Current Server Time: 23:08 (Central Europe)

#apertus IRC Channel Logs

2022/10/13

Timezone: UTC


01:10
aombk2
joined the channel
01:11
aombk
left the channel
02:42
aombk
joined the channel
02:44
aombk2
left the channel
08:28
anuejn
considering ir cut filters basically UQG gives us the best results by far
08:49
se6astian
very interesting!
08:50
se6astian
and great that we have a foundation now to rate the UV/IR filters
08:54
se6astian
is there also method to compare these theoretical values to the actual images once max delivers the first set
08:55
se6astian
to further fine tune or confirm the model
08:58
vup
probably but its not super easy
08:59
vup
or rather comparing is easy
08:59
vup
instead of inserting the simulated pixel values for the color checker patches one would just insert the actually observed ones
09:00
vup
(with some details like normalization of these pixel values that needs to be figured out, but also the color matrix is exposure independent, so that should not make a differentce)
09:01
vup
fine tuning the model would mainly consist of trying to improve our idea of the spectral response of the sensor, which I think would be very tricky to do with just 24 patches + some uncertainties about the exact spectrum of the illumination
09:08
vup
anuejn: afaik the ΔE value you are optimizing only is a proxy for the chromaticity difference, not the luminosity one, right? I am not really sure what is the "most desirable" thing to optimize, but I would have assumed we want to minimize both chromatic and luminosity distance, right
09:09
vup
also, whats the reason for not specifing the illuminiant when converting to lab values?
09:11
anuejn
vup: is it?
09:11
anuejn
if i read the CIEDE2000 paper right, it also considers chromaticity?
09:15
vup
anuejn: ah right, I misread some part somewhere
12:18
McUles[m]
left the channel
12:18
davidak[m]
left the channel
12:18
underpantsgnome[
left the channel
12:25
McUles[m]
joined the channel
12:45
davidak[m]
joined the channel
12:45
underpantsgnome[
joined the channel
15:01
McUles[m]
left the channel
15:01
vup
left the channel
15:01
illwieckz
left the channel
15:01
fredy
left the channel
15:01
fredy
joined the channel
15:02
vup
joined the channel
15:03
illwieckz
joined the channel
15:03
McUles[m]
joined the channel
16:00
Guest8391
joined the channel
16:01
Guest8391
left the channel
16:28
illwieckz
left the channel
16:41
illwieckz
joined the channel
20:29
anuejn
se6astian: which filters do we give out with the devkits?
20:30
anuejn
we probably also want a correction matrix for these
20:41
vup
maybe, but one still probably needs to do calibration with the specific lens one uses anyways, no?
20:41
vup
well actually the effect is not too bad: https://f.coroot.de/20221013_21h41m11s_grim.png
20:42
vup
would be interesting to test how ΔE degrates if we simulate these lenses
20:42
vup
(ref: https://sci-hub.se/https://doi.org/10.1117/12.2233723)
20:42
vup
if anybody else know sources for lens spectral response function lmk
21:16
se6astian
Haida Pro II MC Digital Slim UV / IR 390 / 750 Filter 77mm
22:35
anuejn
vup: interesting :)
22:35
anuejn
se6astian: thanks :)
22:36
anuejn
but you didnt by accident throw one of these into a spectrometer too?
22:36
anuejn
huh oh wow these go pretty high
22:37
anuejn
that might be a bad thing
22:44
se6astian
I can plan to measure one if it's not in the list of measurements yet
22:47
anuejn
it is not sadly
22:47
anuejn
or i didnt find it
22:47
anuejn
if it is the haida filter we have a measurement
22:47
anuejn
but i am not sure if it is this one?
22:47
anuejn
can you confirm or deny that? do you still remember ;)
22:48
vup
how do the filters in a camera with integrated filters look like?
22:54
davidak[m]
i've also been exploring color science lately. i'm fascinated by the idea of using an lut to recreate the colors of an arri alexa on a cheaper camera. if the axiom offered that, as a "cinematic" preset, it could be very popular.
22:54
davidak[m]
here is my research so far: https://pad.nixnet.services/XDgdTJERQv6fr9Iu79iSmQ?view
22:56
davidak[m]
especially check out the perspective of Steve Yedlin in the theory section on color science. i think that fits the open nature of the apertus project very well and we might should use that in our communication. we can also offer the tools to create a custom look
23:17
davidak[m]
have we decided what our main intended audience is yet? is it professional film like hollywood or small budget indie productions? the latter can be intrigued by the possibility to have the ARRI look with a fraction of the cost while the first will want to match the colors to an ARRI camera. maybe they would use our camera as a small throw-away camera that can get destroyed in action scenes. it does have the benefit there that the recorder is
23:17
davidak[m]
modular and not attached to the sensor, so you can fully destroy the sensor without loosing the recording :)
23:36
vup
well, doing proper color calibration and providing a way to get the color of the pxilse in some well defined color space is a precursor to both of these
23:52
aombk2
joined the channel
23:55
aombk
left the channel
00:01
aombk
joined the channel
00:03
aombk2
left the channel