Current Server Time: 22:33 (Central Europe)

#apertus IRC Channel Logs

2015/11/01

Timezone: UTC


01:50
fsteinel
left the channel
01:50
fsteinel_
joined the channel
02:40
intracube
changed nick to: intracube|away
02:47
spookdigit
left the channel
03:52
Bertl_oO
off to bed now ... have a good one everyone!
03:52
Bertl_oO
changed nick to: Bertl_zZ
05:50
slikdigit_
joined the channel
05:50
slikdigit_
changed nick to: spookdigit
06:59
spookdigit
left the channel
07:58
sebix
joined the channel
08:09
sebix
left the channel
11:06
se6astian|away
changed nick to: se6astian
11:48
danieel
se6astian: the article with debayering is really misleading.. some facts are just not true
11:57
se6astian
I am happy to forward your feedback to the author
11:58
danieel
i wonder, there was no review? usually unposted articles appear here, no?
12:13
Bertl_zZ
changed nick to: Bertl
12:13
Bertl
morning folks!
12:15
Bertl
danieel: facts are always true, but I think you meant statements presented as fact
12:16
danieel
correct :)
12:19
Bertl
but I'm sure, sebastian is happy to publish your article if you write one ... it seems to be the slow time of the year :)
12:20
danieel
what about is the most interest in?
12:20
danieel
(got some comparison of past article pageviews / popularity ?)
12:20
Bertl
everything color science related seems to be very popular
12:21
Bertl
compression is probably a runner up
12:29
se6astian
time for a walk
12:29
se6astian
bbl
12:29
se6astian
changed nick to: se6astian|away
12:30
fsteinel_
changed nick to: fsteinel
15:10
Bertl
off for now ... bbl
15:10
Bertl
changed nick to: Bertl_oO
15:19
se6astian|away
changed nick to: se6astian
16:42
intracube|away
changed nick to: intracube
18:15
sebix
joined the channel
18:42
sebix
left the channel
18:46
sebix
joined the channel
18:57
sebix
left the channel
19:07
pozitron
joined the channel
19:15
sebix
joined the channel
19:21
troy_s
Plurality of truths.
19:21
troy_s
Welcome to the Postmodern era for those of you stuck in late Modernism.
19:22
troy_s
danieel: In particular, what did you dislike / consider "untruth" in the article?
19:22
comradekingu
danieel: i can correct the flaws if you point them out
19:22
troy_s
comradekingu: I'm more interested to hear what they are.
19:22
troy_s
(first at least)
19:23
comradekingu
danieel posts the flaws, we factcheck it, and then correct. If in error, i dont think the original author will mind the text changing
19:24
troy_s
comradekingu: Typo here to fix though "...imaries D e-Baye...imaries D e-Bayer..."
19:24
troy_s
Pretty sure that is an errant space.
19:27
troy_s
I will say that the ratio of pixels that are interpolated appears wrong.
19:27
troy_s
I believe that any given typical RGGB styled Bayer array means that
19:28
troy_s
20.8% of your pixels are properly sampled, with subsampled pixels then making up the remaining 79.2 %
19:28
troy_s
;)
19:29
troy_s
Erp that's bad math on my part methinks. Let me run it.
19:29
troy_s
Ah it works out precisely 66.6% Well done.
19:29
troy_s
My bad.
19:30
troy_s
danieel: Can you cite where you see issues?
19:30
comradekingu
The errant space is a <span class="dropcap">D</span>e-Bayering ...
19:30
troy_s
Ah it's supposed to be a drops cap.
19:30
troy_s
But doesn't render that way here.
19:30
comradekingu
it shows as a large capital letter like in newspapers, maybe just a css error on your view
19:30
troy_s
Although I am quite amused at the homebrew demosaicing techniques.
19:31
comradekingu
I think the shoodak stuff should be in a follow-up article
19:31
troy_s
comradekingu: It's in the summary page. Go to apertus.org and you will see the strange look.
19:32
comradekingu
link?
