fft3dGPU 0.8.2 [Archive] - Page 13

AssassiNBG

20th March 2007, 18:10

Yeah, sorry, I figured I should describe my problem better.

It only happens when using vdub's job control to do a 2pass encode. I use XviD for encoding and the following script:

Import("aa.avs ")
Import("F:\DVDRips\2ndtry\masktools-v2.0a30\masktools\LimitedSharpenFaster.avsi")
Loadplugin("F:\DVDRips\2ndtry\RemoveGrain\RemoveGrainSSE3.dll")
Loadplugin("F:\DVDRips\2ndtry\MaskTools-v1.5.8\masktools.dll")
Loadplugin("F:\DVDRips\2ndtry\masktools-v2.0a30\mt_masktools.dll")
Loadplugin("F:\DVDRips\2ndtry\WarpSharp.dll")
Loadplugin("F:\DVDRips\2ndtry\FFT3dGPU.dll")
Loadplugin("F:\DVDRips\2ndtry\RemoveGrain\RepairSSE3.dll")
Loadplugin("F:\DVDRips\2ndtry\RemoveGrain\RSharpenSSE3.dll")
Loadplugin("F:\DVDRips\2ndtry\RemoveGrain\DenoiseSharpen.dll")
Loadplugin("F:\DVDRips\2ndtry\SangNom.dll")
Loadplugin("F:\DVDRips\2ndtry\eDeen.dll")

LoadPlugin("F:\DVDRips\2ndtry\Decomb.dll")

AVISource("F:\DVDRips\2ndtry\project_d2v_vfapi.avi")
ConvertToYUY2()
AssumeTFF()

Telecide(guide=0,post=2,vthresh=43,blend=true)
Decimate(5)

antialiasing()
FFT3DGPU(sigma=3, bt=1)

Crop(8,0,-6,0)
LanczosResize(640,480)
LimitedSharpenFaster()

And when I manually do the first pass, then go and choose second pass, save as avi, and it is fine.

So the crash occurs at the end of the first pass, somewhere around the beginning of the second one. The status of the second one in the job control says Aborted.

The error from the crashdump is the following:
An out-of-bounds memory access (access violation) occurred in module 'FFT3dGPU'...

...writing address 00000059.

Maybe someone could try and reproduce the error by running the script I am using on a small file just for the sake of reproducing the crash. That would really help out in seeing whether the problem is in conflicting DLLs from the script, or just a problem in my configuration. (Although I have no idea what it might be caused by, since I installed the latest versions of DirectX and the ForceWare drivers.)

tsp

20th March 2007, 21:58

AssassiNBG: ok I can reproduce it or maybe a variant of it. It looks like the script from the first pass isn't properly been shut down and that it casing problem with the creating of the script for the second pass. I will try to see if I can fix that. Else maybe Megui might be an option until then.

AssassiNBG

20th March 2007, 22:01

That is some relieving words you have man! I hope you can fully identify and manage to fix the problem.

In the meantime, I'll check that MeGUI out. :)

Thanks!

tsp

21st March 2007, 00:22

It turns up that it is masktool that are causing the problems as the clips before mt_inpand is not destroyed. This simple script causes the same error if the script is opened in virtualdub, closed and reopened without closing virtualdub:

blackness().converttoyv12()
FFT3DGPU(sigma=3, bt=1)
mt_inpand()

I use version 2.24 of masktools. I will try and see if upgrading helps

[EDIT]
Yes it seems like the bug was fixed in 2.26 so get the latest version from the masktools thread (http://forum.doom9.org/showthread.php?p=701539)

AssassiNBG

21st March 2007, 14:52

Hey there,

Did you mean v2.0a30 ? That is the latest version I found in the thread. And it's from June 2006. Also I have it loaded in my script, but perhaps it is conflicting with MaskTools 1.5.8. I'll comment the line loading masktools 1.5.8 and try again.

Thanks!

