Current Server Time: 05:39 (Central Europe)

#apertus IRC Channel Logs

2014/10/25

Timezone: UTC


00:18
Bertl
off to bed now ... have a good one everyone!
00:18
Bertl
changed nick to: Bertl_zZ
00:41
g3gg0__
left the channel
01:14
intracube
left the channel
01:38
wescotte
left the channel
02:51
intracube
joined the channel
03:17
dmjnova
left the channel
03:31
intracube
left the channel
04:35
intracube
joined the channel
05:08
ItsMeLenny
joined the channel
05:33
darthrake
left the channel
07:22
ItsMeLenny
left the channel
07:23
ItsMeLenny
joined the channel
07:27
g3gg0
joined the channel
07:49
intracube
left the channel
07:57
ItsMeLenny
left the channel
08:38
g3gg0
left the channel
08:43
g3gg0
joined the channel
09:30
se6astian|away
changed nick to: se6astian
09:32
se6astian
good morning
09:37
Bertl_zZ
changed nick to: Bertl
09:37
Bertl
morning folks!
10:01
ItsMeLenny
joined the channel
10:31
se6astian
massive wiki restructuring done
10:31
se6astian
https://wiki.apertus.org/index.php?title=Main_Page
10:31
se6astian
https://wiki.apertus.org/index.php?title=Special:RecentChanges
10:31
se6astian
https://wiki.apertus.org/index.php?title=AXIOM_Beta
10:47
Bertl
we will never find anything again ... :)
10:51
se6astian
thats the idea :D
10:55
ItsMeLenny
lolol
10:56
ItsMeLenny
looks pretty though
10:56
ItsMeLenny
much better than a lot of wikis ive seen
10:56
ItsMeLenny
ever heard of wikipedia? so hard to navigate
11:15
se6astian
Live screencast for getting started with phabricator on Tuesday: http://lab.apertus.org/calendar/event/view/1/
11:22
legendin
joined the channel
11:24
se6astian
bb in the evening
11:24
se6astian
changed nick to: se6astian|away
12:36
legendin
left the channel
13:36
Bertl
off for now ... bbl
13:36
Bertl
changed nick to: Bertl_oO
13:50
Rebelj12a
joined the channel
13:52
Rebelj12a
Run for the hills!
13:52
Rebelj12a
left the channel
14:44
_TiN_
left the channel
14:50
Rebelj12a
joined the channel
14:51
Rebelj12a
Finally, its nice to get a large project finished. Alot less stress...
14:51
Rebelj12a
Well until... if your like me right on to the next one XD
15:01
intracube
joined the channel
15:54
troy_s
Beginning to look like the loony bin in here.
16:10
troy_s
What the flying hell? Bertl_oO has the camera now turned into a DSLR?
16:10
troy_s
All of the talk of the 1/4 20 pitch bolt is alarming.
16:13
ItsMeLenny
left the channel
16:45
Bertl_oO
troy_s: as usual there are many different interests and various applications ... I don't see a problem with that, somebody who would like to have a DSLR can get/make a DSLR case
16:45
troy_s
Bertl_oO: And so it begins.
16:46
Bertl_oO
no problem there, it is the least of problems with an open design and the best, we do not even have to do it :)
16:46
troy_s
Make whatever one likes, I would caution against a straying from the original design vision however.
16:46
troy_s
I agree entirely. It wasn't a statement against preventing folks from doing what they like. At all.
16:47
troy_s
Rather that it is critical, from my stupid and useless vantage, to keep a keen eye and focus on the design for the main shipping point.
16:47
Bertl_oO
no worries, we simply cherry pick what we consider suitable
16:47
troy_s
As the design drift is incompatible.
16:48
troy_s
(See Libre software as a basic proof of principle that design drift is absolutely debilitating.)
16:49
troy_s
(And I really wish Seb's original idea to draft women happened. This sucks in here.)
16:49
troy_s
(Toxic trajectory.)
17:05
Bertl_oO
how so?
17:20
troy_s
I have to explain?
17:22
se6astian|away
changed nick to: se6astian
17:25
Bertl_oO
troy_s: yes, please elaborate :)
17:26
se6astian
good evening
17:26
Bertl_oO
changed nick to: Bertl
17:27
troy_s
I believe in culture, and this one is horrifically lopsided. It was brought up before by seb, and damn if I didn't think it was one of the most pressing issues.
17:27
troy_s
Sausage fests brew bad brews.
17:27
Bertl
we have women here, at least one
17:28
Bertl
and we probably had more, if there were more interested in hanging around
17:28
troy_s
Thank god.
17:28
troy_s
Well that is reflexive.
17:28
troy_s
Sort of a bogus argument actually.
17:28
troy_s
(With respect.)
17:28
Bertl
so our part is to make it more interesting :)
17:30
Bertl
I had a number of private talks with women interested in the AXIOM, so there is a certain percentage (below 5% I guess)
17:30
Bertl
but at the moment, the channel doesn't seem very appealing/interesting to them
17:30
troy_s
Shocker. :)
17:30
se6astian
We received another indiegogo contribution today - I can't reproduce this? maybe someone still had the perk displayed in a cache?
17:30
troy_s
Sausage fests beget sausage fest's.
17:31
troy_s
se6astian: Probably.
17:32
Bertl
I think you can still contribute if you get the proper link
17:32
troy_s
Just not a published link now but still active?
