpenfaster eg:
LimitedSharpenFaster(ss_x=1.0, ss_y=1.0, Smode=3, strength=60)

and NOT

LimitedSharpen(ss_x=1.0, ss_y=1.0, Smode=3, strength=60)

If you cut and paste scripts from the web - sometimes these ADD SPACES in and they come up with errors - use the one from masktools I mentioned earlier.

Cheers,
Bitey

Search (on mt_edge) is your friend and will take you to my post (and others) to this post:

http://www.avsforum.com/avs-vb/showthread.php?p=6500838&highlight=mt_edge#post6500838

regards,

Li On

After all that. I find out my notebook, celeron M 1.4 Ghz. Can't play the video very well !

Good stuff to know though. For when I get a better PC.

Thank's all.

What's the minimum recommended hardware specification needed ?

I get DenoiseSharpen.dll, RepairSSE2.dll, RemoveGrainSSE2.dll from the remove grain link.
You're a stubborn guy, Jeremy. I already told you twice that only RemoveGrain*.dll is necessary. You don't need DenoiseSharpen and Repair in order for LimitedSharpen to work.

I put all of these into system32 as well as my avisynth plug-in folder.
That's absolutely unnecessary.

Import("E:\Program Files\AviSynth 2.5\plugins\LimitedSharpenFaster.avsi")
Also unnecessary. Files with extensions .avsi automatically load when in the plugin folder.

Hi Jeremy,

Any particular reasons why your not using resize options?

Jerry

Not really.

It's funny, ffDShow version Dec 21 2005 gcc 4.0.2 (as pictured above) is the only recent version I, too, found so far that seems to run stable. I guess most problems that people report concerning non-working LimitedSharpen or bad performance stem from bad ffDShow releases and not from bad configurations.

I have ffdshow dated Feb 2 2006 11:34:36 that seems fine to me.

- Andy

I say:
Unnecessary. Files with extensions .avsi automatically load when in the plugin folder.
Jeremy then says:
Import("E:\Program Files\AviSynth 2.5\plugins\LimitedSharpenFaster.avsi")

I give up. There's no use in helping someone who refuses to listen.

Jeremy Duncan,

Unless your registry is pointing to a different folder for your plugins than where you have limitedsharpenfaster.avsi - you do not need to load it explicitly. If its not autoloading and your plugins folder is right in the registry then there is a problem with your avisynth install.

Glad to hear you finally got it working :-)

Cheers,
Bitey

Ok, let me give it a try -

Jeremy,
In the ffdshow avisynth text box only type:
LimitedSharpenFaster(ss_x=1.0, ss_y=1.0, Smode=3, strength=60)

Now let us know what happens :)

jidelite

So your test shows that limited sharpenfaster is being used without importing the .avsi.
No. The script file contains the function, but not the function call. You still have to CALL the function, which is why everyone's been telling you to leave out the "Import()" and just call "LimitedSharpenFaster()". The function is already getting loaded without the explicit "Import()" command.

Sounds to me like you're trying to get to step #3 (Using LimitedSharpenFaster) without studying steps #1 (Using Avisynth) and #2 (Using FFdshow). Please study the Avisynth documentation (http://www.avisynth.org/index.php?page=AviSynthManual) carefully.

BTW, ColorMatrix() should be called before anything else in your example.

No. The script file contains the function, but not the function call. You still have to CALL the function, which is why everyone's been telling you to leave out the "Import()" and just call "LimitedSharpenFaster()". The function is already getting loaded without the explicit "Import()" command.

Sounds to me like you're trying to get to step #3 (Using LimitedSharpenFaster) without studying steps #1 (Using Avisynth) and #2 (Using FFdshow). Please study the Avisynth documentation (http://www.avisynth.org/index.php?page=AviSynthManual) carefully.

BTW, ColorMatrix() should be called before anything else in your example.

Ok.

Thanks for the great thread and website. I got everything installed and working with the help here. However, when I try to use the Multi-threaded avisynth and MT.dll I get an access violation and Zoom Player crashes (actually ZP stays open, but unresponsive...must be avisynth that has the memory violation). Anyone else have this problem?

This is on an Athlon 3800+ X2, and Nvidia 6800 GS (PCI-E).

I reinstalled avisynth (to get the old dll) and got it working again. But would like to take advantage of both cores if possible.

Right now I am running it like this: LimitedSharpen(ss_x=1.1,ss_y=1.1,Smode=3,strength=100)

This puts me between 30% and 40% CPU. But I am not doing any resizing yet (can't get it to fill the screen, but that is another problem).

anyone getting a quartz.dll error when choosing resize options of above 2.0X??

