Current Server Time: 11:58 (Central Europe)

#apertus IRC Channel Logs

2022/10/09

Timezone: UTC


06:03
balrog
left the channel
06:12
balrog
joined the channel
15:12
anuejn
vup: polyrhythm se6astian isnt there like pretty accurate spectral response in the datasheet of the image sensor?
15:12
anuejn
cant we just trust this data?
15:12
anuejn
at least for relative values?
15:14
vup
se6astian: can do later
15:15
vup
anuejn: maybe, it doesn't really specify errorbars
15:15
anuejn
could we 'just' derive all our color science from that
15:16
anuejn
or and why is that a naive thought
15:16
vup
So my assumption is they measured exactly one sensor
15:16
anuejn
vup: so no errors ;)
15:16
anuejn
quite plausible
15:17
se6astian
no idea tbh
15:17
se6astian
I wouldnt even know how to translate the response curve to a color interpretation
15:18
vup
Also it doesnt say anything about the response curve wrt intensity
15:20
anuejn
wrt intensity?
15:21
anuejn
but theoretically there is all the data to generate the two color matrices from this diagram if we assume that the sensor is perfectly linear?
16:15
aombk
left the channel
16:19
aombk
joined the channel
18:06
vup
Probably, although it could also be impossible
18:06
vup
(In general, regardless of the data we have)
19:00
anuejn
why or why not?
19:18
Bertl
it's impossible because it won't work with a matrix ;)
19:19
anuejn
what wont work with a matrix?
19:19
anuejn
like, you have to specify two matrices for the dng to be developed correctly
19:19
anuejn
for two different iluminants
19:20
anuejn
to map sensor output into CIEXYZ
19:20
Bertl
we already know, we need something more powerful than a matrix (or two)
19:23
anuejn
for nonlinear response of the image sensor?
19:25
anuejn
but even if we wanted to create something like a 3d lut, couldnt we just "simulate" test pics using the spectral response, the spectrum of some standart iluminant and the absorbing spectrum of our color targets?
19:30
Bertl
yes, in theory we could, we also need to take the lens into account though
19:31
Bertl
I was just commenting on the matrix part, hence the ';)'
20:20
anuejn
okay
20:49
anuejn
huh turns out you can have 3d luts in dng files
20:49
anuejn
but in a kind of odd form
20:50
anuejn
only in hsv form
21:21
anuejn
se6astian: do you have a link to the filter spectral response data for me?
21:22
se6astian
The uv it filters?
21:23
se6astian
*IR
22:00
anuejn
yup
23:44
anuejn
I obtained a pair of plausible color matrices from the datasheet spectral response :)
23:46
anuejn
(see https://github.com/anuejn/cmv12k_color_response/blob/main/color_matrices.ipynb)
23:50
anuejn
and it even kind of works in davinci resolve