PS. I tried without masktools 1.5.8 but apparently the antialiasing script requires it. Specifically it requires the EdgeMask function from MaskTools 1.5.8. I would like to ask what the alternative for EdgeMask is in masktools 2.0+ ? Also if it might help here's the antialiasing function I got from another thread here on doom9 forums:

function antialiasing(clip orig,int "th_luma",int "th_croma",string "type",int "aath") {

# "th_luma" & "th_croma" are the edge detection thres.: lower values=more edges filtered
# "type" is the matrix used for edge detection: with "sobel" (default) only the
# hi-contrast edges, where artefacts are more noticeable, are filtered. If you want
# to test other matrices, read the MaskTools Guide for more info.
# "aath" = anti-aliasing strenght (default should be fine)

th_luma = Default(th_luma, 20)
th_croma = Default(th_croma, 20)
type = Default(type, "sobel")
aath = Default(aath, 48)
ox = orig.width
oy = orig.height
dx = orig.width * 2
dy = orig.height * 2
clp = orig.IsYV12() ? orig : orig.ConvertToYV12()

a=clp
b=clp.Lanczos4Resize(dx,dy).TurnLeft().SangNom(aa=aath).TurnRight().SangNom(aa=aath) \
.LanczosResize(ox,oy)
c=clp.EdgeMask(th_luma,th_luma,th_croma,th_croma,type)
MaskedMerge(a,b,c)

}

Please help me tweak/fix this function to work with masktools 2+. Thanks!

G_M_C

21st March 2007, 20:51

G_M_C: I don't have enough money for a geforce 8800 GTS yet so not much progress here.

AssassiNBG: I need a little more information before I can do anything about it. Like what script you are using. What codec is you using for encoding. Does it happen instantly or after a while. What are the exact error message.

I dont have one either, and dont plan on getting one. I was merely asking, and hoping for, updates since i use FFT3DGPU a lot :)

Boulder

21st March 2007, 21:51

AssassiNBG: did you try using MPEG2Source instead of AVISource (i.e. VFAPI)?

AssassiNBG

23rd March 2007, 20:58

Yeah, I did, but the problem is still present. I guess it's not because of the source. And it is also definitely not because of masktools since I was helped modifying the function so it now works with masktools 2+. But the crashing still occurs.

@tsp Which version exactly works ok for you ? Apparently the latest one doesn't fix the problem for me. :(

tsp

23rd March 2007, 21:31

AssassiNBG: fft3dgpu 0.8.2 with virtualdub 1.6.17 using a two pass job file encoded with xvid 1.2 -127. with you script. Does this script also fail?

Loadplugin("F:\DVDRips\2ndtry\FFT3dGPU.dll")

AVISource("F:\DVDRips\2ndtry\project_d2v_vfapi.avi")
ConvertToYUY2()
FFT3DGPU(sigma=3, bt=1)

if not try commenting out 1 filter at a time in our script until you find out what is causing the problem.

AssassiNBG

26th March 2007, 20:41

Hmm, I tried commenting out each filter since the script above worked fine numerous times. You're probably not going to believe it, but it was LimitedSharpenFaster that causes the crashing.

The cartoon I'm encoding right now can go without a sharpener, but I'm sure I'll need a sharpener for other encodes in the future. Does anyone know any substitute for LSF ? No matter how slow it is, I'm only concerned about quality. Or could someone try fft3dgpu along with LSF please ? Or should I turn Didee (he's written the function right ?) for help ?

Or perhaps if both the developers of LSF and fft3dgpu could team up to try identifying and fixing the problem. I'm willing to provide crashdumps of VDub for the errors, but I'm sure you'll be able to reproduce the effect yourself.

Thanks so far for the help whatever happens!

tsp

26th March 2007, 20:55

AssassiNBG

27th March 2007, 05:56

seesaw (http://forum.doom9.org/showthread.php?t=104701) might be worth trying.
I will create a version of fft3dgpu that should provide a little more usefull error message.
Does it still appears when you use blackness(length=8) instead of AVISource("F:\DVDRips\2ndtry\project_d2v_vfapi.avi")
?

I hvaen't tried with blackness=8, but I will after school later today.

SeeSaw I've tried and didn't really like, because it created jitter (is that what it's called - dots around edges) which makes the otherwise smoothed picture back to what it was (almost). But I'll have to try with a few more denoisers as I've only tried with another fft3dgpu() in the script.

All right, off to school for me.
Later.

