Current Server Time: 21:32 (Central Europe)

#apertus IRC Channel Logs

2014/02/11

Timezone: UTC


23:31
gwelkind
left the channel
23:53
gcolburn
joined the channel
00:12
danhanes
joined the channel
00:46
gcolburn
left the channel
02:02
troy_s
left the channel
02:02
troy_s
joined the channel
02:03
gwelkind
joined the channel
02:10
tyrone_
left the channel
02:59
gwelkind
left the channel
03:19
gwelkind
joined the channel
03:33
danhanes
left the channel
03:48
gcolburn
joined the channel
04:25
gwelkind
left the channel
04:33
gcolburn
left the channel
06:37
se6astian
joined the channel
06:37
se6astian
good morning
06:37
Bertl
morning se6astian!
06:38
se6astian
how did the night go?
06:38
Bertl
quite well, actually
07:31
philippej
joined the channel
07:32
philippej_
joined the channel
07:32
philippej_
left the channel
07:32
philippej
hi all
07:32
philippej
reading past discussions, I think it would be a good idea to start a workflow documentation
07:34
se6astian
morning philippe
07:35
philippej
homemade raw vs dng, single frame per file vs stream, what is done in camera, what is done outside, what is the expected format, what is the less painful way for users, what covers the most used workflow (and what IS the most used workflow)
07:36
philippej
morning se6astian
07:37
philippej
and also, are we talking about testing the workflow for the prototype or the finished camera, what are aiming at?
07:37
SashaC
joined the channel
07:42
se6astian
I have not had time to follow the workflow discuss tbh
07:42
se6astian
hi sasha
07:42
se6astian
*discussion
07:42
SashaC
hi Sebastian, no worries
07:42
philippej
shall we start a goole doc or wiki page?
07:45
se6astian
google doc is quicker I guess
07:45
philippej
ok
07:45
philippej
let's make the ball rolling
07:46
philippej
https://docs.google.com/document/d/1ydH8KhJAbhChMuiSXfOjX8U6lD0tnyxaZa6p3SagWkw/edit?usp=sharing
07:53
se6astian
may I interrupt for a second :)
07:53
se6astian
we have lice HDMI output running !!!!
07:53
se6astian
*not lice though
07:53
se6astian
its rather live!
07:53
mars_
\o/
07:54
Bertl
lice = live + nice :)
07:54
se6astian
haha :)
07:54
philippej
hohooo !
07:54
Bertl
great! how did you do it? :)
07:55
philippej
share something :-)
07:56
se6astian
Bertl, I have no idea how I did it, it was just there somehow, but I know a magician who might once again had been playing tricks on me :D
07:57
se6astian
will snap a picture :)
08:02
philippej
how does it look ?
08:02
se6astian
the colors are unprocessed atm
08:02
se6astian
but for that I would say it looks rather good :)
08:02
SashaC
Congratulations!
08:04
se6astian
its all due to Bertls incredible commitment (did I mention magic skills yet?) !
08:04
se6astian
https://cloud.gerade.org/public.php?service=files&t=88644302ffde4f6e5659627a3c8bbf32&path=/Axiom/alpha/live-hdmi.jpg
08:06
SashaC
Hey it's our favourite hungry caterpillar ;)
08:06
mars_
you should create a small movie ;)
08:10
philippej
congrats Bertl !
08:14
se6astian
its so weird that I am the only person who can really see this now and there is no proper way to share it
08:14
se6astian
what we could do though is a Google hangout with live video
08:16
philippej
yes!
08:17
se6astian
give me a few minutes to move all the stuff in position here :)
08:17
SashaC
can't wait :D
08:20
[1]se6astian
joined the channel
08:21
se6astian
left the channel
08:21
[1]se6astian
changed nick to: se6astian
08:21
se6astian
OK I think I am ready
08:23
se6astian
who wants to be invited ?
08:30
Bertl
off to bed now ... have a good one folks!
08:30
SashaC
Night Bertl, and fantastic work once again!!
08:36
se6astian
yes sleep well, you earned it ;)
08:36
philippej
good night man!
08:39
se6astian
you snapped some pictures, does google save them somewhere automatically?