So why all this softning?
50 in Blur & NR
100 in LimitedSharpenFaster
Would it not be easier to just leave out LimitedSharpenFaster(?),
that would give you the desired soft look...

Henrik

"Edit the Lanzcos4Resize to the size of your monitor.
Edit Picture properties, and Blur & NR to your preference."

LimitedSharpenFaster cleans up the picture.

zambelli
I am running the limited sharp script,with blur and noise reduction.I am also using lancos4 resize like in your guide.My cpu usage is around 10 to 30 percent and usually goes no higher.I still see some jerkyness .If i turn off resize it smooths out.I was wondering where to get graphedit so i could trouble shoot my settings and see what is causing the most problems.

jlyon

Zambelli

I think i have the graphedit software .can you also tell me briefly how you used it to get the frame numbers.Do you select the filter you are using if ffdshow?Under video compressors i see ffdshow ffdshow codec or encoder am i getting close?

jlyon

I would like to know if it is still necessary to add the "converttoyv12()" function in order to fix the color problem noted by Socio in post#274 when using VRM9 renderless mode.

Regards,

Bill

I think i have the graphedit software .can you also tell me briefly how you used it to get the frame numbers.Do you select the filter you are using if ffdshow?Under video compressors i see ffdshow ffdshow codec or encoder am i getting close?
Start up GraphEdit. Hit Ctrl + R to render a media file and choose the file you want to play back. If ffdshow is configured correctly, you should see it in the graph right before the Video Renderer. Disconnect and delete the entire audio graph because it's not necessary for this test.
Hit the Play button, then immediately hit Stop. Go to the Graph menu and uncheck "Use Clock". Right click on the Video Renderer and bring up its properties. Now hit Play again.

The video will play as fast it possibly can given the finite CPU resources. You can see the average framerate in the Renderer properties.

zambelli
thanks for your help.

jlyon

First off, Great job Socio....Limited Sharpen rocks...a better find than ffdshow IMHO.

I have gone through all 25 pages on this topic and have used every combination of avisynth, ffdshow and related plugins on both my laptop (amd) and htpc (p4). I can get it to work and I am happy with the results but I keep getting an error box whenever I start Zoomplayer or change any settings within ffdshow realting to avisynth "Import:couldn't open"C:\Program Files\Avisynth 2.5\plugins\avsfilter.vdf.avs" On both machines...

I saw earlier on that someone else had this issue but was never really solved.

Any suggestions would be appreciated.

Another intersting tidbit. My laptop is a Athalon AMD 64 3700 2.4 and can handle Limited Sharpen faster supersampling at 1.4 followed by Lanczos4 1280x800 resize and hits about 90% cpu usage.

My HTPC is a P4 2.4ghz and struggles with supersampling at 1.0 with Lanczos4 1280x720.

Go AMD!!!!!

After playing/experimenting & lots of watching I like these settings below.
Man, have I tried a helluvalot of combinations and settings !
My projector is 1400x1050 native resolution and my HTPC is Intel 630 OC'd at 4.09 GHz.
( I hope to soon get a dual-core 940/950 CPU and OC it too )

I only have AVISYNTH checked-boxed in ffdshow, and here is my script:

Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpenFaster.avs")
SetMTMode(2)
MT("HQDN3D(1)")
LanczosResize(1440x960)
MT("LimitedSharpenFaster ss_x=1.0,ss_y=1.0,Smode=3,strength=100,overshoot=7)")

My main test clip for this is the start of Chapter 10 in Gladiator and let it run for about 2 minutes.
I like this clip as there are lots of horizontal pans, that let me keep on eye on stutterting.

With the above settings, I am at the performance limit of my 4.09 GHz HTPC. Some notes:

1.
If I up my NR to HQDN3D(2) I can spot a hint of stuttering.
I seem to like HQDN3D more than various mergeluma/RemoveGrain settings as RG seems to consume too much CPU.
I think someone else previously commented they felt RG enhanced jaggies on diagonal lines.

2.
LanczosResize(1440,960) is 2x the DVD's native res.
If I try to scale to my PJ's native res via LanczosResize(1400,1050) I can more easily spot some stuttering.
I attribute this to the fact that 1400x1050 is not an even multiplier of DVD's 720x480, and so involves more math.
Lanczos4Resize consumes too much CPU and yields little visual benefit over juts LanczosResize, as shown previously in this thread.
I visually prefer LanczoResize over BicubicResize.

