#apertus IRC Channel Logs

troy_s:

yes, these two parts had to be separated, only a one-pass grading, something "light" might be applied, just to be watchable

FergusL: How did the editing -> grading pipeline work on your small projects?

ha cool !

will see, maybe I can make some pictures tonight, sec, let me check

and below that 2010.06

well I think that's the very same troy_s suggested we get and offered to source

great news

if I can focus on it, maybe i'll have the first resule of VNG debayer on your raw captures

maybe a well-balanced high CRI CFL bulb could do

wuh ?!

CRI = color rendering index?

yes

CFL bulbs sold for growing plants are full spectrum aka high CRI

but it's still a hacked solution

better wait from someone else approval !

I'm assuming we want the purest source to light the chart

http://en.wikipedia.org/wiki/File:Solar_Spectrum.png

450 - 700 or around, but I guess you know about all that ?

and it makes sense to assume that calibration charts do only address the visible spectrum (at least for normal imaging and moving pictures :)

so the main question is probably, do we aim to recreate sun light or do we aim to have a certain 'color temperature' white

and especially with normal imaging, the actual source isn't that much what matters, what matters is how we calibrate our sensing device !

yes, color temperature

so all we need is a furnace which can do 5780K :)

I wonder if lamp manufacturers specify the color temperature

http://www.narva-bel.de/ressourcen/spektrum/popup/SP_958_NARVA_BIO_vital_T5_RGB.jpg.png that's what I called "CRI chart", I meant a graph actually, it's as stupid as that... just a graph of how the source performs accross the range

can we assume it's betweel 3200 and 4000 ?

does the 958 NARVA BIO specify that?

ah, I get it, this _is_ the spectral output of this specific light source

http://www.amazon.com/Hydroponic-Full-Spectrum-Light-5500K/dp/B005P29K1S I was thinking something like that

now if we knew the spectral response of the sensor (CMOSIS hasn't disclosed that yet) we could calculate the spectrum of the light source :)

CRI is to be read as a "distance from 100"

http://housecraft.ca/wp-content/uploads/2012/09/spectral_responses2.png

yeah, it looks like it has the 'worst' spectral distribution

Bertl: You want a D65 at best. A tungsten 3200. Or at worst, a Kino tungsten or daylight.

but isn't that really harder to find now ?

sounds good, but I guess they will be expensive

ha ok

that's what I suggested with the slit/prism

why is D65 the destination ? the sensor ? the screen ?

just an habit ?

okay, where would I get that?

FergusL: Because stretching the white point adversely impacts ranges of colors via metamerism.

They will likely give you one for free.

equipment means film/photo equipment?

yes, a cinema/video gear rental jouse

You do not need a large light. A pocket par would suffice.

Just make sure the color temp of the light is noted.

so the store should know the exact color temperature

But even a ballpark set of test images is ideal.

Ask for some small shreds of 1/2 and full CTB.

(they will burn out quick so you will need to work briskly once you put it on the barn doors)

but for the size of the setup 800W is huge overkill

I'd say the 120W halogen lamp is already overkill

(HA, time for 3:30am breakfast !)

You need a movie tungsten fixture to assert 3200~

that will give a first idea till we have proper lighting

Just use _one_ lamp

camera

small piece of diffusion and color on doors

super simple

Redhead is as small as it gets

you can get smaller with a fresnel based fixture

but I am certain they will loan a redhead out for nothing.

barn doors = metal doors that slide onto fixture to control the light spill

you normally clip (wooden clothespegs) gel / diffusion to the doors

okay, and 'color' means?

hence "on the doors"

CTB = reference blue gel to cool down (increase color temp)

ah, I see, so you use that to 'condition' the spectrum, yes?

they used to be for migrating tungsten to daylight or back from daylight (hmi) to tungsten

Sense?

Yes. In the old days there were only tungsten balanced stocks and daylight.

So you had to balance for only one if you wanted an achromatic appearing source.