17:34
Bertl
yes, something like that
17:35
se6astian
yes
17:35
troy_s
Still in board dev?
17:36
troy_s
As a total newb in board design, how do you intend to iterate this down to the final form factor?
17:36
troy_s
Is there another semi-shoebox design in process or?
17:36
Bertl
no, we are already working with the Beta dimensions
17:37
Bertl
I've almost finished the interface board, which goes between sensor board and beta main board
17:37
Bertl
both, sensor and interface boards are squares, the width of the Micro/PicoZed
17:38
Bertl
i.e. 2.25x2.25 inch
17:39
Bertl
but we still need to do some iterations to get the electronics right
17:40
Bertl
(although I'll try to get it right on the first one of course :)
17:40
Burridan_
joined the channel
17:40
Bertl
welcome Burridan_!
17:41
Burridan_
left the channel
17:45
troy_s
se6astian: Maybe attach an image of Ronford Baker plate?
17:45
troy_s
The bolt pattern is fixed on one size and variable on another.
17:46
troy_s
(IIRC. The only ones I ever touch are the 5" or so variety)
17:48
se6astian
troy_s: please do that would be great!
17:48
se6astian
any reference material helps
17:49
troy_s
se6astian: Looking for the Alexa bolt pattern or Sony.
17:49
se6astian
also I am wondering how the 3/8" / 1/4" cheeseplate holes are spaced
17:49
se6astian
does every manufacturer do it differently
17:49
se6astian
or is there THE standard?
17:49
troy_s
Yes.
17:49
troy_s
Standard cages in North America are 3/8ths cheese with 1" spacing
17:50
troy_s
Extra is holed out for weight on camera accs.
17:50
troy_s
R3D's cages (aftermarkets) are always 3/8ths with 1/4 for the accessories and holed out
17:51
troy_s
The bolt pattern on the bases of cameras is standardized
17:51
se6astian
please add all that to the thread
17:51
troy_s
And almost every camera I have ever touched has the dual 3/8s for the Ronford Baker quick release clasp to affix to.
17:51
se6astian
very good knowledge
17:51
troy_s
Actually worrying. ;)
17:51
se6astian
will probably summarize everything into the wiki once we have everything together
17:51
troy_s
Asking on a mailing list for 1st ac feedback would be helluva prudent
17:52
troy_s
Because chances are they (there is likely one or two) are silent.
17:52
troy_s
rexbron: PING
17:52
troy_s
se6astian: When you are designing the top cage
17:53
troy_s
Bear in mind that for top mounts you need a good number of mounting holes to base top plates out
17:53
troy_s
And some of those will be standardized (Steadicam top plates) and a good starting point for mountinf
17:53
troy_s
This would also accommodate securing for mounts to other things such as cars etc.
17:54
troy_s
So top and bottom mount holes are a priority to get right.
17:54
se6astian
agreed
17:54
troy_s
My personal advice is to reinforce the corners of that housing (the verts)
17:54
troy_s
And bore 3/8s into the top down
17:55
troy_s
For the four points. That buys you a great purchase point for top plates.
17:55
troy_s
And the top plates can then be customized.
17:55
troy_s
Bottom is a little more standardized.
17:55
troy_s
Sides too, if you can get one or two, are always welcome.
17:55
troy_s
(The added upside is that nothing speaks more âindustry ruggedâ than bolt holes. ;) )
17:56
troy_s
Honestly though, this sort of ânon issueâ as it seems, is a _huge_ issue on set. To the point where some cameras are not even considered due to the fsckery you need to cage them etc.
17:57
troy_s
se6astian: 3/8 16 is the pitch IIRC.
17:59
troy_s
se6astian: Many bodies have a locator pin too. See Alexa.
17:59
troy_s
I believe it rests offset from the dual 3/8s slightly.
17:59
se6astian
all to the thread please, every tiny details!
18:01
troy_s
You are already going to face a design fork. ;)
18:01
troy_s
Because stillsies will want 1/4 20 which is junk for motion picture making.
18:02
troy_s
(If you think of quick release to Tango heads, or doves etc., in motion picture it is exclusively 3/8ths.)
18:02
Bertl
I think it should be trivial to have two base plates for the case
18:03
Bertl
(or an arbitrary number of base plates, we do not need to actually produce them)
18:05
se6astian
or just several holes of 1/4" and 3/8" on the bottom
18:06
Bertl
if there is enough space without rendering the base plate useless, why not
18:06
troy_s
That bottom bolt pattern is critical though.
18:06
troy_s
As is the height from the base to lens
18:07
troy_s
Otherwise you are going to be expecting custom risers and all sorts of other crap.
18:07
troy_s
se6astian: The bolt pattern has to be aligned in a fixed position along the axis.
18:07
se6astian
I get the feeling that the Beta should not yet try to do everything already in the first step
18:08
troy_s
Anyways... That is something to get right.
18:08
troy_s
LOL. It is height and bolt patterns.
18:08
troy_s
Not exactly a huge design leap.
18:08
troy_s
Just make sure to meet the standards, doubly so on a beta.
18:08
troy_s
The part being missed here is testing
18:08
troy_s
.
18:08
troy_s
The beta is designed for testing right?
18:08
troy_s
;)
18:09
Bertl
the height is fixed, modulo the base plate height
18:09
Bertl
the pattern can be adjusted to whatever fits the size
18:09
troy_s
The height is fixed to existing accessories actually.