UPDATE:
So it crashes with blackness(length=8) too.

I tried SeeSaw with a few denoisers but it crashed with each and one of them. It seems that the clip doesn't want to get sharpened. :P I tried degrainmedian, vaguedenoiser and the one that gave me the best result (no jitter) was hqdn3d. But the script crashed VDub with all of them. This is really weird. What could both LSF and SS have in common (except that they're sharpeners) in terms of processing the clip, that causes the crashing ?

swaaye

18th April 2007, 01:15

I picked up an 8800 a month ago and realized that it would be fun to give FFT3DGPU a shot now. I've used the filter with a X800GTO2 (which is basically a underclocked X850XT) previously.

So, I've found that with a 720x480 source off a DVD, I see about 60-85 fps (in VDubMod). I'd say the average is around 70fps. Running a very simple script and doing just a preview of output with no compressor:

MPEG2Source("source",idct=3)
FFT3dGPU(sigma=1.5)

I thought the speed was a bit disappointing as I think my X800GTO2 at 510/600 was more than half as fast. So, I loaded up ATITool and underclocked the 8800 GTX down to 200 MHz core. I saw a very minimal change of at most 10 fps. I'm using the NV 158.19 and tried some unofficial 101.02 drivers (they fixed some game issues I was having with the 97.92s).

Very strange...... GPU temp doesn't really go up. Stays around its idle temp (in games it'll hit 85C sometimes and the fan will spin up to keep it there). CPU usage is only about 10-20%. Core 2 Duo.

Leak

18th April 2007, 09:47

So, I've found that with a 720x480 source off a DVD, I see about 60-85 fps (in VDubMod). I'd say the average is around 70fps. Running a very simple script and doing just a preview of output with no compressor:
That would be because VirtualDub limits the preview frame rate to some value (Screen refresh rate? Dunno...). Try turning off input and output video and just look at the statistics instead.

foxyshadis

18th April 2007, 10:11

swaaye

18th April 2007, 17:56

That would be because VirtualDub limits the preview frame rate to some value (Screen refresh rate? Dunno...). Try turning off input and output video and just look at the statistics instead.

Well, I did uncheck the boxes to view the input and outputs. And yeah, when those are on, it was locked to like 30 fps or so.

swaaye

18th April 2007, 18:05

The quick way to benchmark either decoding or decoding+encoding in vdub is "run video analysis pass", which bypasses the preview. It renders to the encoder chosen, if that's uncompressed it's as if there wasn't one.

Doing this I still see ~65-70 fps. My refresh rate is 60 Hz (LCD).If I just run the test with MPEG2Source("source",idct=3), I see over 300 fps.

In practice, the speed isn't a real issue though. I can run:

MPEG2Source("source",idct=3)
tfm(d2v="source")
tdecimate()
Crop(2,0,-2,-0)
LanczosResize(640,464)

at the same speed with or without FFT3DGPU and with Divx at quality preset 5.

Oh, and I tried full 32-bit mode (precision=2) and didn't see a major speed loss. It was only a few fps.

tsp

18th April 2007, 22:13

swaaye

18th April 2007, 23:53

how does it compare to darknite's result here (http://forum.doom9.org/showthread.php?p=913875#post913875)?
I suspect that there are a bottleneck somewhere in the driver maybe with upload or download to/from the gpu as it is obvious nether cpu or gpu limited.

using:

ColorBars (width=1284,height=1024,pixel_type="YV12")
fft3dgpu(bt=4,sigma=2,sharpen=1,bw=128,bh=128,plane=4,precision=2,degrid=1,sigma3=5,oh=64,ow=64)

200/900 MHz, 8.0-8.3 fps
400/900 MHz, 12.8 - 13.7 fps
574/900 MHz, 12.8-13.7 fps
574/300 MHz, 5 fps
574/600 MHz, 9-9.5 fps
574/700 MHz 9.1-10 fps
574/1000 MHz, 13-14.5 fps
630/1000 MHz, 13-14.5 fps
temp rose to 80C on the GPU.