08:41
philippej
I guess
08:41
philippej
I don't know where
08:42
philippej
but at least for history it' snice
08:42
SashaC
agreed
08:45
se6astian
please share them in google drive when you find them
08:55
se6astian
thanks
08:56
se6astian
so sasha I guess we need a new article :)
09:09
SashaC
I'll be able to start writing it up in just over an hour from now
09:10
SashaC
(Busy working on other stuff at the moment)
09:12
SashaC
I think we'd benefit very much from uploading a 10-15 second clip to Vimeo and then embedding it in the accompanying article
09:16
se6astian
I will have the HDMI recorder tomorrow I guess
09:17
se6astian
ok will twitter briefly about the news now
09:17
se6astian
full article can then follow in the next couple of days
10:08
philippej
left the channel
10:10
philippej
joined the channel
11:19
SashaC
left the channel
11:19
SashaC_
joined the channel
11:20
philippej
left the channel
11:25
philippej
joined the channel
12:47
philippej
by the way, se6astian what hdmi recorder will you use?
12:59
se6astian
atomos samurai
13:44
gcolburn
joined the channel
13:44
gcolburn
left the channel
13:52
SashaC_
left the channel
13:53
troy_s_
joined the channel
13:53
troy_s
left the channel
14:28
philippej
left the channel
15:29
troy_s_
changed nick to: troy_s
15:30
troy_s
Workflow
15:36
troy_s
se6astian: Very cool.
15:36
troy_s
se6astian / Bertl Has there been any progress on the matrix and 1D LUT for (starting) 709 output?
16:00
se6astian
thanks
16:00
se6astian
not sure about LUT, thats more gabe and herberts area
16:00
troy_s
se6astian: Updated that shared doc
16:01
se6astian
great, thanks
17:15
Bertl
morning folks!
17:16
se6astian
hello our hero of the day!
17:17
troy_s
Morn Bertl
17:17
se6astian
actually more hero of the year already :)
17:17
se6astian
anyway, gotta start cooking dinner
17:17
troy_s
Q from above: A) Where is the matrix currently used for profiling?
17:17
troy_s
Q2) Has a transform been generated for going from the camera profile to 709? (And likewise, has a 1D LUT been generated and applied via the board for the transfer curve?)
17:18
Bertl
in the current setup, there is only direct output of the linear rgb values
17:18
gwelkind
joined the channel
17:18
troy_s
Okie
17:18
Bertl
very shortly, there will be a 3x3+3 LUT before the HDMI out
17:18
troy_s
So I guess a few things are needed (and you have to figure out how to get them on board... :) )
17:19
Bertl
the code is already there, it just needs to be integrated and tested
17:19
troy_s
1) The matrix conversion for CMV -> 709 primaries.
17:19
troy_s
2) A 1D LUT for the transfer curve (which is approximately 1.95 straight gamma, but not entirely; there is a linear leg and a curve.
17:19
Bertl
both 3x3 and gamma luts are loadable from userspace
17:19
troy_s
Bertl: Great. 1D LUT though right?
17:19
troy_s
(not straight gamma)
17:20
Bertl
yes, the mathematics are simple for now, the 3x3 is not a 3D lut per se, it works more like a non linear matrix
17:20
troy_s
(also useful for providing sRGB output if needed. The 3x3 remains the same (sRGB has same primaries as 709) but the transfer curve is different.
17:21
troy_s
the matrix is the matrix, I'd be more concerned with the curve.
17:21
troy_s
(Seeing as how it is non uniform)
17:21
Bertl
the +3 are simply 3 12->12 1D luts
17:21
troy_s
Hrm. I see.
17:21
troy_s
Why three?
17:21
troy_s
I'm sure there is a reason I have no clue about.
17:21
Bertl
because it is simpler
17:21
troy_s
Well there you go. :)
17:22
Bertl
you can as well load them with identical values
17:22
troy_s
Gotcha.
17:22
troy_s
Well that's positive.
17:22
Bertl
but it also allows to pack something else into that
17:22
Bertl
like for example part of the color correction
17:22
troy_s
I need to refresh my knowledge on the matrices. Somehow need to figure out the raw primaries of the sensor, and then integrate the white balance as separate
17:23
troy_s
Basically the 3x3 matrix will need to be distilled down (likely through pre-baked matrices)
17:23
Bertl
the advantage of the 3x3 being a lut setup is, that you can simply use example values, no need for calculating the matrix
17:23
troy_s
(Bradford + raw primaries)
17:24
Bertl
i.e. if you can generate a 3D lut which takes RGB and outputs RGB, it is rather simple to calculate the lut entries for the 3x3
17:24
troy_s
Sure.