18:09
troy_s
:)
18:10
troy_s
Otherwise you are asking a whole crop of testers to build aftermarket risers, aluminum shims, and a bunch of other crap.
18:10
Bertl
no, actually the height is fixed to the Beta height, as the sensor is in the center :)
18:10
troy_s
Uh.
18:10
troy_s
God.
18:10
Bertl
and it's fine if folks use riser on a Beta development kit
18:10
Bertl
(if they think they need)
18:10
troy_s
Custom risers to use a matte box.
18:10
troy_s
Genius.
18:10
troy_s
Or any other selection of tools that most folks can borrow from a rental house for cheap or free.
18:11
troy_s
C'mon.
18:11
troy_s
This isn't rocket science.
18:11
troy_s
Get a damn 1st involved.
18:11
Bertl
nothing on the wiki about that
18:11
troy_s
Right. And there is nothing in GIMPs design docs about scene referred imagery.
18:12
troy_s
Your culture is the gravitational field.
18:12
troy_s
So expecting a culture to offer up insights on something they have no clue about is.
18:12
troy_s
Well.
18:12
Bertl
what I'm saying is, if it is so damn important, why don't we have any wiki entry about it?
18:12
troy_s
You get my point surely.
18:12
troy_s
Because. Culture.
18:13
troy_s
How many folks have used the actual tools of the trad
18:13
Bertl
yeah, then folks have to live with the results, that's how life is
18:13
troy_s
LOL
18:13
troy_s
I give up.
18:13
troy_s
(And I _know_ you are smarter than that.)
18:13
Bertl
we can't guess everything which is 'important' to somebody
18:14
troy_s
Look... Who is your audience?
18:14
Bertl
developers for the Beta
18:14
troy_s
You have your current audience and your _intended_ audience
18:14
troy_s
You can only design to one.
18:14
troy_s
Your pick.
18:14
Bertl
I pick developers :)
18:14
troy_s
LOL.
18:14
troy_s
Can't wait to use that camera.
18:14
troy_s
Seriously though...
18:15
troy_s
Think about glass
18:15
troy_s
All of the 6x4 glass
18:15
troy_s
That goes in a tray
18:15
troy_s
That goes in a matte box
18:15
troy_s
One simple ridiculously trivial bit of research
18:15
troy_s
Can or can inhibit the use
18:15
troy_s
Now think about rods and other baseplates
18:15
troy_s
Now think about heads
18:15
troy_s
Now think of levels and ball mounts
18:16
troy_s
All of that. Same issue.
18:16
daFred
joined the channel
18:16
troy_s
And now think about the feedback that you can benefit from when someone manages to convince someone to do a simple shot by dragging the camera out and everything works and fits
18:17
troy_s
Or you don't get that, and your culture grows in another direction.
18:17
troy_s
As an independent shooter, are you wanting to have to go to a machinist and get a CNCd riser for your basic tests or borrow a matte box and do some tests?
18:18
troy_s
Am I being unreasonable?
18:18
troy_s
If so, please let me know
18:18
Bertl
imagine everything fits but the camera cannot finish a single take _because_ there is nobody who developed the software :)
18:18
troy_s
Pretty sure the current culture is long on developers
18:18
troy_s
And short on shooters
18:18
troy_s
Fair?
18:19
Bertl
I don't see that at the moment
18:19
troy_s
I have _personally_ watched you work magic in hours
18:19
troy_s
Well by long I mean âa very few talented devsâ
18:19
troy_s
Not hundreds. I doubt there will ever be hundreds. Pareto estimates.
18:19
Bertl
I think there are many good developers out there, they just do not have a chance to develop
18:20
troy_s
But my point is... This is a camera.
18:20
troy_s
And there is a whole legion of things that
18:20
troy_s
I think we can agree
18:20
troy_s
Having hundreds of developers without anyone wanting to use it
18:20
troy_s
Is equally a dilemma?
18:20
Bertl
agreed
18:20
troy_s
So I don't think our views are far off on this subject.
18:20
Bertl
all I said was, the placement of the sensor is fixed
18:21
troy_s
Sure. And I merely suggested that may be an issue.
18:21
Bertl
there is a minimum height (2.25"/2) which is always there
18:21
troy_s
Camera heights to lenses tend to be typically standardized.
18:21
troy_s
Yes.
18:21
Bertl
and there is a maximum you can go up without getting a shoebox :)
18:21
troy_s
The base size was often off of legacy film height
18:22
troy_s
Which, after plate, was common to be a 4" height.
18:22
Bertl
figure out what the optimal height is, let us know, then we can discuss about the baseplate height
18:22
troy_s
Based off of the duality of rod placements: Arri and Panavision
18:22
Bertl
a 3" baseplate doesn't sound very plactical to me though
18:23
troy_s
Height.
18:23
Bertl
4"-1.125"
18:23
troy_s
Not plate
18:23
troy_s
Height from plate to lens. But the reference that you two need are for the two types of rod sets
18:23
Bertl
so what is 'Height'?
18:23
troy_s
Namely height from rod after plate to lens
18:23
Bertl
plate base to lens center?
18:23
daFred
troy_s: please describe all things that are important for you into the wiki! I even can not read as fast as you are writing :))
18:23
troy_s
Panavision is one standard, Arri is the other.
