Current Server Time: 19:39 (Central Europe)

#apertus IRC Channel Logs

2014/02/25

Timezone: UTC


03:05
troy_s
left the channel
03:05
troy_s
joined the channel
03:39
Bertl
back now ...
05:16
troy_s
left the channel
05:16
troy_s
joined the channel
06:15
djp
hey, do you gus an gals not have a real-live bug-tracking system? (email doesn't count)
07:54
se6astian
joined the channel
07:55
se6astian
good morning
08:36
intracube
joined the channel
10:39
intracube
left the channel
10:42
SashaC
joined the channel
10:54
Bertl
djp: not yet, currently there are no bugs, only features :)
12:13
intracube
joined the channel
13:27
danieel
troy_s: a method to determine the native iso/ei which does not require expensive measurement devices would be handy
13:28
mars_
what measurement devices are needed?
13:58
SashaC
left the channel
14:11
se6astian1
joined the channel
14:13
se6astian
left the channel
14:20
intracube
left the channel
14:47
intracube
joined the channel
15:01
danieel
mars_: i would say a precise illumination sensor (either to measure the light or to adjust it)
15:02
danieel
that is what we do not have
15:12
troy_s
Bertl: The variation in the needs of frame lines is why I suggested TIFFs.
15:13
troy_s
Bertl: With "canned" aspects. The reality is often you are generating customs - some folks want a center dot, some a crosshair, some different aspects, some offset to upper third (think cropping)
15:13
troy_s
danieel: I think you can chart the native. :)
15:14
danieel
how to chart it?
15:14
troy_s
danieel: The crossover point where the EI is equal both up and down in latitude.
15:14
Bertl
troy_s: yes, I understand that, the other side of the coin is that overlaying an alpha blended image costs a lot of resources and thus will reduce the lifetime of the battery
15:14
danieel
explain more please
15:14
troy_s
Bertl: Gotcha.
15:15
troy_s
danieel: Most times adjusting the EI will only offset the latitude.
15:15
Bertl
for example, simply drawing a few horizontal and vertical lines or darkening a few areas doesn't cost anything
15:15
troy_s
Bertl: The problem is that it is more complex often
15:15
Bertl
that's why I'm asking :)
15:16
troy_s
Bertl: A center dot or crosshair, different crop regions (see upper third), and different opacities for the mask
15:16
Bertl
it might be an option to integrate it in the HUD and make it regional or optional
15:16
troy_s
Bertl: And of course monitor out gets a different set to EVF
15:16
Bertl
that's not a problem
15:16
troy_s
Bertl: The overlay data is a different function usually (for the types of data readout)
15:17
troy_s
Bertl: Go crawl over the Alexa and F65 docs.
15:17
troy_s
Bertl: Those menus and functions are the byproduct of an iterative loop.
15:17
Bertl
the process of overlaying images doesn't cost much in the FPGA resource wise, the problem is more with fetching an image from memory over and over
15:17
troy_s
Yep
15:18
troy_s
Anyways... important and slightly more complex than first glance.
15:18
Bertl
yes, ideed
15:18
troy_s
(what isn't)
15:19
mars_
danieel: well, i got an ambient light / spot meter if that helps
15:20
Bertl
mars_: does sebastian know this? :)
15:20
mars_
hehe, let me think
15:20
mars_
http://rofitv.files.wordpress.com/2010/02/sekonic.jpg
15:21
mars_
this one
15:21
danieel
Bertl: fetching the overlay is costly? (of bandwidth or of power)
15:23
Bertl
both, fetching from DDR memory consumes bandwidth and power
15:24
troy_s
danieel: I am sure Bertl can lend some insight into finding the native EI as well. Should be that crossover point of balanced latitude I believe.
