Using denoisers for compresion gain (Tests)

[SILICON]

I made test with several denoisers.
I used a 15 minutes sample of film "die hard", 2.35:1 aspect ratio, don´t anamorfic. Medium quality source ripped of DVD.

I compresed with CCE 2.66.01.07 (demo version) with Bitrate min = 300, max = 2350, Quality = 30, MPEG2 (CVD). All CCE filters on off.
The version of denoisers are the last at 8/02/2004 in the YUV12 flavour.
I used Avisynth 2.54
The CPU as AMD K7 2200+


3D FILTER SIZE(BYTES) SPEED (FPS)
----------------- ----------- -------
GoldDust() 82.272.040 0.61 (LoadPluginEx2)
PixieDust() 91.717.800 0.61 (LoadPluginEx2)
MipSmooth(preset="movieHQ") 93.159.424 1.06
Deen() 93.301.800 1.27
MipSmooth(preset="movieHQ2") 93.693.380 1.10
FluxSmooth(7,7) 96.275.124 1.31
NoMoSmooth() 96.349.288 0.90 (YUY2)
PeachSmoother() 97.484.800 1.22 (YUY2)
STMedianFilter(8,15,4,7) 97.995.892 1.35
Convolution3d(preset="movieHQ") 98.369.668 1.26
NONE 99.940.388 1.43

Temporal FILTER SIZE(BYTES) SPEED (FPS)
----------------- ----------- -------
FaeryDust() 94.209.460 0.61 (LoadPluginEx2)
TemporalSoften(4,4,8,15,2) 95.322.092 1.22
atc() 95.340.012 1.19
TemporalCleaner() 96.688.972 1.29
GrapeSmoother() 97.582.316 1.37 (YUY2)
cnr2() 99.940.388 1.24 (No size change)
antiflicker() 104.109.728 1.23 (YUY2)(More size)

Frecuence FILTER SIZE(BYTES) SPEED (FPS)
----------------- ----------- -------
VagueDenoiser() 94.060.176 1.06
DctFilter(1,1,1,1,1,1,.5,0) 98.769.368 1.38

Spatial FILTER SIZE(BYTES) SPEED (FPS)
----------------- ----------- -------
SmoothHiQ(7,20,25,230,5) 90.603.520 0.44 (YUY2)
MSmooth() 93.319.796 0.62
SpaceDust() 93.457.196 1.22 (LoadPluginEx2)
Unfilter(5,5) 95.039.200 1.39
deen("a2d",2,10,12) 95.358.448 1.27
DNR2() 96.240.152 1.40
UnDot() 99.201.016 1.39
Mergechroma(blur(1.3)) 99.274.728 1.34
Blockbuster(method="blur") 100.005.980 1.37 (Increase size)


Any other test?
Any program for objetive test of quality?

[Soulhunter]

Originally posted by SILICON
I forgot any 3D denoiser?
Cant remember which ones are working in 3D mode... ;)


- Temporalsoften
- DNR2
- TemporalCleaner
- Peachsmoother
- SpaceDust
- SmoothHiQ
- FaeryDust
- PixieDust
- GoldDust
- VagueDenoiser
- NoMoSmooth


Bye

[Chainmax]

VagueDenoiser is spatial only. IIRC, all flavors of Dust (except spacedust) are spatio-temporal (3D). I don't know about the rest.

[Dreassica]

smoothhiq is spatial only, temporalsoften only temporal,temporalcleaner likewise, not sure about nomosmooth.

[SILICON]

- SpaceDust, SmoothHiQ, FaeryDust, PixieDust, GoldDust. Don`t test because is only AviSynth 2.0x.

I don´t know how test its

Anymore filter for test?

[Lefungus]

Just to note that vaguedenoiser isn't a frequency filter, and like it has been said earlier, is spatial only (no 3d).

[Didée]

Originally posted by SILICON
- DUST, SmoothHiQ, FaeryDust, PixieDust, GoldDust. Don`t test because is only AviSynth 2.0x.
But then you are missing the greatest surprise of your comparison!

Use:

LoadPlugin( "...\LoadPluginEx.dll" )
LoadPlugin( "...\dustv5.dll" )

"LoadPluginEx.dll" is provided by the (japanese?) WarpSharp.dll package. By its means, you can load and use all AviSynth-2.0 plugins into AviSynth 2.5.x. First load all your 2.5 plugins, then load LoadPluginEx.dll, and *after this* load the 2.0 plugins you want to use. Works perfect for me and others, each and every day.
I'm looking forward to your findings about PixieDust. You'll be happily surprised by the compression, and be badly surprised by the speed ...

- Didée

[Mug Funky]

sorry to seem a little thick, but how is this comparison different to the sticky thread "speed and compression of different denoisers"?

also, you could try straight "blindPP" as this does a good job (if you're worried about losing detail, set it for deblocking only.)

[Soulhunter]

Originally posted by Mug Funky
sorry to seem a little thick, but how is this comparison different to the sticky thread "speed and compression of different denoisers"?
Not much, but SILICON encoded his stuff to MPEG2... ;)

Bye

[SILICON]

Originally posted by Mug Funky
sorry to seem a little thick, but how is this comparison different to the sticky thread "speed and compression of different denoisers"?

The original idea are the same.

But I test it for CCE in SVCD format.
I have test of new versions of the same filters and of new ones. The sticky thread "speed and compression of different denoisers" are old.

