#apertus IRC Channel Logs

evening

the new images

yes

great!

ok

well, obviously a Macbeth is roughly the same idea

a color checker

send me one and I'll put it up there, it looks like it has the right size and weight

but it's definitely something needed too for camera engineering

still not something I will order myself :)

well, as a donation to Apertus I can still order one and get it sent to your home

I was suprised how good the colors are even without white balance and corrections

yes

a lead grip in the USA but who knows a huge lot about color theory, video signals, colorspace science etc...

since a cinema camera will require exposures at least that fast

you know the guy I just refered to :)

but I need a lot light for that :)

(or in the 20-41ms region)

Well, that's a little subjective

You wouldn't happen to have any other cameras to compare with?

-e

It wouldn't tell anything about the sensor's sensitivity and noise?

(of course I wouldn't be surprised if the RAWs from existing cameras have some post processed filtering going on)

he

this is basically with the cap on (and no black level adjustments)

yeah, well, not sure that works that well, as the camera part is still dark (except for the image)

From the existing pictures it's not possible to tell how bright/dim the subject is.

(showing an uniform white screen)

not yet, but let me describe it to you

the base itself stands on my right side in front of the LG TFT which is placed about 20cm higher than the zedboard

to put the paper at the right place/position I built a simple grid out of 10x3mm wood (used for model airplanes)

in such way, that it also has a 30° angle compare to the table

I see

I can probably find a neon light as well if that is of interest

because the flood light is very yellowish

that's why the white paper looks brighter on the left side than on the right

http://vserver.13thfloor.at/Stuff/AXIOM/RAW/colors_e5ms_flood.png

(I can upload the raws for that as well if there is interest, but it will take some time)

I'll upload the sensor settings for clarification

great to see this, let me take a look !

here the same scene illuminated by the TFT (5ms)

http://vserver.13thfloor.at/Stuff/AXIOM/RAW/colors_e5ms_18Wneon.png

(I changed neon to fluorescent in the other url as well)

http://vserver.13thfloor.at/Stuff/AXIOM/RAW/colors_e40ms_18Wfluorescent.png

I could note that

at 5.6 aperture

I'd go with f/5.6 !

are there a lot of available options ? are these what the complete sensor documentation refers to ?

there are two exposure times, each 24bit

then there is a register to set the number of slopes

(where bottom and top refers to the LVDS readout channels)

with two knees

then there is a PGA divider and PGA gain (that is the analog gain)

and at least 8 registers which are somewhat involved in the imaging process but do not have an explanation

okay, I see

so you can do up to three slopes with two kneepoints

and you can set the voltages and exposure times for each of them independantly

I guess it can be very useful (similar to HDR for photo cameras)

but I'm worried about some artifacts it would create

from the way how it is implemented, I do not really expect any artefacts

there might be issues with neighbouring pixels used to Debayer

like, one pixel was captured before one knee

and there it is, afaik, the Bayer calculation is messed

because it's maths ! it's assuming it's in the same "space"

but maybe I'm misunderstanding it

absolutely

but even if we are talking a really big difference between neighbouring pixels (i.e. a huge contrast) I don't think that it will matter much

but that is something which can be tried and tested via simulation

i.e. you do not need to use a sensor for that

all this will be crucial knowledge for linearizing

familiar with that part ?

(well I know linearization but in the math context :)

like, if you apply a gamma correction

you *have* to tell the computer/software about it in the most precise manner so that it doesn't assume anything

totally-totally-totally

(or the resulting curve measured/tested with a known good gray table)

that's all great

dmj_nova: http://vserver.13thfloor.at/Stuff/AXIOM/RAW/colors_e40ms_TFT_f1.8.png

like, both digital and analog gain ?