15:24
danieel
latitude is dynamic range... how does that relate to base iso
15:25
troy_s
danieel: Because if you look at the charts for latitude on the f65 for examplr
15:25
troy_s
danieel: You will see it never really changes
15:25
Bertl
I'm cleaning up code and fixing a few tools which got broken during the latest updates (double buffered live view) but when I'm done, I'll look into all the calibration stuff
15:25
danieel
what does not change?
15:25
troy_s
danieel: But rather EI shifts the exposure range up and down
15:26
troy_s
danieel: So it always (within reason) stows the same scene referred range of values
15:26
troy_s
danieel: But shifts the amount down or up.
15:26
se6astian1
sorry back now, what might I know
15:26
se6astian1
btw troy I just shot a new chart series for you
15:26
troy_s
(so if you raise the EI, the number of stops down from middle grey goes up, but the number over goes down. ;))
15:27
troy_s
se6astian1: YOU ROCKS
15:27
se6astian1
https://cloud.gerade.org/public.php?service=files&t=88644302ffde4f6e5659627a3c8bbf32&download&path=/Axiom/alpha/daylight2014-02-25.tar.xz
15:27
troy_s
se6astian1: Will do it first thing this morning if you get them up.
15:27
troy_s
Sweet.
15:27
danieel
troy_s: so instead of getting a spot meter you suggest to get an f65? :)
15:28
troy_s
LOL
15:28
troy_s
danieel: No, I am sure Bertl can explain it better. ISO on film was larger crystals of silver where here it is just gain
15:29
troy_s
danieel: So the actual latitude is much more uniform than with silver
15:29
troy_s
danieel: And while analogous, the EI / ISO effectively doesn't adjust the latitude other than offset.
15:30
danieel
yes
15:30
troy_s
danieel: Magic Lantern just did a huge bunch of tests and plotted the EI
15:30
danieel
but we want to get to a fixed value, for unity gain
15:30
troy_s
danieel: Yes and I believe (again open to error) that the point where the latitude over / under paths cross
15:30
troy_s
danieel: Is native.
15:31
troy_s
So if you have 12 stops of latitude
15:31
troy_s
6/6 is nativr
15:31
danieel
ok... so we shoot a fade chart
15:31
danieel
and make different exposurses/gains.. and by shift we determine the base? not sure how that would work out
15:33
Bertl
the thing is, with the different gain and offset setting present in each sensor (not just the cmosis) it is rather complicated to 'calculate' an ISO equivalent for the sensor, you can only do it for a certain setting
15:33
Bertl
OTOH, that is probably all we need for the calibration part
15:33
troy_s
Bertl: Yes and EI / ISO is kind of damn important LOL
15:34
danieel
so with a spot meter, the process would be what?
15:34
danieel
shoot at 360deg shutter?
15:34
Bertl
I'm currently toying with an idea for a slightly different approach to the 'calibration problem'
15:34
troy_s
danieel: Shutter shouldn't matter
15:34
troy_s
danieel: Find your noise floor and blow out ceiling
15:35
Bertl
will get the old chart tonight, and will see if it makes sense to investigate
15:35
danieel
so set shutter to have a non-overexposed image on the spotmeter
15:35
troy_s
danieel: And expose up and down. When those are both hit, the neutral center is native (or darn close)
15:35
troy_s
Bertl: Any easy math way to calculate EI?
15:36
troy_s
Bertl: Or calibrated spot with an emissive patch?
15:37
troy_s
(and given that spots all set middle grey slightly different)
15:37
Bertl
which one :) http://www.wikiradiography.com/page/Exposure+Index
15:38
danieel
had anybody tested the actual EI of other cameras? e.g. if arri's 800 matches sony's 800 ? :)
15:38
danieel
on raw linear data
15:39
danieel
that would be the easiest to compare against - just see where our and their numbers are, for the same gray patch
15:39
troy_s
danieel: That floats I believe
15:39
troy_s
danieel: But 800 at f2.8 would be the same.