Fascinating stuff. I wish we could get ahold of that neat NVIDIA Developer Toolkit! :)
http://www.beyond3d.com/content/articles/10/4

Adub

19th April 2007, 01:34

you should be able to, just search for it on google. Nvidia even has a link to it on their site if I remember. Or maybe that was just the performance viewer.

bill_baroud

19th April 2007, 07:19

Doing this I still see ~65-70 fps. My refresh rate is 60 Hz (LCD).

Do you have something like "wait for vsynch" always activated in your drivers config ? It could be an explanation :)

swaaye

19th April 2007, 17:01

No, I tried forcing vsync off too. Besides, if it was vsync related, it would never hit 70 fps.

http://developer.nvidia.com/page/home.html
I went ahead and tried out NVPerfKit but its built-in "instrumented driver" didn't detect my 8800. I set the NVPerfHub var for FFT3DGPU anyway, to see if it would work anyway, but all that came up was an empty black window.

I also looked at FX Composer, but I really don't have a clue how to set that up with FFT3DGPU.

photoguy123

19th April 2007, 19:09

swaaye,

have you tried processing from your hard drive instead of directly from the DVD?

the data rate is so small it shouldn't make a difference, but at least it would remove one variable from the equation.

swaaye

19th April 2007, 20:01

I'm not processing from the DVD. I always go from the HDD.

With tsp's suggested benchmark I did earlier, I do think the GPU was going full tilt. It got warmer than it does with the processing I'm doing with it normally. And the speed was several times higher than tsp reported for his 7800. So, I think it is working ok. I do think there are some significant bottlenecks that result in the GPU's utilization being not really optimal, but I'm sure these aren't something he can work around easily.

But, I'm only going off the fact that the GPU isn't heating up much. When 8800 is being utilized well you can tell because it heats up to 85C and the fan will spin up a bit to keep it from going higher. I can hear this when playing Oblivion. It's interesting because it literally depends on what you're looking at in the game. The workload apparently changes dramatically and 8800 really reacts to it.

Also interesting is that even with the 8800's massive memory bandwidth, it does seem to be a bit limited by it.

DarkNite

21st April 2007, 13:05

So, I've found that with a 720x480 source off a DVD, I see about 60-85 fps (in VDubMod). I'd say the average is around 70fps.

Something is definitely not right. I get ~108fps with the same script.

ianken

22nd April 2007, 08:57

FWIW it does not run on Vista. Surprise. Whose fault? NVIDIA (the drivers are WEAK) or MS? Dunno. Another low priority item to add to the list I guess.

With the 158.18 drviers fft3dgpu now runs on Vista. Perf rocks on my 8800GTS setup.

shadowhaze

7th July 2007, 20:14

tsp - AWESOME program :thanks: Such an ingenious idea. GPUs are true workhorses. I'm sure you know that NVDA is now launching software for people to use their graphics cards instead of their CPUs for heavy calculations. By using my GPU, I've increased my encoding speed about 3x. It's still relatively slow, but at least it's managable.

Are you still working on the updates? I'd love to be able to use bt=5 like in FFT3DFilter along with a DeHalo option.

A couple of questions. First, I've got my GPU (7900GS) overclocked for 3D @ 595Mhz/1600Mhz, but 2D GPU is at the 525Mhz base clock. Does the program take advantage of my OC? I'm going to download NV Tools - performance since I saw an optimization for that and try to OC the 2d mode (doesn't work right now). Any suggestions for the performance settings?

Second, my card has 256MB of GDDR3. I think that's enough (barely) for one thread encoding, but if I try 2 threads, do I need to put some sort of memory setting in my script? I saw some postes scripts with something like "mem=128".

Thanks again for your help and a great program.

DarkNite

7th July 2007, 23:32

fjhdavid

8th July 2007, 16:52

Dear all,

I use fft3Dfilter from a long time and it is perfect for me but slow!

I tried fft3dGPU with my intel gma 915 graphic card (with direct x 9.0c) and it doesn't work. I use DVD-RB with HC encoder. These card support the pixelshader 2.0 in hardware.

Do I have something special to do? a special tweak?

thanks
francois

shadowhaze

8th July 2007, 21:03

You only need to worry about the memory settings in your script if you have a filterchain that uses copious amounts of memory. If your script isn't that complex you probably won't have to worry about it. Then again, using SetMemoryMax(128) wouldn't harm anything if you didn't need all that memory.