(yes)

indoor lighting is tungsten generally. Night as well.

anyways... it doesn' t matter too much.

and get equal illumination

how well calibrated is a typical picture camera?

e.g. an EOS 10D, do we know the spectral response of that one for example?

but the transform is often baked in low level

Keeners profile their own camera and create either a matrix (again most of the range is linear so easy transform 3x3 matrix) or

Yep

then calculate the transform that gets from A to B

hahaha, my raw to EXR script has been taking a full CPU thread for about 2 minutes now... still working

so basically I need the spectral output of the light source

Bertl: Movie lights are pretty damn close

(again, they _do_ drift (tungsten shifts as the bulb ages)

but by and large you can figure out where they are

Yes

and tungsten is well... tungsten. Physical property of getting it hot (age caveat aside)

Bertl: Feel free to hit me via email.

troy_s: okay, thanks for the information, I appreciate it!

Bertl: But to be honest tungsten is probably a better choice if you can snag some CTB slices

(having stills at various color temps is great)

okidoki, will see what I can do

A redhead with a bit of diffusion and CTB is for the win.

Most rental house peeps are very cool, and always looking to help out indies.

If you treat them well, you can forge a good start to a long relationship (and trying to develop a camera mean you might want to try and keep them on the good side.) :)

that's probably a job for sebastian then, he will know where to ask, thanks for all the input!

have a good one!

Bertl: The gamut is something I am keen to find out.

I wonder what's a decent time for debayering a 4K image with VNG

that sounds a little slow :)

my implementation is silly too

it's not simple interpolation

high quality white :)

actually is *is* working

Pixel 6 , 0 done.

G: 349.375

can I assume /4095 ?

the other way round, 4095 / calc_val

when you shift it right by four, the range will be 0x000 - 0xFFF (so 0 - 4095)

yep

but it doesn't matter if you take 4095 or 4096 :)

if you want to normalize the range to 0-1.0 you have to divide by 4095 (or 4096)

depending on how you define the 1.0 value

i.e. if you divide by 4095, you will reach 1.0

if you divied by 4096 you will get the 'lower end' of a 1/4096th interval

i.e. the center would be at V/4096 + 1/8192 :)

but, it's not completely fucked

yes, I'll have to do that

)

HA, self_20ms isn't Y flipped

I can't tell because mine is indeed flipped

that's quite possible, the flip should be recorded in the sensor register dump

http://www.pasteall.org/47018/python

hardcoded at some point where knowledge lacked

I should also forget python and do it in C++ now

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

hi

how did the rest of the evening go yesterday?

hi sasha

I saw. Very nice :)

yes I saw that se6astian, impressive !

especially regarding lighting

Hey FergusL, I was watching that conversation earlier today.

yes

you're both living close by ?

ha cool

to light the chart for exampel. an HMI/daylight balanced source would be the best but a tungsten-halogen would be fine with CTB (blue gel) on it

houses*

that should be pretty color accurate

morning Bertl

how long did you stay at the happylab? :)

oh no, why not?

:)

and after further investigation, we concluded that the cooling must be insufficient

mars_ will give you the details when he finds the time

and then we waited quite some time for the lasercutter, but around midnight we decided to do the cutting another day

but don't worry, I'll take care of that at the end of the week

maybe the cooling liquid was not hitting the mill tip properly? we tuned it down to only spray a little amount of cooling liquid... maybe that was the problem

why not yes, so what was the fruitful converssation about?

anything else would have disappointed me ;)

very nice, I would love to hear a summary sometime soon

rest assured, you will :)

and might come to IRC at some point, name: Jehan

cool se6astian, do you have a color meter ?

yep

yeah I know. It's not the best reporting, but it's spreading the word for our cause

note that I didn't expect anything else though

Hello guys !

unfortunately I have no color meter, I just got a light meter that I will forward to herbert when we meet the next time

last time I checked the only mode for lossy DNG compression was based on JPEG for bayer pattern and limited to 8bits

Sasha_C: nah, not worth it, I guess any news is good news ATM

you can always create a jpeg if you don't care about artefacts

just that at some point, it might be interesting to have some support for light compression

loosing most of the benefits of raw

how did the milling end up ?

because with this test we're going to calibrate the white balance for the camera

(this is what I understood from troy_s explanations yesterday)

yes

I don't really see any practical use for other temperatures, since it can be corrected in post

se6astian, too bad :-(

great

I have a kinoflo with daylight and tungsten tubes

yes, but will the "5500" or "5600K" marking on the tubes might be enough for such precise calibration ?