19:32
troy_s
comradekingu: Just go to apertus.org
19:33
comradekingu
sometimes its De-Bayering, and sometimes its inconsistently "debayering"
19:33
troy_s
Read the article in question and you will see that the summary snippet has that odd space where the drops cap shows up (which renders fine when you read the whole article.
19:33
troy_s
It's debayering or demosaicing
19:33
troy_s
Or demosaicking
19:33
troy_s
The hyphenation is completely odd as hell whenever I see it.
19:33
troy_s
It's not like it is an existing word that has been 'hacked' to be undone.
19:34
troy_s
I need a coffee.
19:34
danieel
i would go for demosaicing
19:34
troy_s
Hrm. Seems there were a few comments on the colour article. I guess that's positive.
19:34
troy_s
(I didn't think there would be anyone at all interested.)
19:35
danieel
all video cameras with CCD <-- untrue. only highend cams were 3 chip, other/mainstream were bayer
19:35
troy_s
danieel: Probably agree. But that nuance and nitpicky rubbish aside, where did you see an error?
19:36
danieel
or did the author mean tubes? they were usually made that way (3 tubes for color)
19:37
troy_s
I wasn't under the impression any cameras used beam splitters though.
19:38
troy_s
Now that you mention it.
19:38
danieel
"with the introduction of cmos chips" ... well, CFA is totally unrelated of ccd or cmos technolgy
19:38
troy_s
Because CCDs almost certainly don't use beam splitters (the Genesis most certainly didn't.)
19:38
danieel
the genesis is a RGB striped CFA
19:38
danieel
like a LCD
19:39
danieel
and probably nonsquare pixels therefore
19:39
troy_s
Well it is referenced as a CCD in all of their literature.
19:39
troy_s
Hence my sloppy terminology.
19:39
danieel
the Bayer patents is based on different sensitivity to luma than chroma... not specifically colors as this article states
19:40
troy_s
A luma bias is hard to do without mentionning colour.
19:40
troy_s
It's a fair shortcut wouldn't you say?
19:40
danieel
the artefact images are from the wikipedia... but it seems nobody read the article :)
19:40
troy_s
Two greens for every other red and blue? (chromaticity of each notwithstanding)
19:41
troy_s
Artefact?
19:42
danieel
the 6 little square images showing issues with demosaicing
19:42
troy_s
Ah. Yes.
19:42
troy_s
I _did_ read the article you jerk.
19:42
troy_s
I didn't dwell on those images though. Good point.
19:42
danieel
you wrote this debayer article?
19:42
troy_s
God no.
19:42
troy_s
I wrote the colour article.
19:43
danieel
im discussing the debayer article, last one
19:43
troy_s
I know.
19:43
troy_s
I did read that one.
19:43
comradekingu
fixed "De-Bayering" the CCD issue, and "Colour"
19:43
troy_s
Colour / color is whatever
19:43
troy_s
that's a UK / Canada versus USA etc.
19:44
troy_s
I wouldn't worry about that.
19:44
comradekingu
sticking to one througout the article is the winning move
19:44
comradekingu
I do
19:44
troy_s
Yes.
19:44
troy_s
Agree.
19:44
troy_s
I agree on that front.
19:44
comradekingu
If it starts with The secret of the two missing colours, it shouldnt be consistently "color" from there
19:45
danieel
the examples should be really graded same.. and have the same resolution (if they are the same frame).. so that one can do arithmetic subtractiona and see the minor details
19:45
comradekingu
I was going to run this article through the "so-that-people-like-me-understand"-filter, but i dont understand de-bayering well enough
19:48
comradekingu
the whole thing starts with "It is called De-Bayering." and then a picture of something that isnt de-bayering
19:48
danieel
the last paragraph is then confusing - they are developing a new and simple low quality algorithm, but it should improve the industry? :)
19:48
danieel
comradekingu: good point :) that hit me first as well
19:48
troy_s
danieel: Sounds like you are writing an article.