3.
For now, I am here to sharpen "a lot" so I set LSF strength at 100.
Lowering it to 70 or 80 will give more CPU overhead, If I want to do more NR such as starting with a HQDN3D(2).

I may have too much sharpening going on and will continue to evaluate ringing as I watch more DVD content.

I have learned MUCH from those "in the know" and extend my thanks to you all.

- Andy

Has anyone been able to "multithread" the laczos resize? When I try it, I get an error message about "split vertical blah blah blah."

This would really help lower my cpu usage. I can't resize to my native 1360X768 before LSF. I can supersample to 1.1 and resize after though.

It is beneficial to supersample or should I just leave more CPU headroom? It seems that with multithreading that supersampling is not nearly as much of a performance hit.

Jim

hobie_od,

You cant multithread resize.. must be done before, or after the MT call..

Andy

Sounds great! - i'd love to be able to run those settings, but my poor old northwood P4 3.49ish just cant do the business :( (sigh) I'll have to convince the wife I need an upgrade :) - I'll try running a straight 2X resize and see if I can get away with resizing prior, but I dont like my chances!

Cheers,
Bitey

LanczosResize(1440,960) is 2x the DVD's native res.
If I try to scale to my PJ's native res via LanczosResize(1400,1050) I can more easily spot some stuttering.
I attribute this to the fact that 1400x1050 is not an even multiplier of DVD's 720x480, and so involves more math.
Your settings sound very reasonable and I'm glad you found one that's efficient and pleasing to the eye. You're probably right about reasons why resizing to 1400x1050 is too stuttery. That is indeed a strange resolution. It's always best to resize to resolutions that are divisible by 16 in both directions, and if you can resize by an integer factor, that's even better. So 1440x960 is definitely the way to go, even if that requires your video card to scale down a little to fit the screen.

Well I was able to get rid of my error messages by removing all the unnecessary junk in my plugins folder....so I am quite happy about that.

Now my poor HTPC just doesn't have the CPU power to use noise reduction (HQN3D) limited sharpen faster then resize lanczos 4.

Should I drop the resize and concentrate on cleaning up with a noise reduction or vice versa?

Unfortunately my AE900 is in the shop so I can only view on my 17"LCD so it is hard to get a good idea of the final results.

scotty144,

Glad to hear your up and running.

Depends how good your scaler is in your video-card :-)

You'll really have to try both to see what works best for you.

Im in the same boat... Im denoising, lsf then resizing - but im not happy with the strength levels I can run and be 100% stutter free (down to 25 ATM)..

Good luck - let us know how it pans out!

Cheers,
Bitey

What does "Overlay Mixer" in FFDShow do ?

Are the instructions in the beginning of this thread still good? I am going to try limited sharpen with my present setup and don't want to start with the beginning instruction only to find on some later page the initial instructions need to be modified. I have a P4 2.8GHz processor and an ATI Radeon 9800 pro video card. I've been using Dscaler/FFDShow/ZP for the last year without problems. I'm just looking for some quick guidancew before going through all 25 pages of this thread.

Thanks,

Thanks, That was just what I was looking for.

I have been experimenting for some time using Avisynth and have achieved remarkable results in respect to image processing. However, I have been plagued by occasional 'stutter'. Initiially I was of the opinion that this was being caused by lack of processing power, I now know that this is not the case, it is caused by cadence processing or lack of it.

It would appear to me that as soon as ffdshow / Avisynth is introduced in to the processing loop, we lose control of cadence!! This can be proven very simply by running a 3:2 detection test without any processing (TheatreTek) and then introducing ffdshow with a simple filter, any one will do. Theatretek / Nvidia do a wonderful job of dealing with cadence issues and the 3:2 test passes 100% but fails 100% of the time when any external processing is perfomed.

The reason for this as I understand it is that the hardware deinterlace resposibility is moved off board and onto the ffdshow, accordingly we lose the excellent deinterlace that Nvidia has developed over the years.

So! are we all fooling ourselves into beleiving that we have achieving the best post processing available?

I seems to me that we need an additional module within Avisynth to deinterlace as well as Nvidia on their hardware. The deinterlace currently in Avisynth, cannot handle the 3:2 pulldown :(

Before anyone responds with: I don't have interlace issues etc, I need to make clear that I can watch a lot of movies without issue but there are also some movies that I need to switch processing off to watch even though my dual processor always has at least 20% of overhead left on each processor.