15:42
danieel
yes, the test should be using same lens aperture, ideally on studio light and fixed patch
15:59
Bertl
mathematically, the EI should be log2(F**2/t)
16:00
Bertl
where F is the f-stop, and t is the exposure time
16:01
Bertl
F**2 = F² = F^2 and log2 is the logarithm base 2
16:11
danieel
makes sense.. but the BASE ei has what base time?
16:11
troy_s
Bertl: In layman's terms, is that roughly what I described?
16:12
danieel
i think bertl described ei as the base iso multiplier
16:13
Bertl
the exposure index, as far as I know, doesn't depend on the sensor/film
16:14
troy_s
Incoming new photos at 108 megs.
16:14
troy_s
Bertl: No it doesn't.
16:14
troy_s
Bertl: That's the baseline and why it is rather important.
16:14
Bertl
i.e. it is simply how much light will hit the sensor/film at a given exposure/f-stop
16:15
troy_s
Would be great if there were a few more hacks that can code in here.
16:16
Bertl
the sensor itself has a spectral response, which is documented as QE curve in the datasheet
16:17
troy_s
Bertl: If we knew the primaries of the filters, we could get a transformation matrix. :)
16:17
Bertl
depending on the energy of the photon, the sensor will convert a certain number of photons to electrons
16:31
troy_s
se6astian1 / Bertl: are you both on the same firmware? It seems sometimes I get unflipped images?
16:39
Bertl
we are currently investigating this, but it doesn't really depend on the firmware, it depends on the sensor setting and the userspace tools
16:39
Bertl
but we will know shortly
16:39
troy_s
Bertl: Gotcha.
16:40
troy_s
Bertl: It was odd that I was having to flip flop the basic demosaics
16:49
Bertl
so I just checked, with Y-flipping set (the default)
16:50
Bertl
the debayer algorithm on the wiki (imagemagick) for flipped images gives the correct colors
16:50
troy_s
Ugh
16:50
troy_s
Bertl: So I've started in here... the general trend is that the error goes up as the exposure is going up.
16:50
troy_s
Bertl: Which sucks because it means that the response off the sensor is likely prohibiting our efforts to get a low DE.
16:51
troy_s
(as in irregularity or?)
16:51
Bertl
I have a theory that we are getting IR light despite the fact that we are using an IR/UV cut-off filter
16:51
troy_s
Bertl: Why is that?
16:51
Bertl
which might be especially visible in the red channel
16:51
Bertl
bad filter?
16:51
troy_s
Ah.
16:51
troy_s
Not a bad theory.
16:51
troy_s
Just looking at the response of the TI3
16:52
troy_s
Even in > D50, the red channel is really grabbing on.
16:52
troy_s
Obviously the data values are arbitrary, but I'd expect their red filter to not be glomming on (on the photosites) quite as much in that relatively blue light.
16:52
Bertl
if you look at the spectral response, the red channel can handle all the range from 570nm up to 850nm
16:53
troy_s
Bertl: Sure, but equal to the green filter in blue light?
16:53
troy_s
Bertl: I looked at that chart... it just seems that the red is heavily overweighting the values.
16:54
troy_s
Bertl: se6astian1's shot the first time around was probably 6000-6500 and the red and green channel were equal, with blue down probably a stop.
16:54
Bertl
green and blue is significantly less sensitive in the 'lower' IR range
16:54
troy_s
Bertl: Seems odd
16:54
troy_s
Bertl: Any way to test your theory?
16:54
Bertl
we are already testing it with an IR filter (together with the cutoff)
16:55
troy_s
Bertl: The good news is that our DE2k is down to 8.09
16:55
troy_s
Bertl: With our max being 23... let me post the best results I got
16:55
troy_s
Bertl / se6astian1 Here's the 16 ms profcheck results http://www.pasteall.org/49827
16:56
troy_s
Bertl / se6astian1 And here's the TI3 http://www.pasteall.org/49828
16:56
troy_s
(The 16ms (the lowest se6astian1 shot) yielded the best results, which is _entirely_ odd given that the RGB values are being read at 77-78 maximum.)