-might

color temperature, if slightly off, on the whole surface, is not a problem

that's not what I meant

what are you trying to calibrate?

but the kinoflo tubes are the most accurate light sources you can get

we're going to look at R G and B values for a known white, the known white is the one on the chart, we look at the R G and B value captured by the sensor. We then shift the values so that R = G = B, this is white balancing, at this point we know "how much" to shift from the values read by the sensor to get to <Temperature>

but you do the same shift for all the pixels, right?

yes

you might be 1% off, but you won't see it

which imho, means that kino or similar lighting will be good enough

it's in the equation that the issue is : we have to start from a well known base. If we assume 5600K as a base colortemp from the kinos but that it is in fact 5587K (or even 5595K) (depending on age of tube or other factors), we will *constantly* have an unwanted 13 Kelvins shift

(also I might be wrong somewhere, I'm waiting for Bertl's and troy_s's thoughts on this)

I think it would work

off to lunch, brb

Tungsten is probably smoothest. Kino is ill advised. Too many gaps in spectrum.

And no, a color temp meter to keep things simple. Again, the guys and gals at the house can measure the values.

FergusL: There are basically three points white balance enters into things. 1) When shooting. In this case, tungsten is probably the only decent option due to full spectrum with no spikes.

3) Destination white. Another Bradford here.

1) Show up at house for redhead lamp. Measure color temp using their meter. Note it.

3) Slap full CTB on the doors, measure temp and note.

4.1) One at native tungsten output.

4.3) One with full CTB.

you might need what you've missed, pasting log in /query se6astian

what is a PCS conversion ?

and a Bradford ?

There are several ways to move the white point through XYZ and transform it. XYZ scaling (worst - aka Wrong Von Kries), Von Kries, and Bradford

I think I see

XYZ is all unbounded. Think of it as an infinitely large cube rotated like a diamond. XY and Z grow out and up

got it

https://wiki.apertus.org/index.php?title=Camera_Color_Temperature_Calibration_Procedure

there's something wrong in my implementation of VNG, it's all greenish !

se6astian: No credit thanks. I just make it up.

troy_s: I'll just leave my questions here

afaik, white balancing is being at "pure white", there's an exact term for this : equi-something white

it is merely the origin - a point of achromatic.

se6astian: Ultimately, after profiling, you get a pretty good idea as to the primaries of the camera - how the triangle base (and the 3D volume) of the color gamut of capture is.

FergusL: Again, white does not exist.

it is merely a point of reference

Think of it as a map. If I say "Hawaii is six miles south" and you don't know your starting point, all of your map is wrong.

You must know it precisely.

yes, I just realised that

so that 3200K is R = baseR + R, Gbase + R etc... calculated as distances

No. But think map again.

and we take a camera shot of it

so, finally done with useless stuff for today

hurray, hail the tax minister :)

(so I think it can be considered useless work :)

I'm looking for existing dng writters libraries

gabe is working on a dng writer and has written a dng reader AFAIK, probably best to ask him

troy_s: we don't, so we need a reference !

s/skip/postpone/

troy_s: I remember using ENG cameras at school, there was this WB button, when you pressed it, the white became white

In current state the more images we get the better, and as soon as we can play with them, we'll get a better idea

and doing a setup to get a proper picture (lighting, position, sharpness) is somewhat complicated

any image will do to judge based on very subjective aspects :-)

that's why I'm being helpful: coding

dinner time, later

I think with the latest "self portrait" images, there is already a lot to do

what are the next steps coding-wise?

FergusL: here is a nice explnation : http://www.cambridgeincolour.com/tutorials/white-balance.htm

you can bang your head against the wall while I implement it ... after all, the news people write the truth, don't they?

Time to leave for me. See you later everyone !

I wonder whats going on there

not from "inside" facebook

Digitalbolex Joe Rubinstein sees the times are right for small companies building cameras: I’m not saying the big guys won’t be successful, but the Axiom project could very well be successful too.

time for bed