Al test are make with BlinPP (CPU = 4 in MPEG2DEC3)

[mf]

Originally posted by Mug Funky
also, you could try straight "blindPP" as this does a good job (if you're worried about losing detail, set it for deblocking only.)
Another alternative could be: BlindPP(cpu=4).HQDering(smoother="BlindPP", params="cpu2="+Chr(34)+"ooooxx"+Chr(34)) if you're worried about losing detail but still want deringing anyway ;).

[Wilbert]

I have test of new versions of the same filters and of new ones. The sticky thread "speed and compression of different denoisers" are old.
Let me know if you want to replace the sticky with a series of new tests :)

[SILICON]

Originally posted by Lefungus
Just to note that vaguedenoiser isn't a frequency filter, and like it has been said earlier, is spatial only (no 3d).

VagueDenoiser and DCT work in one frame, but makes and DCT tranform for make the changes.
A spatial filter work with nearest points, a frecuency filter work with nearest frequencies.

Am I wrong?

Very good filter for increase the compresion, but is a bit slow. ¿can you speed up?
You can use the DCT code of MPEG2DEC3. It's very fast.

[SILICON]

Originally posted by Didée
Use:
LoadPlugin( "...\LoadPluginEx.dll" )
LoadPlugin( "...\dustv5.dll" )

"LoadPluginEx.dll" is provided by the (japanese?) WarpSharp.dll package.

I used this script:
[CODE]# Load external plugins
LoadPlugin("C:\PLUGINS\MPEG2DEC3.dll")
LoadPlugin("C:\PLUGINS\LoadPluginEx.dll" )
LoadPlugin("C:\PLUGINS\DustV5.dll")
Mpeg2Source("C:\DIEHARD.D2V", IDCT=7, CPU=4, iPP=FALSE)

# Trim 15 minutes of film
TRIM(20000,42500)

# Crop the black borders
CROP(4,70,-4,-72)
LumaFilter()

# Change to CVD resolution
BilinearResize(336,320)

# The filter to test (activate yuy2 only if the filter need it)
# ConvertToYuY2()
FaeryDust()
# ConvertToYv12()

#Restore the black borders
AddBorders(8,128,8,128)

# change for CCE input
Limiter()
ConvertToYUY2()[CODE]

Avisynth Says:
LoadPlugin: Unable to load "C:\PLUGINS\LoadPluginEx.dll"

- LoadPluginEx.dll are in C:\PLUGINS\
- LoadPluginEx.dll have a 3/11/03 22:19 date (60KB)
- Avisynth version 2.54, build: JAN 26 2004 [09:28:02]

Work ok with LoadPluginEx2.dll

_2DCleanYUY2.dll give the same error. I can test it.

[SILICON]

Originally posted by Wilbert
Let me know if you want to replace the sticky with a series of new tests :)

Not replace. When I finish the test of all sugested filters, you can add this test to the previous ones.

[Lefungus]

Originally posted by SILICON
VagueDenoiser and DCT work in one frame, but makes and DCT tranform for make the changes.

Dctfilter and Vaguedenoiser are completely different. One use DCT while the other one use wavelets. So no it's impossible to use dct of mpeg2dec3 as i don't use any DCT transformations anywhere in the plugin.

[dragongodz]

it would be interesting to see how Donald Grafts MSmooth filter does aswell.

[SILICON]

Originally posted by Lefungus
Dctfilter and Vaguedenoiser are completely different. One use DCT while the other one use wavelets. So no it's impossible to use dct of mpeg2dec3 as i don't use any DCT transformations anywhere in the plugin.

Vaguedenoiser don´t work in the frecuence domain?.
I can put Vaguedenoiser in "frecuence filters" or in "Spatial filters"?

Thanks

[SILICON]

Originally posted by dragongodz
it would be interesting to see how Donald Grafts MSmooth filter does aswell.

I made the test with the default params.

The DOC don´t give the best values for real image. The filter don't have presets.

If you know best params for real image, tell me. I can repeat the test.

[MfA]

Silicon, the correct term is transform domain. Frequency domain mostly is taken to mean fourier space. Neither Vaguedenoiser nor DctFilter work in fourier space (due to poor localization noone really does denoising in fourier space).

Spatial denoising on this board tends to be a way of saying it is not spatio-temporal, ie. it uses only spatial information which is true for both Vaguedenoiser and DctFilter. It does not necessarily mean that the filter operates in the image domain only.

Strictly speaking the filters using mipmaps work in a quasi transform domain too, so I dont think the kind of distinction you are trying to make is really a valuable one.

[Soulhunter]

Nice update... !!!

Bye

[Boulder]

Originally posted by SILICON
cnr2() 99.940.388 1.24 (No size change)


CNR2() is a bit flawed. sh0dan did try to fix it but in his last version the scene change detection is broken somehow. The value needs to be very high for the filter to do anything but when the filter kicks in, it is too aggressive.

See this thread: http://forum.doom9.org/showthread.php?s=&threadid=64831

I've been hoping that someone would fix CNR2 for good as it's a good one for any analog capture;)

[ADLANCAS]

I´ve also tested this version 2.51
http://forum.doom9.org/showthread.php?s=&threadid=68745
I've been hoping that someone would fix CNR2 for good as it's a good one for any analog capture
me too!

[windtrader]

This is a nice list of filters and performance measures. Is there any place where there are samples of the original and post filter results? This would be a super addition to this information so folks can decide which filters may be of most benefit in different situations.

[LB]

why don't we test dedot() ?