16:57
troy_s
So my theory that exposure was to blame at this point is not correct.
16:58
troy_s
What's _really_ odd Bertl is that in the GS row (the achromatics) the colors vary wildly.
16:58
troy_s
Bertl: Any idea?
16:58
troy_s
(And by wildly I mean the white box (00) is off by a notable amount compared to say, the box adjacent it, etc.
16:58
Bertl
'the color' means?
16:59
troy_s
Bertl: The data values in terms of achromatic balance shifts dramatically.
16:59
troy_s
Bertl: Example if using 'color' terms (obviously not) the first white box is pinky, the next is pinky, the next is yellow, then directly adjacent is green, then pink again, then green.
17:00
troy_s
So the sensel capturing across the achromatic strips is varying left to right significantly dramatically.
17:00
Bertl
okay
17:00
Bertl
sounds indeed odd
17:00
troy_s
Bertl: Look at the profcheck
17:00
troy_s
Bertl: The first link.
17:00
troy_s
Bertl: And look at rows GS00 to 23 (very bottom)
17:01
troy_s
The DE is very high (in red channel I'd again say) for 00 and 01
17:01
troy_s
the de is only correct-ish in the middle exposure range (10, 11, 12)
17:02
troy_s
So as we increase and decrease away from our 'middle exposure' zone, our deviation appears to be flying off the charts.
17:03
troy_s
Bertl: Also note all of the 13s
17:03
troy_s
That's the blue strip
17:03
troy_s
(top to bottom goes A through L in 13)
17:03
troy_s
Bertl: Notice a trend?
17:03
Bertl
well, it doesn't look like the color changes much
17:03
troy_s
Bertl: Look at the profile check
17:03
troy_s
Bertl: All column 13
17:04
troy_s
Bertl: That's the blue strip on the IT8
17:04
troy_s
Bertl: And remember how I said the DE is lowest on the neutral greys at middle exposureish?
17:04
troy_s
Bertl: Look as we start BRIGHT in the bluey cyans
17:04
Bertl
I'm looking at the GS results ATM
17:04
troy_s
The DE is off the chart
17:04
Bertl
and they seem to have a consisten color shift
17:04
troy_s
the only time it gets into a viable range of 2.++ ish
17:04
troy_s
is as we get down to L
17:04
Bertl
except for GS00-02
17:04
troy_s
(which is somewhere closer to middle grey exposure)
17:05
troy_s
The shift is very uniform from middle grey across the board
17:05
troy_s
and underexposed
17:05
troy_s
same thing
17:05
troy_s
Which we can expect, except the matrix doesn't fit the real world (as in errors)
17:05
troy_s
So something is going on at all non-middle grey exposure values.
17:05
troy_s
Notably on that blue channel.
17:05
troy_s
(it is _way_ off)
17:05
troy_s
DE 20+ is ... uh
17:05
troy_s
Whack.
17:06
troy_s
(Bertl using calc or a spreadsheet with spaces as the column breaks works wonders)
17:09
troy_s
Bertl: So blue is notably out of whack, which show sup on the yellow, blue, and neutral swatches quite dramatically.
17:10
troy_s
And notably in the higher end of exposures and the lower end.
17:10
troy_s
(seems to deviate MUCH more in those ranges)
17:10
troy_s
Bertl: Any idea why? Does that play into your guess about the IR?
17:13
Bertl
not completely, but you never know, I also suspect irritating results from incorrect offsets
17:14
troy_s
Bertl: Worst swatches are J22, A16, F03
17:14
Bertl
for example, with the cap on, the default offset setting gives a peak around 250/4096
17:14
troy_s
Which are low end of exposure, high, low respectively.
17:15
Bertl
and most importantly, on this peak all four channels converge
17:15
Bertl
i.e. any difference resulting from IR exposure in red, will be eliminated at the low end offset
17:15
Bertl
probably resulting in quite a strange profile
17:16
troy_s
Bertl: http://www.pasteall.org/pic/show.php?id=67408
17:16
troy_s
Bertl: That's the diagnostics chart (for the 24ms version as an aside) so you have chart values to look at.
