Current Server Time: 07:35 (Central Europe)

#apertus IRC Channel Logs

2022/10/12

Timezone: UTC


00:04
mustafa__
left the channel
00:25
aombk2
joined the channel
00:27
aombk
left the channel
01:16
se6astian
left the channel
01:16
polyrhythm
left the channel
01:16
mustafa_
left the channel
01:16
Spirit532
left the channel
01:16
elafon
left the channel
01:16
mustafa_
joined the channel
01:20
se6astian
joined the channel
01:20
polyrhythm
joined the channel
01:21
elafon
joined the channel
04:57
balrog
left the channel
06:23
balrog
joined the channel
06:42
anuejn
left the channel
06:42
vup
left the channel
06:42
tpw_rules
left the channel
06:44
tpw_rules
joined the channel
06:45
anuejn
joined the channel
06:45
vup
joined the channel
07:07
anuejn
polyrhythm: that is sad indeed :(
07:08
anuejn
but understandable since CDNG is such a monstrosity that it is basically impossible to implement all of it
07:09
anuejn
I see that resolve is a valuable tool but I also think that we can build a solution for somehow getting our magic into resolve ;)
07:15
anuejn
but yeah i am also looking forward to seeing your results :)
07:16
anuejn
also: do you have any idea if there is a public dataset on spectral reflectance of different realworld stuff (different skin tones, foliage, ...)?
07:53
Spirit532
joined the channel
08:24
vup
polyrhythm: right, but DNG does specify that the transformation matrices are supposed to bring the camera space into XYZ_D50
08:24
vup
(also I stopped being lazy and read the DNG spec myself, there is some pretty crazy stuff in there)
08:24
vup
(for example DNG 1.6.0.0 allows you to specify a custom illuminant by its spectral data
08:26
vup
oh also: does one not need to correct for the wavelength dependent absorption of the lens glass?
08:44
anuejn
I suppose
08:44
anuejn
there is even an ISO standart lens: https://colour.readthedocs.io/en/develop/generated/colour.SDS_LENSES.html
08:45
anuejn
though its spectrum is relatively boring
08:46
se6astian
hmm, as there are tons of different lenses and coating I cant imagine the camera needs to account for what those are doing...
08:47
anuejn
I mean it _could_ but in practice it shouldnt make much of a difference
08:48
anuejn
In the end I get the feeling that colour-science is more more about colour than about science ;)
08:49
polyrhythm
the general idea is that you try to do the calibration with a lens that doesn't have an obvious color cast. modern lenses definitely qualify, especially when put into controlled lighting and looking at a very matte surface like a color chart
08:50
polyrhythm
modern lenses are so good, you really have to try very hard to get them to exhibit any kind of obvious flaws. although in the case of e-mount lenses i'm not sure what is great. i'm used to PL-mount cinema stuff
08:53
polyrhythm
anuejn: the standaard x-rite color chart is a good spectral reference for basic diffuse reflectance of foliage, skin tones, etc.
08:53
polyrhythm
other than that, there's not a specific need in the games/film industry for amassing a spectral DB of material types (yet...!) because most production rendering is taking RGB values and rendering just with that, or in the very specific case of a few studios like weta, they upsample RGB to spectral. hardly anybody is going out in the world to measure spectral data and use that info directly.
08:53
polyrhythm
also because what we think is a relatively straightforward value is actually not really.
08:53
polyrhythm
for example, if you take a spectrometer and sample human skin, you'll come up with some crazy dark value, because human skin doesn't really diffuse much light, it scatters it below the surface.
08:53
polyrhythm
this is all a fancy way of saying that you first have to define what kind of spectral reflectance data is meaningful to you
08:54
anuejn
I see
08:55
polyrhythm
there's lots of stuff like this for RGB values tho https://physicallybased.info/
08:56
polyrhythm
which again is all witchcraft fakery for the most part, totally just taken from neutralized photographs rather than any kind of rigorous spectral measurement
08:56
polyrhythm
and in rec 709 primaries
08:57
anuejn
interesting
08:58
polyrhythm
for games we usually go out with a color chart and just photograph a bunch of materials with a polarizing filter, then we use the chart to color neutralize and the polarizer to try and get rid of specular influence
08:58
polyrhythm
it works pretty decent
08:58
anuejn
I was especially interested in data outside of the visible spectrum to get an idea about the effect of uv/ir cut filters
08:58
anuejn
but that is probably rather hard to get
08:58
polyrhythm
there's a device that can do it rather easily i think, and it's not too pricy, hang on
08:59
anuejn
Like, I would really like to get my hands on a hyperspectral-camera one day :D
08:59
anuejn
polyrhythm: sounds like a solid method
09:01
polyrhythm
https://www.nixsensor.com/nix-spectro/
09:01
polyrhythm
this thing, pretty badass
09:01
polyrhythm
i was evaluating it for use at a game studio
09:02
anuejn
honestly: to me that does not sound to interesting
09:02
anuejn
since it has only 31 channels
09:02
anuejn
and does it have a builtin light?
09:03
anuejn
ok it does
09:03
anuejn
thats nice
09:03
polyrhythm
yeah of course, it has a D50 illuminant if i recall correctly
09:03
polyrhythm
it actually has to position it at a certain angle to defeat specular interference, it's not a straight on thing
09:03
polyrhythm
and yes it buckets the spectral data, but that's still more than enough for most use cases that they're trying to tackle anyways
09:04
polyrhythm
at it's price point there's no competition
09:04
polyrhythm
the next step up is like an order of magnitude more expensive
09:05
anuejn
yeah all in all it seems like a pretty good package
09:05
anuejn
also that its portable is a big plus
09:06
polyrhythm
yeah, plus phone app blah blah. now: how it relates to making a better camera sensor characterization, ehh....not so sure. :) but a fun toy
09:06
polyrhythm
what we need is the opposite, which is a LED or something that can very accurately emit certain wavelengths of light so we can see what the sensor does with it
09:07
anuejn
or a spectrometer and a beamsplitter
09:08
anuejn
so that we can just calibrate out the insufficiencies of the light source
09:08
anuejn
(+ a prism of course)
10:57
vup
well more likely a diffraction grating
10:57
vup
the problem is the intensity calibration
10:57
vup
most cheap spectrometers are not (relative) intensity calibrated
10:58
vup
so for calibration you would still need a light source with a know spectrum
11:06
vup
which is again expensive: https://www.oceaninsight.com/products/light-sources/calibration-sources/radiometric-calibrated/
11:06
vup
https://www.princetoninstruments.com/products/technologies-family/intellical
11:06
vup
left the channel
11:07
vup
joined the channel
11:09
anuejn
probably we could calibrate with some university equipment?
11:09
anuejn
either you or se6astian can access?
11:18
se6astian
not sure...
11:23
vup
anuejn: maybe / possibly
19:47
aombk2
left the channel
19:47
aombk
joined the channel
20:07
aombk2
joined the channel
20:08
aombk
left the channel
22:52
anuejn
there is some more color-science at https://github.com/anuejn/cmv12k_color_response/blob/main/color_matrices.ipynb
22:52
anuejn
this incooperates filters and minimization of the color measured in a perception-matched way
22:53
anuejn
the results are already quite good and probably as far as I want to go for now
22:54
anuejn
if you have more ideas on what I should try, feel free to throw them at me
23:47
aombk
joined the channel
23:48
aombk2
left the channel