DarkNite - I think I need it. My script is as follows:

Import("C:\Program Files\AviSynth 2.5\plugins\LimitedSharpenFaster.avsi")
Import("C:\Program Files\AviSynth 2.5\plugins\FastLineDarken v1.3.avsi")
AVISource("C:\Documents and Settings\pc owner\My Documents\Bleach to Dvd files\FLCL to DVD\converted video\FLCL 01 Furi Kuri.avi")
DeDot()
VagueDenoiser(threshold=1.5,method=1,nsteps=6,chromaT=2.0)
FFT3DGPU(sigma=2.3,bw=32,bh=32,bt=3,sharpen=0.3,mode=1,precision=2,ow=16,oh=16,wintype=0)
LimitedSharpenFaster(Smode=4)
FastLineDarken(strength=32,thinning=0)

Where do I put "SetMemoryMax(128)" in the script? Also, if my video card has 256mb of GDDR3, should I use 128 or 256? I really don't want to slow anything down, but I do need more stability.

Thanks.

Thanks.

shadowhaze

8th July 2007, 21:16

Dear all,

I tried fft3dGPU with my intel gma 915 graphic card (with direct x 9.0c) and it doesn't work. I use DVD-RB with HC encoder. These card support the pixelshader 2.0 in hardware.

Do I have something special to do? a special tweak?

I'm assuming when you say you have an Intel 915 graphics card, you mean you have Intel integrated graphics (on the motherboard) rather than a discrete video card.

I haven't been using this program long, but I understand the idea behind it. Unless you have a separate graphics card (ex. NVidea 7900GT or 8800GTs, or ATI x1950XT), you can't use this program effectively, if at all. Integrated graphics are not designed for heavy work or hard core gaming - just picture and basic video stuff. You need a stand alone (discrete) graphics card for this program to work effectively, if at all. The better your graphics card, the faster your processing (ex. you'll get much higher FPS w/ a nVidia 8800GTS than with a nVidia 7800GT). There's a list of "good" and "bad" graphics cards in the read me file.

Hope this helps.

tsp

8th July 2007, 21:25

fjhdavid: There are no such thing as intel GMA 915. The intel 915 chipset contains the intel GMA 900. It doesn't support hardware vertex processing and I think that it might cause the problems. Anyway I think it would be faster to use fft3dfilter than fft3dgpu with an integrated intel GPU :)

shadowhaze: As fft3dgpu uses directx 3d the Geforce will run at the 3d speed so you will see some improvement by overclocking. Just be carefull not to overclock too much as fft3dgpu is very sensitive to it (it will produce artifacts that will be saved in the final video)

Also SetMemoryMax only affects how much mainmemory avisynth will use for its cache not how much gpu memory fft3dgpu will use. If fft3dgpu runs out of GPU memory it will swap some of the textures to main memory. This might cause some slow down. So SetMemoryMax should not be necessary. Best thing to do is to try running two threads.

Prim3

9th July 2007, 05:55

lol

1.40FPS

Mpeg2Dec3_Mpeg2Source("C:\Documents and Settings\Makaveli\Desktop\Test Vob 1\test1.d2v")
SetPlanarLegacyAlignment(true)

FFT3DGPU(sigma=3.0, bt=3, bh=128, bw=128, mode=1)

ConvertToYV12

It's probably my x264 settings. What would you recommend for best settings tsp? Time, is not a problem for me. It's all about the quality, for me. Can it also be the bh, bw?

Also, with Fizick's filter, it's about the same. Around 1.40FPS. So, I'm pretty sure it's my x264 settings.

And I'm running 8800 GTS 320MB.

shadowhaze

9th July 2007, 20:24

lol

1.40FPS

OUCH! I've refined my script a little (see below) as I was having trouble with LimitedSharpenFaster, and now I'm averaging 20FPS encoding avi w/ HCenc. That's about 3 times faster than FFT3dfilter in the same chain.

