Current Server Time: 10:14 (Central Europe)

#apertus IRC Channel Logs

2022/02/05

Timezone: UTC


00:53
balrog
left the channel
01:30
aombk
left the channel
01:32
aombk
joined the channel
04:39
balrog
joined the channel
05:54
Bertl_oO
off to bed now ... have a good one everyone!
05:54
Bertl_oO
changed nick to: Bertl_zZ
08:30
se6astian
good da
08:30
se6astian
y
09:22
vup
some more flatfield analysis
09:23
vup
there seem to be a lot of pixels that get very low pixel values and show very nonlinear response curves in that region
09:23
vup
https://files.niemo.de/corr_1.9932959999999997_by_13.993392.pdf
09:24
vup
https://files.niemo.de/corr_2.9964_by_13.993392.pdf
09:24
vup
https://files.niemo.de/corr_6.996432_by_13.993392.pdf
09:25
vup
(The flatfields for each exposure is a average of 256 flatfields)
09:27
vup
The plots show a 2d hexbin histogram of the original pixel value on the x axis and the ratio between the original pixel value divided by the value of that pixel for the 13ms exposure
09:28
vup
If the response curve of all pixels were the same we would expect no correlation between x and y
09:28
vup
But there clearly is
09:29
vup
(Or atleast no correlation if the response curve were linear)
09:38
vup
Any thoughts?
09:39
vup
I guess we will see more when se6astian has time to capture more flatfields
09:39
vup
All in all it looks like flatfield calibration will be very important atm
11:07
se6astian
Great progress
11:07
se6astian
I am in Linz currently demanteling the kitchen of the mother in law
11:08
se6astian
Will capture more next week
11:49
anuejn
vup: veery interesting
11:50
anuejn
looks kinda bad though
11:50
anuejn
I hoped everything would be much more linear
11:51
anuejn
though I dont know how to interpret these plots exactly
13:37
Bertl_zZ
changed nick to: Bertl
13:37
Bertl
morning folks!
13:43
vup
anuejn: well i don't think its bad really
13:44
vup
Looks pretty easy to model actually
13:44
vup
For exactly linear you would expect the same y for every x
13:44
vup
(If you double the exposure, the value should double)
13:45
vup
And it seems like this actually is the case for many pixels
13:45
vup
(Atleast roughly)
13:46
vup
Oh also the per color version of this is pretty interesting
14:06
Bertl
hmm, it looks like half of the sensel have one factor and the other half a different one, no?
14:10
vup
ah whoops
14:11
vup
these were already just one color
14:11
vup
i updated them inplace to show all pixels and each of {even,odd} rows, {even,odd} columns
14:11
vup
Bertl: well not really, there seem to be pixels that are more sensitive, that seem to have a approx linear factor
14:13
vup
and then there are pixels that are less sensitive, that seem nonlinear, maybe exponential? (ie doubeling the original pixel value seems to double the factor between the exposrues
14:13
vup
anuejn: ^ updated them implace with per color plots
14:16
anuejn
thats interesting
14:23
Bertl
vup: hmm, is there a pattern to the linear vs the nonlinear one? what's the ratio between the 'linear' and the 'nonlinear' ones?
14:23
se6astian
Do you consider an offset image and a factor image is still the best way for compensation?
16:34
vup
Well darkframe / offset image seems definitely like a good choice
16:34
vup
Maybe with a constant and a nonuniform (exposure scaled) part
16:35
vup
But for linearization it seems like a simple factor probably wont cut it
16:36
vup
Bertl: the obvious pattern is that the nonlinear ones seem to only occur in odd rows
16:36
vup
The rest ill look into later
16:48
sidsh
joined the channel
20:21
sidsh
left the channel