18:24
troy_s
I cannot daFred... I must shower or suffer divorce. :)
18:24
Bertl
yeah, I would also appreciate a drawing with measurements/dimensions
18:24
troy_s
This isn't important to me. At all. But for your independent filmmaker audience it is a critical point.
18:24
troy_s
I would too. Hence... I will say it again: GET A FIRST.
18:24
troy_s
LOL
18:24
daFred
but you can help us!
18:25
troy_s
Someone here nag rexbron
18:25
troy_s
He is a first
18:25
Bertl
troy_s: bring you'r wife here, this will raise the percentage of women dramatically and we will convince her that you do not need to shower right now :)
18:25
troy_s
He can get you the specs.
18:25
troy_s
LOL
18:25
troy_s
You get a color nerd. That's it.
18:26
troy_s
Seriously though, at this phase of design, the project desperately needs an experienced first.
18:26
troy_s
Shouldn't be too hard to find.
18:27
troy_s
rexbron has been interested and helped previously. Nag him via email.
18:27
troy_s
He will certainly be willing to help.
18:27
troy_s
now showers.
18:28
Bertl
se6astian: please be so kind and add a 'nag rexbron' task
18:28
Bertl
(with link to this irc log)
18:29
troy_s
I will email another now. See if I can get dims for you.
18:32
troy_s
Done.
18:32
Bertl
thanks!
18:39
gdims
joined the channel
18:40
Bertl
welcome gdims!
18:45
troy_s
Bertl: Some googling http://matthewduclos.wordpress.com/2014/01/27/iris-rods-a-simple-explanation/
18:47
troy_s
(I also didn't get into the impact remote motors would have on oopsieing here. :) )
18:48
troy_s
Bertl: To explain further, that blueprint that Duclos has is from the rod mount plates to CoL.
18:49
troy_s
The rod mount plates will land those rods precisely in those positions.
18:49
troy_s
Hence the requirement to nail the bolt position and lens height such that an off-the-shelf system can be immediately slammed onto the camera and used.
18:52
troy_s
19mm rod plate http://www.abelcine.com/store/Chrosziel-401-F235-Bridge-Plate-for-F23-F35-F65-ALEXA-with-19mm-Rods/
18:52
daFred
left the channel
18:52
daFred
joined the channel
18:57
daFred
we have this arri sketch in the wiki since 2012. ( https://wiki.apertus.org/index.php?title=Rods ) you are welcome to add information to this page...
19:02
troy_s
http://www.ocon.com/inspiration/labs/rod-standards-explained/
19:04
troy_s
daFred: too many years around libre software; I loathe wikis
19:05
daFred
so you can use http://lab.apertus.org
19:07
troy_s
Sure.
19:07
daFred
I think it's hard to collect infos from the irc logs :)
19:07
troy_s
I will leave that for more able bodied types.
19:07
troy_s
It sure is.
19:08
troy_s
And it is not pleasurable adding stuffs to wikis and other things.
19:08
troy_s
Sort of a double-edged sword.
19:08
daFred
what should we use? dropbox?
19:09
troy_s
I think Phab is great. But the folks that need to know have read it.
19:11
daFred
there will be help: see @13:15 se6astian -> "Live screencast"
19:12
daFred
for most of us it's new too
19:13
daFred
http://lab.apertus.org/calendar/event/view/1/
19:17
intracube
hi
19:17
daFred
hi
19:18
intracube
I'm wondering if it would be worth adding a task on phabricator about OLPF solutions
19:20
intracube
even if it's treated as an optional extra, it should be a clear option when ordering the Beta IMO
19:21
daFred
sure, sooner or later we would need it.
19:21
intracube
rather than just leave it to users to figure out
19:24
se6astian
please add any wish/feature to phabricator you consider remotely useful
19:24
intracube
se6astian: yup, will do
19:24
se6astian
the discussion there will lead to a decision if we can deal with it already or at a later point
19:24
intracube
if there's any intention of using the Beta to shoot usable footage, it's essential
19:25
intracube
Philip Bloom has been quite vocal about the issue in his blog, etc
19:25
se6astian
there are also software solutions to reduce moire and step artifacts
19:25
se6astian
any info/links you have to back up the requirements please also add to the wish in phabricator
19:26
intracube
ok
19:26
Bertl
also note that finding suitable parts/devices is essential for any hardware addition
19:26
Bertl
(not just OLPF)
19:27
intracube
some aliasing artefacts would probably be really difficult to get rid of in post
19:27
intracube
like rainbow/false colour gradients on fine meshes
19:28
Bertl
nothing you can't get rid of with a sufficiently brutal gauss :)
19:28
Sonic4Spuds
joined the channel
19:28
intracube
Bertl: you don't mean a simple gaussian blur to the image??
19:28
intracube
o.O
19:28
Bertl
question: what does the OLPF actually do?
19:29
intracube
optically filters of high frequencies in the scene before the light hits the sensor
19:29
Bertl
well, not hight frequencies of light, more high frequencies of change
19:29
intracube
closely related to nyquist formula
19:30
intracube
Bertl: yep
19:30
Bertl
correct, so, basically what effect has that on the sensel?
19:30
Bertl
there are three cases we have to look at here:
19:30
Bertl
a pattern which is (way) larger than a sensel
19:31
Bertl
a pattern roughly the size of a sensel
19:31
Bertl
and a pattern which is (way) smaller than a sensel
19:31
intracube
Bertl: my knowledge here isn't so good...