19:48
danieel
that prism stuff should go away
19:49
troy_s
debayering is very simple
19:49
troy_s
You have a bayer
19:49
troy_s
you need to subsample (yes my math friend beat me to death over this) the "gaps"
19:49
comradekingu
Or could call the article " The secret of the two missing colours"
19:49
troy_s
or more truthfully, you are sampling the signal at the interval that it doesn't exist.
19:49
troy_s
So at least I agree with both of you on two points;
19:50
comradekingu
and then make a thing out of how its either the prism, or the debayering/demosaic, and that the latter is often proprietary
19:50
troy_s
1) That the prism stuffs is rather misdirecting to a degree. Perhaps a historical bit on the tricolor film would have been more reasonable.
19:50
comradekingu
"The secret of the missing colours" is a good name
19:50
troy_s
2) The last bit on the homebrew amateur demosaic is also extra
19:50
troy_s
by the way, looked at the new "P3" iMac
19:50
troy_s
Impressive display it seems
19:51
comradekingu
who made it?
19:51
troy_s
couldn't test the panel, but I suspect it is the same panel in the HP z27 and the others
19:51
danieel
it is not proprietary - they are several algorithms described.. in scientific papers and so on.. but competitive cameras use their own hidden essence here :)
19:51
troy_s
The good news is that the blue and red primaries are darn close to the P3
19:51
comradekingu
yes, but the whole article is a low-down on how to understand and use it
19:51
troy_s
green is shy
19:51
danieel
*there
19:51
troy_s
I'm VERY interested to know how the hell OSX handles that mess though
19:51
comradekingu
posing as just 1/3 of those things
19:52
troy_s
Because the UI elements would be of two states:
19:52
troy_s
1) NOT color managed
19:52
troy_s
2) Color managed
19:52
troy_s
if the (1) folks are going to see batshit crazy wrong colours in colour pickers etc.
19:52
troy_s
if (2) it's going to break a crapload of software that knows what it is doing and is expecting the UI to simply display the colours as-is.
19:53
sebix
left the channel
19:54
danieel
the profile is give to color managed software, while non color managed parts gets the profile applied in the OS ? i do not see any issue here
19:56
comradekingu
shouldnt there be a mention of faevon sensors in this article?
19:58
danieel
you cant mention everything.. then it comes to question if RGBE should be mentioned, or CMY sensors
19:59
danieel
or the striped RGB genesis :)
19:59
danieel
what the article should mention is that simple debayer is done that way: *** (math formula)
19:59
comradekingu
Is that everything then?
19:59
danieel
and that new one is doing it differently, and what is the improvement (processing time, quality, ... whatever)
20:00
comradekingu
That is a different article on shoodak imo
20:00
dmjnova
danieel: RGBE?
20:00
danieel
dmjnova: http://www.dpreview.com/articles/1471104084/sonyrgbeccd
20:00
troy_s
danieel: You are missing my point.
20:00
danieel
(actually i own a Sony F828)
20:00
troy_s
Let's say you load an icon
20:00
troy_s
That icon is RGB
20:00
troy_s
so it hands say, 0.1,0.2,0.3 display referred RGB to the display
20:01
troy_s
Let's pretend that is sRGB
20:01
danieel
the screen is sRGB?
20:01
troy_s
(as is going to be the most typical case)
20:01
troy_s
No the display is P3
20:01
troy_s
The new iMac
20:01
dmjnova
Ah, the cyan version of something I've been wanting for a while
20:01
danieel
so the understanding of icon file is sRGB
20:01
dmjnova
RGBIr
20:01
troy_s
so now Apple's magic ColorSync goes "oh this is untagged (or tagged) so I'll assume sRGB and convert to DCI-P3. MAGIC!!!!"
20:01
troy_s
But now let's consider an industrial application like Nuke... it shows a colour picker.
20:02
danieel
dmjnova: yes, there are RGBIr, and also RGBW :) sooo many options
20:02
troy_s
Say a wheel.
20:02
troy_s
Nuke goes "Ok... we've adjusted the colour wheel as per the configuration file. Let's display it."