17:24
troy_s
Not my strong point, but I know that RGB -> XYZ ->RGB can be baked into a single 3x3
17:25
troy_s
then tack on the transfer via the 1D
17:26
gwelkind
left the channel
17:35
guest
Bertl: zero black level is achievable on sensor? (e.g. the adder allows signed operation?) or that is part of some calibration on CMV ?
17:37
Bertl
the 3x3 is signed, and the CMV allows offset values to get zero black
17:41
gwelkind
joined the channel
18:25
rexbron_
If anyone has SMPTE access, you can get the RP on the matrix math to get from camera RGB to a Display referred whitepoint
20:02
troy_s
RP?
20:02
troy_s
rexbron_: I'd think that isn't too difficult actually.
20:03
troy_s
rexbron_: Camera RGB to D65 via Bradford
20:03
troy_s
rexbron_: Been looking at the DNG code. What a nightmare.
20:03
troy_s
rexbron_: It's a ridiculously cumbersome mess.
20:16
se6astian
what exactly do you mean with "the DNG code"?
20:16
troy_s
se6astian: The Adobe SDK
20:16
troy_s
se6astian: Which covers the digital negative format they implemented. (Well not just anyone... Thomas Knoll himself)
20:29
se6astian
I see, before we used a modified libtiff to write dng files
20:30
se6astian
was about 10x faster than doing the same with the adobe sdk
20:38
troy_s
Likely
20:38
troy_s
I'm still not sold on DNG
20:38
troy_s
I think it's a wank
20:38
troy_s
But I don't want to rule out exporting to DNG for the adventurous, and it would be via OIIO.
20:39
troy_s
It's just a lot of extra cruft that I can't see all that much usefulness out of
20:45
troy_s
Bertl: What's the chances of onboard Huffman encoding?
20:45
troy_s
Bertl: Is that board fast enough to keep up or?
20:46
Bertl
chances are not that bad, there have been a number of projects/implementations as early as 2008 (for FPGA)
20:46
Bertl
OTOH, the data rate is quite high with 4K/300FPS :)
20:53
Bertl
but of course, we will investigate this for Axiom
20:59
troy_s
Bertl: That's positive. I could see a bottleneck at that 300FPS zone.
20:59
troy_s
Bertl: Because whatever the case, I can see rexbron_'s point about compression. Huffman makes solid sense.
21:00
troy_s
Bertl: I could write a stub quickly, but that doesn't cut down on cost of storage during shoot.
21:01
Bertl
I think some kind of compression should be doable, either run length coding or deflate or huffman
21:01
troy_s
Huffman I guess is ideal.
21:40
se6astian
left the channel
21:42
jucar
left the channel
22:18
jucar
joined the channel
22:23
troy_s
Bertl: Ping
22:23
troy_s
Bertl: Trying to encode to a jpeg2000 image format using convert, convert -size 4096x3072 -depth 16 rawfoobar.rgb output.jpg
22:23
troy_s
it is giving me a three image picture.
22:23
troy_s
(Three images side by each.)
22:24
Bertl
what is rawfoobar.rgb?
22:25
troy_s
Bertl: I found it out. Rename it to .gray and it interprets it as a greyscale image.
22:25
troy_s
I wonder how to do a linear encoding in JPEG2000.
22:26
Bertl
ah, so raw data, just specify what it is
22:26
Bertl
i.e. use gray:rawfoobar.rgb
22:27
Bertl
jpeg2k btw, uses jp2 or jpc
22:28
Bertl
so if you want convert to create that either use this extension or again specify the format
22:32
troy_s
Got it worked out.
22:32
troy_s
convert -size 4096x3072 -depth 16 rawfoobar.rgb output.jp2
22:32
troy_s
Appears to do the trick
22:32
troy_s
(Aside from the color mangling.
22:32
troy_s
Not sure if JP2 can store linearized data.
22:40
Bertl
not sure how the encoder would know :)