19:31
Bertl
you agree?
19:32
intracube
I think if you've got a repeating pattern smaller than the sensels, the high frequencies will add/subtract in non-expected ways
19:32
intracube
giving 'rainbow pattern' effects
19:33
Bertl
so, what does the OLPF do?
19:33
intracube
AFAIK it essentially blurs the image very slightly
19:34
Bertl
yes, either that or it creates four versions of the image
19:34
intracube
so there are no spacial frequencies that would cause interference with the sensels
19:34
Bertl
(slighly shifted)
19:34
intracube
but again, I'm -really- not the right person to discuss this with :P
19:34
Bertl
in any case, it is a blur at the light level
19:35
Bertl
large enough to avoid 'single' sensel effects and small enough to be acceptable
19:35
troy_s
Interpolation is the primary determinant for false color
19:35
troy_s
(IIRC)
19:36
Bertl
that is probably true
19:37
troy_s
I have been looking into it a bit.
19:37
troy_s
Some are aggressive cheats.
19:38
troy_s
Hence why I think Glen Chan's idea to leverage the components against the result is damn genius.
19:38
troy_s
He proposes (for YCbCr subsampling, but same idea)
19:38
intracube
troy_s: any link for that?
19:38
gdims
left the channel
19:38
troy_s
That you âcageâ the result such that you can't get completely erroneous results
19:39
troy_s
(You need a gamut to do this of course)
19:39
troy_s
intracube: Google Glen Chan Toward Better Chroma Subsampling.
19:39
intracube
troy_s: does this rely on the raw sensor data to work?
19:39
Bertl
(and add it to the wiki please :)
19:40
intracube
troy_s: if user records over HDMI to Atomos recorder they'd be screwed?
19:40
intracube
as alias artefacts would be burned into the image
19:40
intracube
?
19:41
troy_s
Yes screwed.
19:42
troy_s
Yes. Exactly. Post raw is âbakedâ.
19:42
troy_s
The idea is quite simple and elegant really... If you visualize worst case on a subsample is stripes of say, green and black.
19:43
troy_s
If we assume worst worse (not realistic) of non-narrow band post spectral captured on sensor
19:43
troy_s
The green vertical rows would be totally black
19:43
troy_s
When we interpolate, the typical algos average / borrow / curve from adjacents
19:44
troy_s
So the non-black stripes would contaminate the no data region
19:44
troy_s
In Glen's idea, you would leverage the known data (green sensel) to trap the result in the acceptable defined range.
19:45
troy_s
Outlined in that paper. It could be pushed into the XYZ domain too to further the results.
19:47
troy_s
(In the case of a black stripe, the luminance of 0 would reveal exactly one color that is acceptable. ;))
19:47
Bertl
note that HDMI doesn't mean interpolated
19:47
Bertl
it could as well be raw over HDMI
19:48
troy_s
Hrm. Never much thought of that.
19:48
troy_s
Good point.
19:49
intracube
Bertl: which is why I said HDMI to Atomos :P
19:50
intracube
so imo still a strong case for a optical solution on the sensor
19:50
Bertl
well, even the Atomos supports 'raw' recording, granted, only with 8 bit which won't make you happy
19:50
Bertl
the reason we switched to YCbCr is because the Zedboard has a stupid HDMI design :)
19:51
troy_s
YCbCr isn't exactly debilitating.
19:51
troy_s
But the encode would sort of smudge the data
19:52
Bertl
again, depends on the interpolation
19:52
troy_s
YCbCr will always add crosstalk from raw.
19:52
Bertl
YCbCr with equal number of components?
19:52
troy_s
Hence sort of make it impossible to unwind.
19:52
troy_s
RGB raw to YCbCr would be crosstalk.
19:53
Bertl
you are basically saying that RGB -> YCbCr is not reversible
19:53
Bertl
(which I disagree, as it is a simple matrix operation)
19:54
troy_s
Depends on intent
19:54
Bertl
as long as you do not interpolate or decimate, you're fine
19:54
troy_s
You could of course pass raw RGB as YCbCr
19:54
troy_s
But if the intent is to go from raw RGB âprimariesâ to broadcast 709 YCbCr, then the result could be a matrix with offsets.
19:55
troy_s
Still... Crosstalk at the quantization level.
19:55
troy_s
So no dice.
19:56
Bertl
anyway moot point, we have no plans to transport raw over YCbCr :)
19:56
troy_s
LOL
19:56
troy_s
Goof.
19:56
troy_s
Good.
19:57
troy_s
Not goof. Typo!
19:58
Bertl
feel free to goof around for good here :)
19:59
troy_s
Bertl you goof!
20:07
bcallebaut
joined the channel
20:09
bcallebaut
left the channel
20:14
intracube
troy_s: if you mean this: http://www.glennchan.info/articles/technical/chroma/chroma1.htm
20:14
intracube
it doesn't seem related to optical moire/aliasing
20:14
troy_s
I think that is it.
20:15
troy_s
Of course it does.
20:15
troy_s
It is subsampling. Precisely the same issue.
20:15
intracube
it's aboout upsampling existing subsampled data?
