Strange behavior when building profile
Posted: Wed Sep 14, 2011 10:07 am
Hello.
Files and screenshot are here : http://dl.free.fr/gfrQvSKt6
With v2.2, when manualy building profile, i've noticed that a specific value could only increase, never decrease (except for the 1rst time when value had never been assigned).
Otherwise, for exemple, 1rst value of Red channel, can go the 1rst time to -10%, and after, you can sample others area, value may eventualy increase ,up to +15% for exemple, but afterward, will never decrease, and will in final state stays at +15% if there is no area with more than +15% sampled during manual building.
Logical, in way you are supposed to keep the most noisy value/part.
With 2.6, value can go up and down. During process, your 1rst value of Red channel can go up to +10%, after going to -15%, going after to +30%, and have a final value of -5%...
This behavior is illogical for me, because you can in this way have a final profile with values lower than the most noisy part, even lower than the most representative noisy part.
With 3.0, going more and more strange...
First, when i've 1rst try to manualy build a profile i've noticed :
- Even after a reset of profile, profile viewer said there is clipping in B chanel !! (Image3). How is it possible wihtout even having made a sample ?
- Quality stay at 8% and never change, despite how many areas i'm sampling.
- Values of noise were so HUGE, that i decided to make some test.
I've sampled the exact same area from same picture with v2.2 and v3.0.
In the provided pictures (screenshots), you can see :
- Image1 : Sampling of picture with v2.2
Result was :
R : -68% G : -50% B : -43%
- Image2 : Sampling sampe area with v3.0
Result was :
R : +261% G : +169% B : +1012% !!!!!!!!
You can see both profiles open in the same times with v3.0, both created from the same area of the same picture (Image4).
Is it realy normal ???
Considering these behavior, i consider v2.6 suspect, and v3.0 absolutely untrustable for creating noise profiles, and for now, i'm still using v2.2, wich behavior seems the most realiable.
Files and screenshot are here : http://dl.free.fr/gfrQvSKt6
With v2.2, when manualy building profile, i've noticed that a specific value could only increase, never decrease (except for the 1rst time when value had never been assigned).
Otherwise, for exemple, 1rst value of Red channel, can go the 1rst time to -10%, and after, you can sample others area, value may eventualy increase ,up to +15% for exemple, but afterward, will never decrease, and will in final state stays at +15% if there is no area with more than +15% sampled during manual building.
Logical, in way you are supposed to keep the most noisy value/part.
With 2.6, value can go up and down. During process, your 1rst value of Red channel can go up to +10%, after going to -15%, going after to +30%, and have a final value of -5%...
This behavior is illogical for me, because you can in this way have a final profile with values lower than the most noisy part, even lower than the most representative noisy part.
With 3.0, going more and more strange...
First, when i've 1rst try to manualy build a profile i've noticed :
- Even after a reset of profile, profile viewer said there is clipping in B chanel !! (Image3). How is it possible wihtout even having made a sample ?
- Quality stay at 8% and never change, despite how many areas i'm sampling.
- Values of noise were so HUGE, that i decided to make some test.
I've sampled the exact same area from same picture with v2.2 and v3.0.
In the provided pictures (screenshots), you can see :
- Image1 : Sampling of picture with v2.2
Result was :
R : -68% G : -50% B : -43%
- Image2 : Sampling sampe area with v3.0
Result was :
R : +261% G : +169% B : +1012% !!!!!!!!
You can see both profiles open in the same times with v3.0, both created from the same area of the same picture (Image4).
Is it realy normal ???
Considering these behavior, i consider v2.6 suspect, and v3.0 absolutely untrustable for creating noise profiles, and for now, i'm still using v2.2, wich behavior seems the most realiable.