#apertus IRC Channel Logs

Bertl: We could do a few things so that we can capture an ideal image.

Bertl: B) Output the mean error / DE error.

etc. etc.

but I still need to figure out what that error actually means

Bertl: It doesn't matter what it means

Bertl: Given that the image is sampled from a region etc.

It _could_ be related to noise, but my gut says that it is less of an issue.

no, it cannot

because the perfect chart has no noise at all

Bertl: That doesn't mean that noise isn't impacting the matrix.

Bertl: Example - noise happens quite badly in the red spectrum on a lot of sensors (reds / magenta)

Bertl: So of course that _could_ be weighting the swatches in an undesirable fashion. But again, I say we will know more when we can get some solid daylight exposures.

Bertl: It's blurred and buried.

the perfect charts are artificially created

Bertl: You are missing my point. You've pulled from pixels right? Those pixels have noise. Or rather... let's call it irregular firing sensels

But the sensels themselves are being filled up with data of some sort and some of that is irregular.

noise is deviation from a normal

*area

differences across color areas cannot be accounted as noise, they are 'the profile'

but yes, we can do it somewhat interactive i.e. chasing for the smallest number reported by the tool without knowing why, how or what it means

another thing we could do very easily (and interactive) is to check for uniform illumination

Bertl: You want a median average of the whole swatch. Which is why noise plays a role.

The uniform illumination is deadly easy to check with the TI3 chart.

File rather. A hutch is better of course because it has white slots around the image so you can EASILY tell based on the RGB sampled values.

Bertl: It will illustrate how much of the swatch it is meaning / averaging.

Bertl: At any rate... I should get the VNG code done then we are a step ahead.

Bertl: We want the _whole_ swatch sampled using various pixels.

I agree, but what I'm saying is that even with this, the results are the same

yes, but not because of the noise :)

Bertl: No. I only said that noise will play a role, but rather I'm not too concerned about noise at the moment.

I want to see a DE2000 of 3 or less.

yes, at the beginning I though the unusually high FPN causes the calibration tools to go crazy

I'm not sure. The problem with a matrix error is that

that's why I created the perfect charts in the first place

As in if you think of the colors as existing in fixed points in 3D cartesian space, the 'ideals' simply cannot be fit against what we are gathering.

The errors are ridiculously high.

(they were slightly better to be fair. SLIGHTLY)

Still super crap.

10 is "throw the thing away" bad.

Bertl: I'm going to say my guess is the absolutely crap samples. The tungsten is not working with the shift present on the sensor. At all.

The sensor has a HEAVY skew to reds.

Is there a way we can nerf the sensitivity of the various gains?

(The reality is that they are SO far apart that the green will fill up so fast that it will be spilling into the blue and red anyways causing nasty fringing)

Your daylight was way underexposed last time I looked at it.

I didn't see a significant gain on the daylight. What image was it again?

Huh?

Green sits way higher on luminance.

http://vserver.13thfloor.at/Stuff/AXIOM/RAW/scene_daylight_211ms_c2.png

Bertl: Is that a 16 bit PNG?

(very underexposed.)

yes, but in this case the red is clearly not dominant

btw QE = quantum efficiency

Which is why I'm suggesting daylight is mandatory at this point. I can't be bothered with tungsten at all (and heaven knows it might have an adverse effect in tungsten light after bradford shift)

Bertl: That's just saying that "Whatever the filters are set to will fill up with each photon smoothly"

Yep

(Hence my concern with the filters)

Yep.

(Think if we had filters that were all at strange angles of hue)

(green and blue rotated around to red)

Scares me.

https://github.com/apertus-open-source-cinema/alpha-hardware/blob/master/Datasheets/datasheet_CMV12000%20v1.12.pdf

the only additional filter is the IR/UV cutoff

Bertl: What's the imagemagick perspective command in order of corners?

pairs of 'source' 'destination'

Gotcha

Bertl: Your convert isn't working on convert \( -size 4096x3072 -depth 16 gray:IT8-tungsten-halogen-Run03-9.1.2014-15ms.raw16 \) \( -clone 0 -crop -1-1 \) \( -clone 0 -crop -1+0 \) \( -clone 0 -crop +0-1 \) -sample 2048x1536 \( -clone 0,1 -average \) -delete 0,1 +swap -combine c-IT8-tungsten-15ms.tiff

ON your scene raw file

Getting unexpected EOF

?

It's your scene_daylight_211

there you go, unpack it first

compressed

Bertl: Average error 13 on daylight.

Average DE 8.6

(Much better than tungsten, for a super underexposed image.

Bertl: Or medium role.

(It is definitely tinting the achromatic values)

Bertl: It gives me confidence that a well exposed daylight will give us good results.

*see

Lowest yet.

off to bed now ... have a good one everyone!

morning folks!

good evening

herro

Bertl: Still wonder about PCIe for sensor board. It just can't possibly be firm enough.

Perhaps a machined box around each, with PCIe exposed?

not that I think that PCIe is the optimal way to go

no idea what 'the days of backfocus' are :)

Adjust the planes between sensor and lens

Or wiggle

how so? it will be screwed to the lens mount?

Anyways... it worries me.

what would be your suggestion?

I would rather the camera perform focus as best as it can than be able to swap sensors.

even in completely non modular designs, no?

My only thought was that PCIe has two points of torque

And foo-to-PCIe

and that could end with a catastrophic effect on a device designed for imaging.

Modular requires an ecosystem. The audience that is keen to design boards and flip flop elements and such is, in the great archetype coordinate system in the sky, a vast matrix transform away from a (estimated!) archetype keen on imaging and cinema.

I don't have one and I never wanted a board with one

Given the audience.