20:16
intracube
which isn't the same thing
20:16
intracube
and not about prefiltering in analogue domain before digitally sampling
20:16
intracube
which is where OLPF/moire/aliasing issues crop up
20:17
intracube
as in you want to filter the source at less than half the sampling frequency
20:20
troy_s
The sensor is chroma subsampling
20:20
intracube
troy_s: correct me if I'm wrong, but the point where the light is spacially sampled is when it hits the sensor
20:21
troy_s
Only difference between the YCbCr and Bayer is citing
20:21
intracube
any filtering has to be done before this point (in the optical/analogue domain)
20:22
intracube
for example, if you've got repeating black/red bars at several times the spacial frequency of the red photocites (but not an -exact- multiple of)
20:22
intracube
you're going to end up seeing weird stuff
20:24
intracube
sensor (or anything after in the chain) has no way of knowing the true spacial frequency of the bars
20:24
intracube
anyway
20:24
intracube
shuts up for now
20:26
troy_s
intracube: false color is because of guesses at missing sensel positions
20:26
troy_s
Exact same issue
20:26
troy_s
Remember, the data off the sensor isn't really RGB, it is simply gathered post spectral numbers
20:26
Bertl
intracube: may I suggest to devise some practical tests?
20:27
troy_s
The citing (position) is guessing at missing sensel positions.
20:27
troy_s
Same issue Glen addresses in a portion of that work.
20:27
troy_s
Scaling and guessing. No magic.
20:27
Bertl
intracube: i.e. create some actual test pattern with a known test procedure, so that it becomes easy to evaluate effects and the benefits of certain OLPF solutions there?
20:32
intracube
"false color is because of guesses at missing sensel positions"
20:32
intracube
I don't know what that means
20:51
intracube
Bertl: it would just be a repeating colour pattern that's a slightly different spacial frequency to one of the sensel colours
20:52
Bertl
that's fine, as long as it is properly specified and the procedure to test with it can be recreated
21:11
wescotte
joined the channel
21:17
troy_s
intracube: Take a hole in an image. Now guess (via scaling or such) at the color in the hole via the surrounding colors.
21:17
troy_s
intracube: You can see why the color in the middle is wonk if the surrounding colors are saturated or something and the color in the hole is not.
21:19
intracube
troy_s: Bertl: https://lh4.googleusercontent.com/-7SjeYd-KE-A/VEwTpE5up-I/AAAAAAAAA64/x9CmTa-MnuA/w409-h722-no/sensor_sampling_nyquist.png
21:20
intracube
a) common sensor layout
21:20
troy_s
I am pretty familiar with sensor layouts right?
21:20
intracube
b) looking just at red photosites
21:20
troy_s
Exactly
21:20
troy_s
So now think that you have holes in your image
21:20
intracube
c) high frequency pattern shining on red sensels
21:21
intracube
d) gives false gradient
21:21
troy_s
Your lens resolves to those sensels
21:21
troy_s
And you have gaps in the data
21:21
troy_s
That is all VNG / Amaze etc resolves. Nothing magic.
21:21
troy_s
It guesses at the colors
21:21
troy_s
Bad colors come up because the values on say, two channels, is discontinuous with the values on the guessed hole gap
21:22
intracube
I still can't see how this can be fixed after the event
21:22
troy_s
Which is why I said to look at Glen Chan's work, because it guesses and restricts based on the color domain
21:23
troy_s
Ignore everything about what you think you know and think only in terms of gaps in data
21:23
troy_s
It is vastly easier to understand.
21:23
intracube
troy_s: this wouldn't really be an issue on a monochrome sensor, right
21:23
troy_s
The red gaps in your image, if it were purely red gradient, would weight the blue and green collectors accprdingly
21:23
troy_s
Exactly
21:24
intracube
but because of RGB sensor layout, effectively the digital sampling filter width is too narrow
21:24
troy_s
Because every single resolved unit of light records _all_ information.
21:24
intracube
therefore aliasing artefacts
21:24
troy_s
It is nothing more than âOk I have the value for green at this lens resolved spot, what is blue and red?"
21:25
troy_s
And based on your guesses, you can get bad (really bad) guesses
21:25
troy_s
Example is an agnostic sort of thing that just scales
21:25
troy_s
Say the actual color is a highly saturated color
21:25
troy_s
And luminous
21:25
troy_s
So green well at the site is very full
21:25
intracube
"<troy_s> The red gaps in your image, if it were purely red gradient, would weight the blue and green collectors accprdingly"
21:25
intracube
why would red bars register at all on green/blue sensels?
21:26
troy_s
But depending on the color, maybe the red and blue adjacent wells aren't saturated properly for that particular site and the guess would be _way_ off.
21:26
troy_s
Well the filters are non-narrow band
21:26
troy_s
As in post spectral data will fill the well for certain colors that are non - pure
21:27
troy_s
And illuminant and other such rubbish will refract / reflect differently (hence post spectral)
21:28
Bertl
it isn't hard to find colors which only register on sensel with a specific filter, btw
21:29
troy_s
Bertl: Disagree. Too complex.
21:29
Bertl
for the cmv12k, pick 450nm, 550nm and 650nm for example
21:29
troy_s
Bertl: The frequencies are totally post spectral.
21:29
troy_s
Hence not as easy as it seems. Try profiling art works on media to get a clearer picture. Virtually impossible.
21:29
troy_s
But not a huge deal.
21:30
Bertl
pattern could be created by LEDs which have a rather narrow bandwidth
21:31
troy_s
Sure. But your glass too would get in the way.
21:31
troy_s
It is sort of a fool's errand really. Watched too many folks try.
