(I've put in some comparison screenshots and scripts below for the discussion.)
Scintilla wrote:
That never fails to amuse me.
Haha, uhm... Oops. C'mon, it's such a pretty name!! *duck*
By the way (as I guess you've already found out), FFT3DGPU ought to work exactly the same as FFT3DFilter...
Yepper. Though I haven't run the original filter, the GPU version is smashingly beautiful. Given GPU's origins, I'd imagine they'd be mightily comparable. The
very worst parts of the video (extremely high contrast areas which are rare) still show buzzing, but the rest of it is squeaky clean.
I did find the guide quite helpful. *stage whisper* But I surely do wish there was a section on rainbow noise. *duck*
I did get the latest version of the plugin (I always check if I can) and there was not a previous version installed. I have not re-tested it since finding the GPU version. I... admit to fear...

That, and since I bought a new video board, I figure I might as well put it to some use.
And now for something completely different:
Here's a screenshot of the ever-sexy Yuki Eri, aka Uesugi Eri. (It's always bothered me that Shindo calls him by a dead man's last name, but... nevermind.) The screenshot, as you can see, is like the worst one I could find, full of mosquito noise until the cows come home.
It was cleaned with the following script:
Code: Select all
LoadPlugin("C:\Program Files\AviSynth 2.5\plugins\DGDecode.dll")
MPEG2Source("C:\Users\Administrator\DVDMovie\VTS_01_1Test.d2v",cpu=6)
TDeint(mode=2,mtnmode=3,blim=100)
Crop(8,0,-8,-0)
Lanczosresize(640,480)
FluxSmoothT(18)
FFT3dGPU(sigma=8,bw=64,bh=64,bt=4,ow=32,oh=32,sharpen=0.7,plane=4,precision=1)
VagueDenoiser(threshold=5,method=3,nsteps=8,chromaT=0)
Here are the results.
I'm guessing if I tried to encode with that script, my machine would crash out (again). Though I suppose I could try overclocking my video board, cranking the fan to 100, and giving it a shot. Oh wait, just remembered that the overclocking software is not in operation. Hmmmm....
Nevertheless, at that strength, FluxSmooth starts to invent its own artifacts, so it's not practical to use on an entire episode. (This is why VaugeDenoiser comes last in the script, to clean up artifacts left by FluxSmooth.)
Here's the weird part. This is
THE VERY NEXT FRAME, unfiltered except for resize and deint.
The cleanup used on this frame was:
Code: Select all
LoadPlugin("C:\Program Files\AviSynth 2.5\plugins\DGDecode.dll")
MPEG2Source("C:\Users\Administrator\DVDMovie\VTS_01_1Test.d2v",cpu=6)
TDeint(mode=2,mtnmode=3,blim=100)
Crop(8,0,-8,-0)
Lanczosresize(640,480)
FluxSmoothT(13)
FFT3dGPU(sigma=4,bw=32,bh=32,bt=4,ow=16,oh=16,sharpen=0.7,plane=4,precision=1)
VagueDenoiser(threshold=4,method=3,nsteps=6,chromaT=0)
Here are the results.
Conclusion: Anime encoders are on crack.
The small segment those two frames come from is an interesting and frightening mix of clean and noisy-as-hell frames. It almost looks like you'd have to do that on purpose.
I've settled on using the moderately strong script in the second example on the whole video and then tweaking sections as necessary if I need to later on.
If there are any suggestions on tweaking the script (i.e. you try something and get better results than I did), I'm all ears. <^.^>
(I have a thing for bishounen antiheroes. Sue me. Sesshoumaru, I'm yours when you want me...)