newer or better filter than dfttest 1.1 ?

[survivant001]

I have a really blocky video (made from a digital camera) that I want to convert and deblock.

I read that dfttest could do the job, but I don't know if there is a newer filter.

the setting that I had is : dfttest(tbsize=5)

the source can be downloaded there http://www.mediafire.com/?mzywrzxd9xs

(38 megs)

[Mystery Keeper]

dfttest is denoiser, not deblocker.

[survivant001]

sorry about the confusion. so there is a denoiser better than dfttest for my source ? I found dfttest better than fftgpu

[jethro]

I've had good results with MC_Spuds() with a movie recorded from camera: much lower bitrate and no real loss of details. Blocking was also automagically gone (more than with Deblock(). But it's not quick obviously..

[Adub]

Yes, of course, there is something better than dfttest.

dfttest with motion compensation.

Which is probably what happens in MC_Spuds().

[survivant001]

thanks. I look MC_spuds(). it's seam to use lot of others filters. THe problem, is to find which settings to use, there are so many.

I compare theses filters this morning.. the source was good, so the result were really similar.
test1 :
FFT3dGPU(video,sigma=1.5, bw=32, bh=32, ow=16, oh=16, sharpen=0)

test2 :
video=dfttest(video,tbsize=5)

test 3:
video=mc_spuds(video,frames=2,strength=4)

will try with the source on the first post.

[Dark Shikari]

dfttest is denoiser, not deblocker.I'd say its one of the best deblockers out there, actually.

[survivant001]

I try with the movie on the first post. with FFT3dGPU(movie,sigma=1.5, bw=32, bh=32, ow=16, oh=16, sharpen=0.4) it's awful.

with ddftest it's really nice

and with mc_spuds.. it's worst than ddftest, but a little better than fft3dgpu.

[Adub]

MC_Spuds() is all about the correct settings. If you spend enough time tweaking it, then you will get excellent results. Better than just using dfttest or just using fft3dgpu.

[survivant001]

thanks

I think I'll start a thread for the input of others. I'll do a webapge with samples and the filters config that produce the best result (for the eyes).

[martino]

with samples and the filters config that produce the best result (for the eyes).
Highly dependant on the actual sample. Meaning, if filter a is "best" for clip x, it may not be for y.

[survivant001]

exactly. Perfectly agree with that, but it still be useful as a guide. We will never obtain a "work for all" solution, but the user will be able to check what differents settings does to a clip, and custom them if needed.

The problem is there so many filters, settings that it hard to know where to start. the same for the x264 settings. I have a command line (ripbot and automkv) that I can use, and I'll be happy with the results, but if I want I can tweak the settings for a special case, but in my case, the command line that I use is enough.

What I want to do, is the same thing with filters. If I have a blocking movie, I will use a specific filter (maybe not the most optimize quality.. but with the default settings found, I'll be happy with that).

If I have a nosy clip.. I 'll use others settings..

in the worst case, the user can try all the settings found in the guide, and try them all, and choose the best for him. It's what I want to do.

[Blue_MiSfit]

It's interesting how fft denoisers tend to do a nice job of deblocking as well!

I use fft3dgpu all the time for this when I re-encode heavily compressed 1080p MPEG-2. It does a great job of deblocking and denoising.

I've noticed DFTTest does well too, but is prohibitively slow for HD, since there's no GPU version... yet :D

~MiSfit

[survivant001]

It's interesting how fft denoisers tend to do a nice job of deblocking as well!

I use fft3dgpu all the time for this when I re-encode heavily compressed 1080p MPEG-2. It does a great job of deblocking and denoising.

I've noticed DFTTest does well too, but is prohibitively slow for HD, since there's no GPU version... yet :D

~MiSfit

good to know that I'm not alone to use fft3dgpu for reencoding :)

and I find dfttest so slow.. I'll continue to use fft3dgpu when needed, and for the clips recorded of my children I'll use dftttest or mc_sprud (need to play with the setting).

with ftt3dgpu I do around 60fps for the 1th pass, but only 5-8 fps for dfttest. Most of the time, ftt3dgpu will do the job.

[Blue_MiSfit]

Yeah, it's pretty good, as long as you're careful with your sigma2,3,4, parameters, and check for banding using histogram(mode="luma"). I almost always dither the result with gradfunkmirror(1.51)

~MiSfit

[survivant001]

ok, now you lost me (banding and gradfunkmirror ) :) can you give me a sample of your avs script ? I will check the params settings with AVSP.

[thetoof]

Banding is having stripes of colors instead of nice gradients and gradfun is what solves the problem by dithering, which is adding "intelligent" grain to hide the stripes.