21:32
Bertl
I actually had very good results with illuminating the sensor via LEDs directly (see contraption :)
21:32
troy_s
intracube: Needless to say, you can see part of the problem in "ideal" stripes
21:32
Bertl
it would be relatively simple to project a pattern onto the sensor for example
21:32
troy_s
How do you know what is at the gaps?
21:32
troy_s
True. That might be a fun test.
21:33
troy_s
Although you probably need a lens in there somewhere to focus.
21:33
intracube
troy_s: I still don't understand the solution/workaround you're suggesting
21:33
troy_s
Still, an RGB sensor isn't spectral. Hence crosstalk and unidirectional.
21:34
troy_s
intracube: You look at the data and use piecemeal color to restrict the result.
21:34
intracube
troy_s: sorry, does not compute :/
21:34
troy_s
Once you know the luminance qualities of the sensor for example, via a profile, you know valid and invalid values for s given range.
21:34
intracube
I'd need pretty pictures to explain it
21:34
troy_s
So a degree of the false colors can be ruled out.
21:35
troy_s
Red and black stripes.
21:35
troy_s
The red fires. The blue and green do not
21:35
troy_s
In a typical up sample, you effectively scale the red.
21:36
troy_s
So our example, the red data plane is solid
21:36
troy_s
And if we upscale, we get 0.5 red
21:36
troy_s
Erf 1.0 red
21:36
troy_s
If our blue and green are 0.0
21:36
troy_s
The value would still be 0.5
21:36
troy_s
Pulling the value doen
21:36
intracube
"<troy_s> So our example, the red data plane is solid"
21:37
troy_s
Every red sensel is full
21:37
intracube
the point is it wouldn't be solid - there would be false gradients all over it
21:37
troy_s
Because it is a stripe image
21:37
troy_s
It would be full red on only the red plane
21:37
troy_s
If we consider green and blue
21:37
troy_s
It would be 0.0 at those positions, pulling our scale values down to say, 0.5
21:37
troy_s
Wrong!
21:38
troy_s
But if we know the sensor, we can say âRed can never be 0.5 when blue and green luminance is zeroâ
21:38
troy_s
(Grotesque oversimplification)
21:38
intracube
troy_s: sorry, I can't follow this
21:39
Bertl
troy_s: and slipery slope as well :)
21:39
Bertl
anyway, the argument for an OLPF is simple:
21:40
troy_s
Bertl: Indeed.
21:40
Bertl
take a color which is actually a mix between red and green for example
21:40
troy_s
I will leave it at that for now. Read Glen's piece intracube
21:40
Bertl
let's say it registers on the sensor 50/50
21:40
troy_s
Bertl: my point is with a profiled sensor those guesses are more accurate. :)
21:40
Bertl
now take a pattern which is about a sensel wide
21:41
Bertl
this will give you any color between red and green
21:41
Bertl
depending on where the pattern starts
21:41
Bertl
if you have an OLPF which actually smears the pattern in such way that it appears as uniform color
21:42
Bertl
then both sensel (red and green) will always give the expected result
21:42
intracube
Bertl: yes
21:42
Bertl
there is no way to compensate for this in software or post processing
21:43
Bertl
that said, good OLPF seem to be quite expensive, and can also be added to the lens system
21:44
Bertl
(or fitted into the lens mount at some point)
21:44
Bertl
and for typical setups they are not really required, because those patterns are rarely seen in the wild
21:44
intracube
Bertl: I don't know if the OLPF distance from sensor matters
21:45
Bertl
some of the effects can also be compensated in a real world scenario by using certain assumptions on the color (that's what troy_s is referring to)
21:45
intracube
Bertl: if you're shooting nature/landscapes - yep
21:45
Bertl
i.e. some color combinations or color changes are not to be expected
21:45
intracube
man-made environments on the other hand....
21:46
Bertl
a typical phone camera will have no OLPF and it does quite fine in man-made environments
21:47
Bertl
note that the problem is the same with any bayer pattern sensor
21:47
troy_s
Bingo.
21:47
Bertl
the only way to avoid it is to either use separate sensors (one per color) or to have stacked sensel (i.e. RGB in one pixel)
21:47
troy_s
intracube: It is more about the non-data than anything else
21:48
troy_s
Smearing the data is a cheat
21:48
troy_s
And has other inferior elements.
22:02
Rebelj12a
Thats what I get, missed alot. Hey bertl good stuff on the DSLR, great marketing tool. Include some standard dslr shooting functions and even a separate possible viewfinder attachment. I have more than a few photographers I know interested in an open source DSLR that could double for video. If anything Beta is almost made for it.
22:04
Bertl
no idea what you're talking about, but glad that you liked it :)
22:06
Rebelj12a
Its two separate emails for marketing and engineering both related I need to send yet. XD, im glad to see the idea at least is on your guy's mind.
22:09
Rebelj12a
Ok so besides derailing the topic OLPF is not fixing any deficiency in the sensor per se it is smearing the colors so to speak?
22:09
Rebelj12a
Interesting.
22:09
Rebelj12a
So its not that its more accurate its like its more averaging the colors.
22:09
Rebelj12a
which in theory is more accurate or pleasing.
