Current Server Time: 09:39 (Central Europe)

#apertus IRC Channel Logs

2014/01/01

Timezone: UTC


23:12
troy_s
Greets all.
00:14
fuzzydolphin
joined the channel
00:41
fuzzydolphin
left the channel
05:17
Bertl
troy_s: ad linking, sometimes the order of libraries is important
05:18
Bertl
i.e. it might help to add -lpthread a second time at the end for example
05:20
Bertl
ad lut transformation, I think it is a good approach to have the LUT (3x3 + 3) implemented in C/C++ so that we can test it without actually requiring the prototype, and it should be as flexible as well (i.e. we can certainly load the LUTs from files)
05:21
Bertl
(i.e. use it without recompiling)
05:23
Bertl
for the OIIO I still think that a 'generic' raw importer/plugin would be the best way to go, it can start as version 0.1 with just support for the axiom raws, but it should be possible to extend it to more formats as we go
05:23
Bertl
like for example changing the bitdepth from 16 to 12/10/8
05:23
Bertl
(we will soon switch form .raw16 to .raw12 to reduce the filesize
05:23
Bertl
)
05:24
Bertl
and maybe different heights as well (we might want to reduce the height to grab more frames)
05:27
jucar
joined the channel
06:00
jucar
left the channel
08:15
Bertl
off to bed ... have a good one everyone!
09:35
se6astian
joined the channel
09:45
se6astian
good morning!
09:55
mars_
morning!
09:59
se6astian
happy new year! :)
15:27
gcolburn
joined the channel
15:27
gcolburn
hello
16:15
se6astian
hi gabe
16:17
[1]se6astian
joined the channel
16:19
gcolburn
Thanks for fixing the typo in the makefile
16:19
gcolburn
I assume after that you can get it to run?
16:19
se6astian
left the channel
16:19
[1]se6astian
changed nick to: se6astian
16:27
se6astian
yes, it works great
16:27
se6astian
it uses a tungesten matrix right?
16:27
gcolburn
good. I think the LUT is in a format hebert can use
16:27
gcolburn
yeah. I can't remember if it was using the images you took last week or not
16:27
gcolburn
but it should be tungesten
16:30
gcolburn
are you using the new firmware?
16:41
se6astian
yes, we set it up together 2 days ago
16:42
se6astian
works great so far
16:42
se6astian
I will capture a new set of IT8 images as exposurerow as troy suggested later today
16:42
se6astian
anything else you need captured?
16:45
gcolburn
nope
16:46
gcolburn
I was just curious if you've been able to see if the image quality seems to have improved
16:53
se6astian
hmm, hard to say,
16:53
se6astian
I never had the artifacts caused by training data missalignment
16:53
se6astian
as we had them on the other sensor
16:54
se6astian
I guess that is fixed as well now
16:54
se6astian
btw we can capture up to 7 frames at full FPS
16:54
se6astian
as much as the RAM can hold
16:55
se6astian
so if we reduce the number of read out lines to roughly a third (1080 lines) we can maybe store around 20 frames in FullHD
17:02
gcolburn
great
17:02
gcolburn
but for streaming over HDMI we don't need to store all those frames simultaneously right?
17:22
se6astian
no, HDMI is a different topic afaik
17:47
troy_s
If you can get a sun lit / no sky IT8 (no glare) that would be excellent.
17:53
se6astian
not easily I am afraid, no sunny weather here and I am bound to my desk with the prototype cables until the battery pack solution is ready
17:54
se6astian
I have the kinoflo and color temperature meter here still though
17:55
Bertl
morning everyone!
17:56
Bertl
se6astian: I need a new run of the noise5 with more light, maybe also point the camera at a white area
17:57
Bertl
for the frame sequence, I would even consider 768, as we will still get 4096 pixels horizontally
17:58
Bertl
i.e. some kind of really really wide format, which would give about a second of footage
17:58
Bertl
(but that needs a minor change to the snap2, nothing critical though)
18:07
se6astian
morning Bertl
18:08
se6astian
noise5, will do!
18:08
Bertl
thanks! appreciated!
18:09
se6astian
are you profiling FPN or what kind of noise compensation are you working on atm?
18:09
Bertl
both, FPN and the row noise as they are entangled
18:10
Bertl
when testing your ADC cutoff issues, I found that a major part of the noise we saw is still in the image
18:11
se6astian
I see
18:11
Bertl
