Hello,
some more thoughts on the recent Olympus ยต43 models which presumably all share
the same sensor (E-M5, E-M10, E-PL5, E-PL6, E-PM2, E-P5).
Looking at the green channel of my E-P5 noise data and the values already
contained in DT's noiseprofiles.h for E-M5 and E-PL5 I noticed that my "a"
coefficient data points for ISO 2000 and 5000 might have been outliers and a
linear regression for the a-parameter (coeff. 1.5869e-08) could easily fix
that. Also the values for ISO 25600 may be used from the other models.
For the second parameter (b) things are not as simple, as it seems (see
attached DT-Noise_Olympus_E-M5_PL5_P5.png). The E-P5 data seems to suggest a
better behaviour up to and including ISO 8000 and then a jump to the worse
values of the E-M5 and E-PL5 at higher ISOs. Also, looking at a more detailed
graph (attached detail.png) up to ISO 3200 the E-P5 data seems much more
consistent, which in turn might indicate problems with interpolation of
missing b-values from the data in noiseprofiles.h for E-M5 and E-PL5 - right?
Besides the two attached plots, this is the best I could come up with so far
(certainly not more than 4 significant digits, data format usable for gnuplot,
only use green channels 1:3 or 1:6):
# regression: a = c*ISO (green channel, 1:3) 1.5869e-08 -> fixed ISO 2000 & 5000
# ISO 25600 taken from comparison to e-pl5 and e-m5 (same sensor)
# the values for green channel parameter b seem to be better for e-p5 up to
# ISO 8000 (better raw AD pipeline?) and then jump to the level of e-pl5/e-m5
200 4.44424187741465e-06 2.99855385072016e-06 1.08317199045895e-05
1.09677288263148e-08 9.04107099379896e-09 1.30291773934123e-08
250 5.56404568178042e-06 3.88782991026818e-06 1.52908824923757e-05
1.27627405571592e-08 1.06119420248892e-08 1.27942248601444e-08
320 6.78322923719586e-06 4.75643030927288e-06 1.85181122371483e-05
1.87557682468392e-08 1.53786861708833e-08 1.88038157506827e-08
400 1.04253812695938e-05 7.02374112282618e-06 2.31162817450733e-05
9.25984889839442e-09 7.79100363069323e-09 1.10494506611547e-08
500 1.26764254735444e-05 8.77815265904943e-06 3.24627782499901e-05
1.28713080714169e-08 1.05051836289739e-08 1.56390030553272e-08
640 1.50679168050577e-05 1.05737316413622e-05 4.06051974483925e-05
1.88179942046864e-08 1.54452060621139e-08 2.16370443491548e-08
800 1.85649982247609e-05 1.28150239502753e-05 5.08890579846725e-05
2.79079089036449e-08 2.28431098551382e-08 2.89754478125412e-08
1000 2.588685845435e-05 1.73818666739912e-05 6.39979523258199e-05
2.96740728193177e-08 2.38036561743747e-08 3.35456323419447e-08
1250 3.15717606338564e-05 2.19365827330852e-05 8.54927744902944e-05
4.0904651722172e-08 3.14323709142334e-08 4.25515306524108e-08
1600 3.81893590563908e-05 2.65713214081756e-05 9.79425216236713e-05
7.69151262328425e-08 6.22034013445868e-08 8.26285789994819e-08
2000 5.05306120757209e-05 3.17380000000000e-05 0.000136001257466992
9.36349966938621e-08 7.1397612621607e-08 1.07149354741741e-07
2500 6.10291318571623e-05 4.05546149865647e-05 0.000135298475680528
9.67001727379876e-08 6.74763362268718e-08 1.10272344388142e-07
3200 7.07150976715582e-05 4.84280738873547e-05 0.000157879395132559
1.14458973588607e-07 7.10891425269051e-08 1.26281774710791e-07
4000 0.000104621666701426 6.10851569889013e-05 0.000200520644053834
1.64926977693232e-07 1.07097866815198e-07 1.88935612885867e-07
5000 0.000155023418468101 7.79345000000000e-05 0.000289810885864649
2.64271269003561e-07 1.69423371258086e-07 2.94429977510558e-07
6400 0.000139542899549738 9.63564051003203e-05 0.000307751147062418
3.55545863858681e-07 2.14196621792007e-07 3.68309469223969e-07
8000 0.000244436488769286 0.000130007112785455 0.000387159500653682
4.60616790189685e-07 2.65493005763617e-07 4.78607186628961e-07
10000 0.000289365872315861 0.000154285743005418 0.000454407855408172
4.52846213810293e-06 3.07664963070376e-06 7.85852002087236e-07
12800 0.000358362598252066 0.00018803731599601 0.000469162340435641
4.61868441354983e-06 3.80000488555163e-06 1.14684667121615e-06
25600 3.99e-04 3.99e-04 3.99e-04 1.2e-05 1.2e-05 1.2e-05
------------------------------------------------------------------------------
HPCC Systems Open Source Big Data Platform from LexisNexis Risk Solutions
Find What Matters Most in Your Big Data with HPCC Systems
Open Source. Fast. Scalable. Simple. Ideal for Dirty Data.
Leverages Graph Analysis for Fast Processing & Easy Data Exploration
http://p.sf.net/sfu/hpccsystems
_______________________________________________
Darktable-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/darktable-users