22:10
Bertl
pleasing yes, accurate, no
22:10
Rebelj12a
accurate from a subjective point of view of a person watching the video, of course depending on circumstances XD
22:11
Bertl
there are certain patterns which cause the same problems with our eyes
22:11
Bertl
but usually they are different from the ones which make the camera sensor trip
22:11
intracube
Rebelj12a: without an OLPF you can get colour patterns like: http://photographylife.com/wp-content/uploads/2012/02/Moir%C3%A9.jpg
22:12
Bertl
with it too, just needs a different 'original' pattern :)
22:13
Rebelj12a
Ah yeah thats bad, but that is fixable in post. And for instance, the 5d Mark II and III doesnt have that problem as far as ive seen. Yet it also has an OLPF available on the market.
22:13
Bertl
if it really doesn't have that problem, then it has a non bayer sensor
22:14
Rebelj12a
Well for raw it has to use a bayer though. Wouldn't it?
22:14
intracube
Rebelj12a: not easily fixable in post
22:14
Rebelj12a
Ah well i havent ran into the problem myself, and I know for photos at least its easy. A simple tool.
22:14
Bertl
Rebelj12a: if it has a bayer pattern based sensor, then it cannot avoid this problem :)
22:15
Bertl
but it can paper over it :)
22:15
Rebelj12a
ahh
22:15
Rebelj12a
I see.
22:18
se6astian
time for bed
22:18
se6astian
good night
22:18
se6astian
changed nick to: se6astian|away
22:18
Rebelj12a
Night
22:34
daFred
left the channel
22:45
troy_s
intracube: The OLPF is a cheat based on interpolation.
22:46
Rebelj12a
ahhhh so for instance the olpf filter for the 5d Mark II isnt necessarily a good thing. Especially if it isnt made well I know OLPF can cause all sorts of problems.
22:46
troy_s
intracube: You really need to appreciate the impact of interpolation on the gaps. It is, in no uncertain terms, huge.
22:46
troy_s
intracube: It is _as huge_ as how you scale your YCbCr data. Huge. Huge. Huge.
22:46
troy_s
And you can get more or less moire / zippers / bad color based on the variants.
22:51
intracube
hides
22:58
troy_s
intracube: I want to get the cubic b prefilter on with some XYZ work post, as the prefilter _is_ frequency domain
22:59
troy_s
And my gut says it will destroy 99.9% of the current debayers on a number of axes.
22:59
troy_s
intracube: Can you code?
23:00
Rebelj12a
Oooh destroying things, is that good?
23:12
intracube
troy_s: not competently
23:17
__anton__
joined the channel
23:17
Bertl
wb __anton__!
23:18
__anton__
Hi Bertl :)
23:18
__anton__
Quick question: if I'd like to follow the project is there mailing list I need to be looking at in addition to IRC?
23:18
Rebelj12a
Blast destroying things is fun haha
23:19
Rebelj12a
Hey anton
23:19
Bertl
__anton__: I would suggest subscribing to the axiom-dev mailinglist, but the action is mostly here and on lab as it seems
23:19
Bertl
(lab = the frabricator thingy :)
23:19
Bertl
*phabricator
23:20
Rebelj12a
Need to get on the lab
23:26
__anton__
Heh, I'm supposed to be a competent computer user. I got a google account. Yet can not figure a way to join axiom-dev. I get as far as here: https://groups.google.com/forum/#!browse but then I can not find the group via search :(
23:31
wescotte
__anton__: https://www.apertus.org/mailinglists
23:31
__anton__
wescotte: were you able to join the group?
23:32
wescotte
__anton__: Haven't tried but will now.
23:32
__anton__
wescotte: that's my question, what exactly do I need to do to join?
23:32
wescotte
__anton__: no idea :) looking into it myself
23:34
wescotte
__anton__: I don't see any info on how to subscribe either... Maybe somebody here knows otherwise I bet if you just send an email to axiom-dev@googlegroups.com it will send you instructions
23:35
wescotte
oh wait nevermind.. That's "internal to team"
23:35
wescotte
so it's not a public mailing list
23:38
Bertl
just send an email to sebastian, he will add you
23:57
troy_s
intracube: It is probably easier to see the issues when you peel apart the data.
23:57
troy_s
intracube: Worth dicking in code.
23:57
__anton__
Bertl: thx
23:57
__anton__
troy_s: I've got (mostly a hobby) interest in an alternative body design for Axiom
23:57
__anton__
troy_s: So did you say camera bottom could mimic Arri Alexa?
23:57
__anton__
troy_s: I've found this http://www.arrimedia.com/downloads/download/116
23:57
__anton__
troy_s: it says optical center is 99.8mm over plate, there are two 3/8 holes 10mm deep, 40mm apart, also some weird oval hole there 7mm wide 8mm long 4.5mm deep
23:57
__anton__
troy_s: (and 15mm LWS rods seem to be built in..)
23:57
__anton__
troy_s: Is this your ideal camera bottom design?
23:58
troy_s
__anton__: I didn't day it could, I said it, given the audience in question, must.
23:58
__anton__
troy_s: :) what's that oval hole for?
23:58
Rebelj12a
You dont want to know
23:59
troy_s
__anton__: As well as hitting the existing specification for height to lens to coexist with the existing gear (that is _horrifically_ expensive to buy for most indies.)
23:59
troy_s
__anton__: I suspect locator. Let me look.
23:59
__anton__
troy_s: page 4
23:59
Rebelj12a
Hm indeed, although I can post the specs of my Redrock MicroFollow Focus That should be pretty standard
23:59
troy_s
Just looking at your quote that is locator.