Test material for the cadence test is: Silicon Optix HQV Technology 3:2 Detection Test HQV Benchmark Version 1.4

Other cadence tests fail but this particular one is crytal clear in that it works fine without processing and fails 100% of the time when any processing is introduces as indicated earlier.

Beamer
"I have been experimenting for some time using Avisynth and have achieved remarkable results in respect to image processing. However, I have been plagued by occasional 'stutter'. Initiially I was of the opinion that this was being caused by lack of processing power, I now know that this is not the case, it is caused by cadence processing or lack of it.

It would appear to me that as soon as ffdshow / Avisynth is introduced in to the processing loop, we lose control of cadence!! This can be proven very simply by running a 3:2 detection test without any processing (TheatreTek) and then introducing ffdshow with a simple filter, any one will do. Theatretek / Nvidia do a wonderful job of dealing with cadence issues and the 3:2 test passes 100% but fails 100% of the time when any external processing is perfomed.

The reason for this as I understand it is that the hardware deinterlace resposibility is moved off board and onto the ffdshow, accordingly we lose the excellent deinterlace that Nvidia has developed over the years.

So! are we all fooling ourselves into beleiving that we have achieving the best post processing available?

I seems to me that we need an additional module within Avisynth to deinterlace as well as Nvidia on their hardware. The deinterlace currently in Avisynth, cannot handle the 3:2 pulldown :(

Before anyone responds with: I don't have interlace issues etc, I need to make clear that I can watch a lot of movies without issue but there are also some movies that I need to switch processing off to watch even though my dual processor always has at least 20% of overhead left on each processor.

Test material for the cadence test is: Silicon Optix HQV Technology 3:2 Detection Test HQV Benchmark Version 1.4

Other cadence tests fail but this particular one is crytal clear in that it works fine without processing and fails 100% of the time when any processing is introduces as indicated earlier."

So. It's the deinterlacer !

Hey, Beamer. I posted this at doom9.org (http://forum.doom9.org/showthread.php?p=798314#post798314)
There's folks over there that I'm sure can help.

Well, Beamer.

The kind folks over at doom9.org linked to this; http://www.doom9.org/ivtc-tut.htm

They said getting a deinterlacer to perform IVTC (23.976fps) won't work very well.

Maybe you can tell me if the 3:2 pulldown deinterlace problem ffdshow has, has anything to do with 23.976fps ?

Looking at it now. They assume ffdshow deinterlaces 3:2 pulldown at IVTC fps.
But I may be wrong. Have to ask them.

Im not sure if this is entirely the cause for some stutter, im in PAL land (no 3:2) and still find some obscure stuttering when using LSF on some occasions.. eg. I can play test clip and see stutter, then play it back again straight away and not see it. These observations have been posted before (by myself and others) and there doesnt appear to be an easy solution (or any solution) - the only general rule of thumb is that the more processing power you have, the less likely you are to see it..

eg. Ive just removed the overclock on my HTPC (back to 3.0ghz from ~3.5Ghz) and am able to play the same clips that I thaught were maxxed out when overclocking seem to play fine.

I did also increate the setmtmode(x) to setmtmode(5) up from setmtmode(2) - which may have had an affect - infact using those settings, with a slower CPU im able to increase my strength up 5..

This may all change tomorrow - it seems to be the nature of the beast :-)

Cheers,
Bitey

Well, Beamer.

The kind folks over at doom9.org linked to this; http://www.doom9.org/ivtc-tut.htm

They said getting a deinterlacer to perform IVTC (23.976fps) won't work very well.

Maybe you can tell me if the 3:2 pulldown deinterlace problem ffdshow has, has anything to do with 23.976fps ?

Looking at it now. They assume ffdshow deinterlaces 3:2 pulldown at IVTC fps.
But I may be wrong. Have to ask them.

The test I'm using is 23.976 Fps, after using any form of external processing the frame rate displayed by TT is 57 Fps - 59.75 Fps. The frame rate starts at 57 Fps and slowly creeps up. When run natively the frame rate is shown as 23.85 Fps and is fairly constant.

I followed the thread on Doom and have dowloaded TIVTC & Tdeintv but have not had any luck getting them to work but I have only spent a few minutes on it so I'm guessing I have probably not installed something correctly.

If WE can get a resolution to this cadence issue, I really believe we will have a fixed a problem that a lot of people percieve as stutter / judder. I'm confident that if high end scalers can resolve this issue, we should also be able to achieve it!