17:17
Bertl
the gray GS23 supports the offset theory
17:18
Bertl
the color changes on GS00-03 are probably due to the low chromatic component in bright gray swatches
17:18
troy_s
Bertl: Explain?
17:18
Bertl
so the twist in the profile kicks in when reconstructing
17:19
Bertl
the chromatic component is very small compared to other fileds
17:19
troy_s
(Side note to all passive onlookers, please do not consider the "color" in that diagonstic image as being sRGB. It isn't. It is arbitrary and undefined. What you are seeing is nothing more than raw data being dumped to sRGB)
17:19
Bertl
*fields
17:19
Bertl
so any error is amplified there and gives a tint in one or the other direction
17:19
troy_s
Bertl: There's something going on.
17:20
troy_s
It shouldn't be that out of whack. Not in a million zillion years.
17:20
troy_s
I mean you can see the variation even in that sRGB view of the raw.
17:21
troy_s
Bertl: And look at the TI3 data for the GS row (the achromatics)
17:21
troy_s
Bertl: Look how retarded the blue channel appears.
17:22
troy_s
Bertl: Or rather, look how hyper the red channel is?
17:22
troy_s
I would expect red to be the most regressed / retarded of the lot due to the luminance in red
17:23
troy_s
but for some reason it seems to be getting hyper excited when we drift from middle grey exposures.
17:23
troy_s
Bertl: Let me do the 'best exposure' and see how it shifts.
17:25
troy_s
se6astian1: By the way, fantastic exposure bracket for this lot.
17:25
troy_s
se6astian1: Really great range.
17:27
troy_s
Bertl: Here's 28ms (with channels peaking at around 92/93 RGB)
17:27
troy_s
http://www.pasteall.org/49830
17:28
troy_s
Bertl: and the TI3
17:28
troy_s
http://www.pasteall.org/49831
17:28
troy_s
(Well high is 94+)
17:28
troy_s
Bertl: Notice the error on G03 here... 1.63
17:29
troy_s
That's decent.
17:29
troy_s
F17 however? 40.
17:29
troy_s
Hell F03 is 1.01!
17:29
troy_s
Sorry reading wrong damn values.
17:29
troy_s
Anyways, look at the DE readouts (the profcheck) on that more ideal exposed chart.
17:32
troy_s
Bertl: Lowest DE in that according to a hasty sort is E20.
17:32
troy_s
(A red swatch)
17:37
troy_s
Bertl: Anyways. I'm done. Any ideas as to what is going on?
17:39
Bertl
we will get some clues shortly
17:39
Bertl
a number of IR filters are on the way, they will arrive in an hour or so
17:40
Bertl
I have some tests to do to clarify what happens in the IR range
17:40
Bertl
(if anything happens there :)
17:40
troy_s
Hrm. Interesting.
17:40
Bertl
and I will also investigate and correct the offsets
17:40
troy_s
But you and se6 are in different regions, so it is unlikely to test with proper charts?
17:40
troy_s
What are the offsets?
17:41
Bertl
more importantly it is already dark
17:41
troy_s
The per-channel offsets?
17:41
troy_s
LOL
17:41
Bertl
so you won't get any day-light charts in the next twelve hours :)
17:41
troy_s
The daylight isn't THAT big of an issue.
17:41
troy_s
What the daylight chart revealed was that we have a lower level issue.
17:41
Bertl
there are two ADC offsets, one for the top ADCs and one for the bottom ADCs
17:42
troy_s
You should be easily able to get a profile under tungsten, but not with the things being out of whack as they are.
17:42
troy_s
Per channel?
17:42
Bertl
nope
17:42
troy_s
Hrm. So what would adjusting those offsets do?