20:02
dmjnova
danieel: Hmm...not seen one, but would love such a filter for the CMV12k
20:03
troy_s
Then ColorSync goes "Oh wait... this is untagged data... our default is that untagged data is sRGB... Let's transform sRGB to DCI-P3's iMac profile. MAGIC!!!!"
20:07
danieel
that would happen if app based color management is unrealted to OS.. i think the OS knows what is and is not tagged :)
20:08
danieel
there will be no funny magic
20:08
danieel
maybe in windows :))
20:08
troy_s
danieel: Trust me
20:08
troy_s
That OS can't be smart
20:08
troy_s
It will fuck things up
20:08
troy_s
badly
20:08
danieel
only if the app is uncooperative :)
20:08
troy_s
When writing applications, you can't have hidden and secret "magic" transforms.
20:09
troy_s
If I want to write RGB values to screen, you have to let me because you have NO IDEA WHAT I AM WRITING
20:09
troy_s
Am I writing a linearized texture? Am I writing RGB data that has no spec'd (in profile in system) "space"?
20:09
troy_s
Am I writing internally controlled values?
20:09
se6astian
thanks for the feedback, I adapted the first few paragraphs of the article a bit to correct the CCD vs CMOS issue
20:09
troy_s
You simply can NEVER magic the crap.
20:10
troy_s
danieel: Also, you seem to imply that the ICC system is "the one true colour management" system
20:10
troy_s
Which is fundamentally a bad idea.
20:10
troy_s
Largely because ICCs are specifically _display referred_ colour management with a keen eye on graphic design.
20:10
comradekingu
Isnt bayering originally the three layers in celluloid film?
20:11
troy_s
They do that remarkably well and admirably, including Black Point Compensation and Total Area Coverage for inks etc.
20:11
troy_s
comradekingu: No
20:11
troy_s
comradekingu: Colour film literally has three layers
20:11
danieel
comradekingu: if you would use the correct term: demosaicing, then you would see no - there is no mosaic
20:11
troy_s
comradekingu: And in terms of film, aside from silver issues in terms of grain size, there is a precisely 1:1 correspondence with the region on each
20:11
comradekingu
where did bayering come from then?
20:11
troy_s
It's a name. Bayer.
20:12
troy_s
Kodak.
20:12
intracube
troy_s: doesn't film actually have way more in practice?
20:12
comradekingu
year?
20:12
troy_s
The peep that invented the actual checkerboard.
20:12
intracube
like two layers for each primary?
20:12
intracube
different grain sizes and such??
20:12
intracube
sticky ? key
20:12
intracube
<- nitpick
20:13
intracube
and hello :)
20:14
intracube
sorry for derailing the convo
20:21
troy_s
intracube: For our purposes in this discussion, if we think of three layers of film as three "planes" of arbitrary chromaticity, there is a 1:1 relationship between the coordinates and each plane
20:21
troy_s
intracube: Sorry was on phone.
20:22
intracube
np. just thought I'd killed the conversation with my rather pointless interjection :)
20:22
troy_s
intracube: So when the lens resolves a line say, the line crosses through (again, silver nuances aside) one unit per plane.
20:23
intracube
the resolution is stacked one-on-top. not interleaved like sensor bayer patterns
20:23
intracube
yep
20:26
troy_s
Exactly.
20:26
troy_s
Although I believe some CCDs do something akin to that (or some sensor somewhere)
20:27
troy_s
Can neither confirm nor deny, but I was under the impression that some Foveons might do something like that.
20:27
danieel
foveon does stack 3 layers on top of each other
20:27
dmjnova
that was the whole deal with foveon
20:27
troy_s
intracube: The criticical part about that (which is _precisely_ why a good film emulation LUT is 3D and not just a contrast curve 1D LUT)
20:27
danieel
the latest one then uses a combination of smaller/larger pixels :) so it is 4 small on top + 1 under them + 1 under that
20:27
troy_s
intracube: Is that when you cast light through three unique silver based layers, the colours "crosstalk" in _very_ unique ways for a number of reasons.