[DeDot()
VagueDenoiser(threshold=1.5,method=1,nsteps=6,chromaT=2.0)
FFT3DGPU(sigma=2,bw=32,bh=32,bt=3,sharpen=1,mode=1,precision=2,ow=16,oh=16,wintype=0)
FastLineDarken(strength=32,thinning=0)]

It's probably my x264 settings. What would you recommend for best settings tsp? Time, is not a problem for me. It's all about the quality, for me. Can it also be the bh, bw?

Also, with Fizick's filter, it's about the same. Around 1.40FPS. So, I'm pretty sure it's my x264 settings.

And I'm running 8800 GTS 320MB.

8800 GTS - Nice! :) That should be twice as fast as mine - 7900GS (assuming same CPU, etc.). I've tried x264 encoding and got weird results. I suggest you do a comparison of FFT3dGPU vs FFT3dfilter using XVID instead of x264. That way you can determine whether it's x264 or something else.

Also, your antialiasing and anisotropic settings may cause problems. I have mine set at application controlled for encoding. Maybe tsp can comment on what effect, if any, these have on FFT3dGPU.

shadowhaze

9th July 2007, 20:39

shadowhaze: As fft3dgpu uses directx 3d the Geforce will run at the 3d speed so you will see some improvement by overclocking. Just be carefull not to overclock too much as fft3dgpu is very sensitive to it (it will produce artifacts that will be saved in the final video).

Oops :rolleyes: I have mine just about maxed out (595MHz GPU/1.68Ghz DDR3). I didn't see any artifacts in the vob file during a quick scan, but I'll drop it down a little just in case.

Also SetMemoryMax only affects how much mainmemory avisynth will use for its cache not how much gpu memory fft3dgpu will use. If fft3dgpu runs out of GPU memory it will swap some of the textures to main memory. This might cause some slow down. So SetMemoryMax should not be necessary. Best thing to do is to try running two threads.

I'm using HCenc so I can't run two threads, but I could try it w/ QuEnc as test since HCenc is supposed to support multi-threading in the next version.

Please see my post above re: the question on antialiasing and anisotropic settings. Also, after reading the FFT3dfilter thread, I thought bw & bh = 32 (default) were usually the best settings. Any thoughts? Finally, please let me know if you're planning to update FFT3dGPU anytime soon. I'd like an option to use bt=5 (bt=4 doesn't see that great), and I'm not sure if this version has the Gaussian High Pass Filter for sharpening introduced in FFT3dfilter v1.7. In fact, I'm not sure what version of FFT3dfilter equates with FFT3dGPU version 0.8.1.

Thanks for the reply and great program.

tsp

11th July 2007, 22:06

shadowhaze: It is best to set the antialiasing and anisotropic to application controlled as it is not used by fft3dgpu.
Also when the encoder uses more than 1 thread it usually only fetch one frame at a time from avisynth so there shouldn't be such a big difference from avisynth's perspective.
Also fft3dgpu 0.8.1 is closest to fft3dfilter 1.8. I plan to introduce the dehalo and bt=5 some day (but no ETA yet)

Prim3: if you only get 1.4 FPS it sounds like it is x264 that are the limiting factor. The best settings depends very much on the source so you will have to experiment (but do use precision=2 for best quality).

survivant

15th July 2007, 02:29

I try fft3dgpu 0.8.2 but I can't get it to work form me. THe image is corrupted. I have a screenshot here

http://24.202.22.217:8282/fft3dGPU.png

I have Windows XP, ATI X800 Pro 256 megs. DirectX 9.0c.

the script is simple.. it work if a remove the line for fft3dpgu and replace it by fft3d

G_M_C

15th July 2007, 10:59

survivant

15th July 2007, 11:21

G_M_C

15th July 2007, 13:34

movie = mpeg2source("C:\DVD\encode\automkv\intro.d2v",info=3)
function getOrder(clip c) {
order = GetParity(c) ? 1 : 0
Return order }
#filter1start

#test3

movie = tfm(movie,d2v="C:\DVD\encode\automkv\temp\movie.d2v").tdecimate()
#resizestart

#test4

last = movie
#filter2start

#test5

crop(16,16,-16,-16)
Spline36Resize(704,400)

fft3dGPU(mode=1,precision=2)

ColorMatrix(mode="Rec.709->Rec.601")

When you try this one, do you get better results ?

mpeg2source("C:\DVD\encode\automkv\intro.d2v",info=3)
ColorMatrix(mode="Rec.709->Rec.601",hints=true)
tfm(d2v="C:\DVD\encode\automkv\temp\movie.d2v").tdecimate()
crop(16,16,-16,-16)
Spline36Resize(704,400)
fft3dGPU(sigma=0.20,sigma2=0.25, sigma3=0.35,sigma4=0.50,precision=2)

Else try to get newest version of DX9c and/or updated drivers; And try my script again.