i.e. the voltages seem to clip somewhere, but the digitization is in the active range, this can be used to extract FPN without any need for uniform lighting
18:13
Bertl
the current setup is probably suboptimal (the register settings, but I haven't searched for the optimal setup yet and I want to see if that theory actually holds first)
18:15
se6astian
should I apply the row noise reduction fix as suggested by cmosis by setting the sun dark spot protection lower? or is that already in your noise5.reg file?
18:16
Bertl
for now, don't change anything, it's just a test
18:16
se6astian
ok
18:16
Bertl
if you like, you can experiment with enhancing the FPN pattern over the row noise pattern
18:17
Bertl
the basic indicator is the following:
18:17
Bertl
run a single frame through convert -normalize and look for uniformity across the whole image
18:18
Bertl
i.e. the edges should have the same gray and different color channels should not be visible
18:18
Bertl
you should also see the FPN emphasiszed and the dynamic row noise appearing 3-4 times over the whole image
18:18
Bertl
(for each frame at a different position)
18:51
troy_s_
joined the channel
18:53
rexbron
joined the channel
18:53
rexbron
left the channel
18:53
rexbron
joined the channel
18:58
rexbron_
left the channel
18:58
troy_s
left the channel
19:12
se6astian
Am I right that the best method to compare the row/FPN noise results is to use very short exposure times to reduce thermal noise influence and to use darkframes with closes lens cap to only see noise and no scene content?
19:12
se6astian
40µs seems to be the shortest possible exposuretime
19:13
Bertl
I presume the strange character before the 's' is a µ i.e. you mean us?
19:14
se6astian
yes :) damn my irc client
19:15
Bertl
yes, in theory we can get lower, but I don't think that it will be necessary/useful
19:15
Bertl
OTOH, as recent tests have shown, at least part of the FPN can be seen in extremely long exposures (reaching the zero point) as well
19:16
Bertl
so it really depends on what noise/offset/distortion we want to capture :)
19:19
se6astian
still cameras can go as low as 1/8.000s = 125us with shutter time, so with 40us I think we are very good already anyway
19:23
se6astian
one issue with using convert -normalize: wouldnt that hide any effects of lowered row noise when changing register as it will just scale up values higher?
19:27
gcolburn
left the channel
19:33
gcolburn
joined the channel
19:44
Bertl
yes, of course, normalize will only show the most prominent noise over the whole range
19:44
Bertl
i.e. if the FPN noise is very large, row noise will almost disappear
19:45
Bertl
for calculations we need the absolute values anyway, but for inspection normalize is a good idea to see what noise type is the significant one
20:05
jucar
joined the channel
20:18
guesst
se6astian: thermal noise in your GS mode makes still significant part (66ms in last readout row?)
20:20
philippej
joined the channel
20:31
Bertl
in CMOS circutiry with GS the thermal and 1/f noise happens at the amplification/readout
20:33
Bertl
the readout itself happens with 32 channels @ 300Mhz (at the moment)
20:33
Bertl
so it will take roughly 60-80us
20:34
Bertl
7.8ms for the entire band
20:50
guesst
are you sure? in my experience, the image degrades while waiting for readout... maybe i should not call it noise, rather thermal generation
20:51
Bertl
yes, it will degrade a little, but that is probably not even measureable unless you have really really long readout times
20:51
Bertl
i.e. the noise at readout/amplification will probably be a few magnitudes higher
20:51
guesst
66ms is wrong by me, it should be 1/150 (150fps) right?
20:52
guesst
6.66 ms
20:52
Bertl
yep, that's the readout time (roughly) regardless of the exposure time
20:52
Bertl
I also think that the FOT can be considered an idle time
20:53
Bertl
i.e. the CMV12k sensor doesn't do anything relevant during FOT
20:53
guesst
hard to tell.. the docs are incomplete
20:53
Bertl
agreed
20:54
Bertl
we will ask cmosis for more details on this subject in the near future
21:09
philippej
Happy new year everyone !
21:42
philippej
left the channel
21:58
jucar
left the channel
22:27
se6astian
good night!
22:28
se6astian
left the channel