20:28
troy_s
Hence why you _must_ use a 3D LUT to achieve that warping of the RGB values.
20:28
troy_s
danieel: That's very interesting. So a quad of FOUR sensors, then one under, then one under?
20:28
intracube
troy_s: that's interesting (and something I've been experimenting quite a bit)
20:28
intracube
not methodically, though
20:28
troy_s
intracube: It isn't just like some sort of ideal wavelength filter.
20:28
danieel
troy_s: http://img1.lesnumeriques.com/news/34/34708/Foveon_Quattro.jpg
20:29
intracube
of course, yeah
20:29
danieel
the layers are not RGB, but rather (RGB),(RG),(R)
20:29
troy_s
intracube: It's like a blender... the first layer leeches out the information it wants and mangles it slightly because of chemicals into the recording, then the next layer gets handed a mangled version of light and does the same, then the last gets the final mangled mess.
20:29
troy_s
intracube: Folks get confused with crosstalk, but that is precisely the best medium to really understand it.
20:30
troy_s
intracube: Like when visual effects folks want to pull keys in the native camera space
20:30
troy_s
intracube: Due to crosstalk if you bend it to another one.
20:30
troy_s
intracube: If you take a demosaiced image and try to key from it, you are already one generation away from the metal, which means you are dealing with sampled values that are not quite exactly what the lens saw.
20:31
troy_s
If you go another generation away by changing colour spaces, your keys get even more distanced, in particular due to the magic in the subsampling.
20:31
intracube
yep
20:31
troy_s
Hard thing to try and explain it seems.
20:32
intracube
troy_s: and w/regard to the film process. it's been tweaked and adjusted over the decades, the lack of accuracy has become something desirable
20:32
intracube
the 'film look'
20:34
troy_s
intracube: It's a magic box that is a great little entry point for creativity I guess.
20:34
troy_s
intracube: But from a practical point, say mixing CGI with the stock, you have to be careful there.
20:34
troy_s
(hence why the 3D LUT is critical)
20:35
intracube
danieel: does the stacked sensor design impact the sensitivity much?
20:35
danieel
never saw a datasheet of the foveon sensors
20:35
danieel
so regarding sensitivity i have no data
20:35
intracube
ah ok
20:36
intracube
I thought there was generally overlap between the filters for the primaries
20:36
intracube
so in practice, the blue filter would be slightly filtering some green wavelengths
20:37
danieel
where?
20:37
danieel
in the stacked sensor there are no filters.. it is just absorbed into the silicon
20:38
intracube
ok, but is the silicon effectively absorbing some wavelengths and letting others pass to the lower layers?
20:38
troy_s
There should be some crosstalk for certain.
20:38
intracube
if the overlap is anything like: https://www.cs.auckland.ac.nz/courses/compsci708s1c/lectures/Glect-html/rgb-ideal.GIF
20:38
danieel
yes, longer waveforms (red/ir) passes deeper
20:38
troy_s
(Which isn't a bad thing necessarily, just makes the transform a little messier.)
20:38
intracube
I would have thought a small amount of loss would be unavoidable
20:44
intracube
misses that natural 90s film look
20:44
danieel
what natural?
20:45
intracube
fairly unprocessed (other than traditional timing)
20:45
intracube
pre digital intermediate/digital colour correcting
20:46
danieel
you can still shoot on film :)
20:46
intracube
indeed. wasn't there crowd funding to get a film factory in Italy back up and running?
20:46
intracube
wonders what happened to it
20:46
danieel
it was for stills, wasnt?
20:47
intracube
I'm not sure. it might have been this: http://www.filmferrania.it/
20:48
danieel
but even if they do make 8/16mm.. where to develop it?
20:49
danieel
offering a complex service (as Kodak focused on) would be nice
20:49
troy_s
intracube: That's a disconnect between understanding colour coupled with a rather ideological bit of cruft around nostalgia, authenticity, and other strange things.