PS: Used low settings for sigma's. when your source is allready clean enough you really should not have to use sigma's > 1. For DVD and/or HD sources i allway use settings < 1 when the source is clean allready.

survivant

15th July 2007, 13:46

PS: Used low settings for sigma's. when your source is allready clean enough you really should not have to use sigma's > 1. For DVD and/or HD sources i allway use settings < 1 when the source is clean allready.

thanks.

I posted the last script that I use.. but even the simplest one won't work.. look at the image I posted.

movie.avs

DGDecode_mpeg2source("C:\DVD\encode\megui\intro.d2v",info=3)

the line for fft3dgpu

fft3dgpu()

or fft3dgpu(movie)

or I try the line from the manual

fft3dGPU(mode=1,precision=2)

it give me always a corrupted image.

I have recent DirectX and Drivers.

there is a way to launch fft3dGPU with debug information or verbose mode.. I could run it and post the result.

thanks

survivant

16th July 2007, 22:44

Reinstall my drivers.. and still the same problem

survivant

17th July 2007, 00:57

I jsut install DirectX 9.0c June 2007

and when I try to script in AVSP, I obtain an error : Error Creatind Direct3D Device

G_M_C

17th July 2007, 13:09

Hopefully TSP has the time to look at it soon, cause it seems to be a problem witch no-one else can adress. You might wanna use the non-gpu FFT3D in the mean time, as a slow (but hopefully temporary) work-around.

survivant

17th July 2007, 15:21

exactly. I'll continue to use FFT3D for now.. but it's 2fps :(

tsp

17th July 2007, 18:57

survivant: could you try different mode=0,1,2 and/or oldfft=false/true and see if it makes any difference?

survivant

17th July 2007, 19:28

OK, I'll tonight. send my all the tests that you want that I do and I'll send you the results.

I try all the combinaisons but give the same results.

here the avs (with few lines in comments)

mpeg2source("C:\DVD\encode\automkv\temp\movie.d2v",info=3)

#FFT3dGPU()
#FFT3dGPU(mode=0)
FFT3dGPU(mode=1,oldfft=true)

G_M_C

19th July 2007, 06:55

Archiver

20th July 2007, 01:27

@survivant

I just came by to see if there is an updated version of fft3dgpu.

I use fft3dgpu (0.8.2.0) since some time with different Nvidia cards. When there was any (visual) output issue, it did not matter what card-model, driver and directx version i used, i always got some weird output. Replacing fft3dgpu with fft3dfilter solved the symptoms, but was not useable, due to poor performance.

The video-output issues simply had either to do with the input format and or the usage of fft3dgpu parameters. My experiences led me to the following:

- use mode=1
- use bt=3 (bt=4 -> can worsen results)
- use precision=0 (precision=2 -> can worse results)
- make sure you set the "interlaced=" parameter properly !!
- bh, bw values higher than default 32 can lead to decreased quality

These hints are more likely to get max. quality, except that the correct interlaced parameter value is vital.

I saw your sample pic, even i do not know what the original should look alike, it reminds me of the fact that fft3dgpu is very sensitive for "input modulos" !!

Make sure you do not have "weird" input resolutions, always make sure that you always have (at least !) modulo 4 = 0.

tsp made a great tool, but in some cases it needs some experience and or experimenting with it's quirks, to get the "same" results as with the usage of fft3dfilter.

Hope you can make it run properly, fft3dgpu is really a fantastic tool, especially when it denoises in such quality and much faster than real-time. ;-)

I hope these hints can help solve your problems, if not, they might help others. At least they made me love fft3dgpu ;-)