Yes but you are a nerd beyond all mere mortals as well.

But on point, that sensor mount has to be butch as hell.

or just moving around with the lens mount :)

But yes of course.

(Random voting in is random noise out under the seductive illusion of information)

Bertl: What is your current short range goal?

the short range goal is to get the image pipeline finalized

Won't happen with what we have.

I can fudge one (an image)

(Se6's charts look blurry as hell)

The real issue is exposure is stinky too.

You don't have a chart do you?

but se6astian_ has two and can take images I guess

Bertl: Drat. Need math help.

yes

se6astian_: What's the exposure of those images in you rpack?

se6astian_: Next shot set

se6astian_: So that I have some reference as to what is one stop hot etc.

Yes

So did you have an idea where baseline middle grey average exposure was?

se6astian_: Good. I thought I was worried a bit when you said "So you changed your mind?"

se6astian_: I think that the 8-14ms range is probably about a stop hot.

And there was.

So hopefully we can figure out with live HDMI a better shot set

se6astian_: NO chance of a daylight set with refined exposures between now and Berlin?

se6astian_: Because it would be a crying shame if you guys don't have a proper matrix.

lol

I've got a matrix that is pretty far out of whack (DE2k of 8.5 which is our best yet)

well better than nothing I would say

se6astian_: I have OCIO working now with Blender and other such OCIO enabled apps so evaluation is quick.

8.5 is not better than nothing. Looks ... nightmare.

well currently we have no matrix and still output a HDMI signal, I doubt it will get worse than that

(Probably some whack filter / tungsten issue)

also I need to get lucky with weather/sunlight

ok, gotta go afk for a bit

se6astian_: Even cloud would be good enough here to evaluate.

se6astian_: Quick request... shoot the chart then pull the chart and show the background generic scene too.)

slikdigit: LOL. Howdy.

good! I'm gonna try to go to lgm this year, was wondering if apertus would be at lgm

But who knows. se6astian_ just ran out. You can ask him.

slikdigit: Hanging in there. Wasted way more time trying to get a decent profile out of the camera than I'd have wanted.

Wish Gabe were around... he'd likely be able to chip in.

or laboratory

so, troy_s, the camera you are making a profile for is an apertus axiom?

hi slikdigit, before I had out to bed, we likely wont be at LGM

but we will be in Berlin on Thursday

https://www.apertus.org/berlin-2014

troy_s, cloud cover noted, I will see what I can do

slikdigit: Are you familiar with color management at all?

only as much as you'd expect an end user of blender to be; I'm familiar with some of the concpets, and I've *seen* some of the math, but not really studied it or coded for it

I know enough to be scared by it and be glad things mostly work by themselves

so, I guess there is more than one prototype camera built so far?

slikdigit: Damn you didn't page me.

ah, well, I usually only page if I have an expectation that chat is gonna be realtime

slikdigit: Fail point one, especially in Blender, color _does not_ work by itself mostly. Especially in Blender, some horrible choices have been made and it is a mess.

Fair chance right now for that realtime thing. LOL

slikdigit: In a future hope, I'd really hope you take it upon yourself to learn the concepts of color. I'd be happy to start up an email exchange with you that you can back and forth with me on things.

slikdigit: Even more importantly is when they have rather massive projects that they work on and ignore how the pixels end up. ;)

slikdigit: Sometimes you have to know enough to appreciate the problem.

slikdigit: Some can't even see the errors when painting in a curved one. You've probably seen it and know why linearized data is the _only_ way to combine colors based on real-world physics.

true true

slikdigit: It's roughly like a confused rabble learning from itself.

most people are self taught I think around blender, so there tends to be a bit of that, yes

(the questions can often force you to think about things in ways you thought you understood things, but reveal gaps.)

however, I think it does work out in the long term (even if it causes problem in medium or short term) as mistakes tend to 'average out' and good ideas survive longer

ah, I was unaware of that

Check the comments of the ToS blog. There's a post where Gez says "This is continuing on the blender-vfx list.)

It _can_

but there are some HUGE holes that need addressing.

It's an interesting subject once you wrap your head around it.

And as an arty farty fool, i can't help focus on it when it comes to creating work. It's so fundamentally critical to what your audience sees. From posters to videos to images to you name it.

The color wheel is a big one for example.

I really wish you would.

Take it slowly.

For me, having walked probably 1% of the whole path

I eventually found a good lot of them

any good textbooks?

I think that's one of the issues for many people

yeah

I think that's an issue for people getting into this stuff

(JW asked meto do a CGcookie thing, but sadly shows have gotten in the way. I do hack away on The Hitchiker's Guide to Color for Digital Artists a bit at a time.)

Chat with people.

To be honest, once you learn the _basics_ the rest is all pretty damn easy to wrap your head around.

slikdigit: How's Tube?

I'll just make it black and white (0 and 1)

character animation will likely be completely done in march

since without finished textures we can't call lighting 'final'

You realize that even Black and White has color management? :)

that's why I avoid grey

LOL

oh shit! looks like I just finished tube ;)

Fresh questions are great.

Can't wait to see how the project ends up.

cool, thanks

well, fateh is planning a documentary project, and we got curious as to the status of apertus project

so I started reading up on it

Having a real life and a real job since independent life challenges all of that.

And it takes still pictures.

And Herbert just got HDMI downconversion working... no profile yet thanks to idiot boy here and the images so there's no real Rec.709 output. :(

we are still in the 'get funding' stage

slikdigit: I was rather skeptical at first, but Herb's a rockstar.

slikdigit: And there are some rather big brainer types.

slikdigit: So I'd think that it is the first sort of project that I think stands a hope in hell.

Thus far, most decisions seem pretty reasonable and well thought out.

Which is always positive.