Am 28.12.2017 um 11:35 schrieb FK:
> Am 27.12.2017 um 22:41 schrieb Matthias Bodenbinder:
>> But my initial question is more from a theoretical point of view. I am
>> just curious about the algorithm so to speak. Does it make a
>> difference if denoise profile, wavelet, unifom blend color is applied
>> before or after denoise profile, wavelet, unifom blend lightness. Does
>> the order matter at all?
>> Matthias
>>
> Hi,
> 
> wow, three denoise steps. Does take some computing time ;-). I use two:

Yes, thats true. But the result is good.

> 
> #1: wavelets, strength: 0.25, blend mode: color 95%
> #2: non-local-means, strength: 1.0, blend mode: lightness 40% - 50%

I tried your settings as well, but for high ISO I get best results with 3 
profile denoise instances. It is
#1: wavelet; color
#2: wavelet; lightness
#3: non-local: average

I keep strength at 1 and opacity at 50%. This is my default setting for ISO 
higher than 10.000. For even higher ISO ike 25.600 I go with opacity 75% and 
100% for #2 and #3.

In the meantime I tested different instance sequences: 1-2-3 / 2-1-3 / 3-1-2 / 
3-2-1
I compared exported 16bit PNG. Just by viewing the pictures I did not find 
differences. Then I loaded the images into gimp 2.9.8 in different layers. When 
applying overlay mode "difference" the result is a pure black picture with just 
a very very few changed pixels in the noise. So for me, by all pratical means, 
the resulting images are indistinguishable.


> 
> I did some tests some time ago and found these settings (and this order)
> o.k. though it slowed down the performance quite a bit.
> Not sure whether the order changed much, but i.m.o. it makes sense to
> get rid of the color-noise first and smooth the result afterwards a bit.
> But that's just trial and error here.
> 
> Would be interested in the opinion of someone with some insight
> knowledge too :-)
> 
> Regards,
> 
>    Felix
> 
>>
>>> In the exact order you see in the right sidepanel, from bottom to top.
>>>
>>>> I have the following scenario:
>>>>
>>>> I apply the profiled denoise module 3 times. All three blended "uniformly" 
>>>> but with a differen blend mode:
>>>> #1: wavelets, blend mode: color, 50%
>>>> #2: wavelets, blend mode: lightness, 50%
>>>> #3: non-local-means, blend mode: average, 50%
>>>>
>>>> I am now wondering if the order in which these 3 instances are applied 
>>>> makes a difference.
>>>>
>>>> Matthias
>>> Roman.
>>>
> 
> ____________________________________________________________________________
> darktable user mailing list
> to unsubscribe send a mail to [email protected]
> 
> 


____________________________________________________________________________
darktable user mailing list
to unsubscribe send a mail to [email protected]

Reply via email to