17:42
Bertl
they definitely introduce a nonlinearity
17:42
troy_s
Bring the skew / fanning of the three channel curves into alignment?
17:42
troy_s
Gotcha.
17:42
troy_s
That makes good sense.
17:43
troy_s
So how do you dial that in?
17:43
Bertl
with the filters, I also get the 'old' IT8.7 chart back
17:43
troy_s
The scanner chart.
17:44
Bertl
precisely
17:44
Bertl
should work for this purpose if I can get the lighting right
17:45
Bertl
I also have a neutral gray here somewhere
17:45
troy_s
Bertl: The neutral grey should at least get you a good reference point on the over / under exposure firings.
17:45
troy_s
That might help to dial in those ADC offsets?
17:46
Bertl
the offsets should be simple to adjust with the GS chart from the IT8.7
17:46
troy_s
You can probably use the error as a guide, as I think that is a linear error.
17:46
troy_s
(not the DE)
17:50
Bertl
I can do a bunch of tricks with the gamma lut
17:50
Bertl
i.e. use special 'markers' in the lut profile to adjust settings with the live view
17:50
troy_s
How does that help here?
17:51
Bertl
for example, have a peak in the red channel at 50%
17:51
troy_s
I mean viewing output can be a 3D LUT and could likely correct for any mess
17:51
troy_s
But in reality, that completely botched response is affecting image quality significantly.
17:51
troy_s
So the root issue is more important than cheating something closeish.
17:51
Bertl
yes, I'm not talking about correction, I'm talking about detection
17:51
troy_s
?
17:52
Bertl
let me give an example
17:52
troy_s
So use the liveview as a diagnostics device of some sort?
17:52
Bertl
yes
17:52
Bertl
you know what a gauss curve is?
17:52
troy_s
Nope.
17:53
troy_s
And the good news is I am stupid.
17:53
Bertl
okay, no problem, let's say we use a curve which is low (0) almost everywhere, except around 50% of the full red range
17:53
troy_s
Bell curve? Parabola?
17:53
Bertl
around 50%, it suddenly gets bright up to the full component
17:54
Bertl
bell shaped curve is the one I mean
17:54
Bertl
this doesn't affect the image capture at all, but it will visually highlight the areas at 50% red
17:54
troy_s
Ah gotcha.
17:54
Bertl
(as perceived from the sensor)
17:54
troy_s
So sort of a false-color readout.
17:55
Bertl
yes
17:55
troy_s
False colors are handy as hell.
17:55
Bertl
I can do the same for the high and low end as well
17:55
Bertl
and of course, I can also apply it to all channels
17:56
Bertl
with the 4x4 matrix we implemented, I can also mix all the channels together and use this
17:57
Bertl
so there are a number of options to simplify adjusting various things with help of the live view
17:57
Bertl
like for example even illumination :)
17:58
troy_s
Well whatever. Get it fixed in the next twenty minutes. kthxbai.
17:58
troy_s
(lulz)
18:01
Bertl
hehe, will do :)
18:19
se6astian1
left the channel
18:20
se6astian
joined the channel
18:21
ThatCantBe
left the channel
18:23
ThatCantBe
joined the channel
18:23
ThatCantBe
changed nick to: 20WABDAGN
19:16
jucar
left the channel
19:33
jucar
joined the channel
19:43
Bertl
off for a nap ... bbl
19:53
sb0
joined the channel
19:54
sb0
hey
19:54
sb0
met you guys at the photonics congress in Berlin last week. I'm the Milkymist/EHSM founder
19:58
se6astian
Hi! welcome Sébastien
19:59
se6astian
feel free to hang out
19:59
se6astian
we have logs available on irc.apertus.org as well
21:28
troy_s
Bertl: Done yet?
22:20
se6astian
time for bed..
22:20
se6astian
good night
22:20
se6astian
left the channel
22:45
FergusL
oh hi sb0, didn't expect seeing you here, I'm FergusL and I hang around on #leloop and other channels