20:50
troy_s
intracube: There's no reason at _all_ that with a decent entry RGB point you can't get Insert-Your-Favorite-Stock-Magic-Here.
20:50
troy_s
Precisely down to the latitude if you want.
20:53
intracube
troy_s: of course, yep. but movies usually go for a heavy stylised look these days
20:53
intracube
danieel: it'd be great if they offered a full service
20:54
troy_s
intracube: Not terribly fair to comment on creative choices is it?
20:55
troy_s
(I mean it's totally fair, but sort of... nebulous in the highest level view?)
20:55
intracube
absolutely, it is :P
20:55
troy_s
(we could certainly comment on whether we think a look is "working" for a given piece, whatever the goals and aims)
20:56
intracube
it's personal preference, of course
20:56
comradekingu
whos editing the page, i cant save now
20:56
intracube
would just prefer fewer films to have lifted blue shadows and very rolled of yellow highlights ;)
20:56
comradekingu
i reworked the whole article
21:06
se6astian
we have been reddited once more: https://www.reddit.com/r/cinematography/comments/3r1z1b/where_did_that_open_source_4k_camera_end_up/
21:07
comradekingu
Nitpickers, meticulously start your collective engines https://apertus.org/node/413/revisions/3801/view
21:09
comradekingu
"The part where the market progressed, does not encompass the evolvement carried out by apertus"
21:10
comradekingu
which is what i would post if i was a redditor
21:10
comradekingu
I bet that guy would say the trolltech phone wasnt important either
21:11
comradekingu
And as always, http://apertus.org/home-wip is relevant to the difference between industry and apertus
21:13
troy_s
what the hell is envolvement?
21:13
troy_s
involvement?
21:13
comradekingu
where does it say that?
21:15
troy_s
Ah nevermind. I read your comment above.
21:15
troy_s
Thought it was a posted comment.
21:15
troy_s
I was wondering what envolvement was... but I see you meant involvement.
21:15
troy_s
(English crap)
21:15
intracube
comradekingu: I get access denied for your link (I'm already logged in)
21:15
troy_s
I am not certain their thoughts are entirely off base regarding costs for certain.
21:16
comradekingu
sebastian has to publish or semi-publish it, i dont have write-access
21:16
troy_s
It can kill a project, but the costs of scale aren't something afforded to our two humble devs.
21:17
intracube
comradekingu: I mean I can't view it
21:17
comradekingu
If what you want is 4K, 4K can be had. If you want an open source camera, it can be had in a 4K variety
21:17
comradekingu
I know, its just sitting around pending approval
21:18
troy_s
I am deadly interested to see how the global shutter behaves.
21:18
comradekingu
But do post an update of process, grants, sensor info and such on the reddit
21:48
dmjnova
left the channel
22:07
danieel
se6astian / Bertl_oO : check how does XCN15034 apply to microzed, to avoid later issues (my friend discovered that "feature" in a pretty expensive way prior it was published)
22:08
se6astian
you mean http://www.xilinx.com/support/documentation/customer_notices/xcn15034.pdf I presume?
22:09
danieel
yes
22:10
se6astian
thanks for the notice
22:10
se6astian
Bertl_oO: will be able to determine if that is relevant for us
22:10
danieel
hes _oO, i do not know if it raises attention, so remind him or make a tak
22:10
danieel
*task
22:10
Bertl_oO
no worries, it does :)
22:11
danieel
good then :)
22:13
Bertl_oO
PR_POR_B seems to be tied to the 1.5V PG output
22:14
Bertl_oO
(on the MicroZed) which in turn is sequenced after 1V/1V8
22:14
Bertl_oO
actually it is the 3V3 PG output, which is sequenced last
22:15
Bertl_oO
so I think that should be fine for us
22:32
pozitron
left the channel
22:47
se6astian
time to sleep
22:47
se6astian
good night
22:47
se6astian
changed nick to: se6astian|away
23:14
dmjnova
joined the channel
23:57
Bertl_oO
off to bed now ... have a good one everyone!
23:57
Bertl_oO
changed nick to: Bertl_zZ