it is more a sensor setting which depends on the bit depth

i.e. you probably want to choose 12/10/8 bit, but not the digital gain

but as it is open source and open hardware, you're free to change it anyway :)

the sensor can transfer data in 12/10 or 8 bit chunks

the offset is basically one adjustment you can do

I understand "digital" better now

i.e. if you have 8 bit, you want code 255 represent the highest/lightest value and 0 the lowest/darkest one

and while analog gain gives you a 'better' range, digital gain just throws away bits :)

Yeah digital gain is unlikely to be useful

better to save some metadata describing the color transform to perform later

it is really tricky to focus ATM, I have to get the hdmi out working soon :)

focusing at F/1.8 can be challenging

I guess there are focus marks on the barrel ?

and that you know precisely the distance, but indeed that's not enough

(not sure they are related :)

hmm, that is on the other side IMHO

let me take a picture for you :)

http://ecx.images-amazon.com/images/I/51CnjgPJ%2BjL.jpg

so, the 22 means, with f/22 I get about this range, yes?

okay, got it

thanks for the explanation!

and do you also know what that tiny switch is for (on the top right in the picture)?

hm

yeah, that would make sense, as it has AF (according to the spec)

yes

with longer exposures, something strange starts to happen

and do you also know what that tiny switch is for (on the top right in the picture)?http://vserver.13thfloor.at/Stuff/AXIOM/RAW/colors_e1.5s_TFT_f11.png

http://vserver.13thfloor.at/Stuff/AXIOM/RAW/colors_e1.5s_TFT_f11.png

yep

that were 4K

not sure about that, looks more like some kind of statistical effect

oh

i.e. certain pixels reach some kind of limit and then go bright

OTOH, exposure times of more than a second are probably quite unrealistic for a movie camera

nope, no calibration, nothing, just raw data

yes, definitely

I think it sould be done everytime the camera starts up, some kind of light calibration

what's the procedure?

okay, no problem, just I can't think of any sane calibration which could be done without some kind of reference input

here you can see the same pixels are brighter

could it be sync issues in the charge/discharge cycle ?

I mean, at a video rate

ok

http://vserver.13thfloor.at/Stuff/AXIOM/RAW/cap_e3s.png

I'm pretty sure you can substract that from the other images

http://vserver.13thfloor.at/Stuff/AXIOM/RAW/colors_e3s_TFT_f22_clean.png

dust on the bottom ?

so that is definitely something instrinsic to the sensor

yes

anyway, I'm off to bed now ... so you win tonight :)

haha, I got it

you won a bunch of fancy pictures you can show around *G*

anyway, thanks for the chat, it was quite informative

thank you, you too

hi sasha

Hi Sebastian. I saw the email train today, yeah!

now that herbert has cleaner images

How are you today / tonight?

hope you're able to handle everything that's getting you stressed

Using http://vserver.13thfloor.at/Stuff/AXIOM/RAW/colors_400ms.png as an example, I can see a pattern noise (in the shape of vertical and horizontal lines) really evidently on the yellow band

mars_,yes that would be great, maybe you can bring them when we meet on Monday?

yeah, i can bring them with me

morning everyone!

Sasha_C: I think we still have a bunch of different effects

which in case of our sensor connection and read-out will give a two line pattern

or more general, even rows bottom, odd rows top readout

hi troy_s :)

ah, lots of new? folks on the channel

but there is a slight timezone shift as he is in North America

that's fine, we are usually up late anyway

yes

I just got up, had breakfast and I'm working my way through my inbox :)

http://vserver.13thfloor.at/Stuff/AXIOM/RAW/colors_500ms.png might be new depening on when you set the border to "yesterday"

hello se6astian, indeed the border for "yesterday" was this night at around... maybe 4 ? or 5 ?

se6astian: you should really get an irc client which is on 24/7 or some kind of irc bouncer

and I just ordered some sensor cleaning tools

http://vserver.13thfloor.at/Stuff/AXIOM/RAW/colors_e3s_TFT_f22.png (this is what FergusL is referring to)

ha cool about the cleaning tools

what did you order ? "swabs" ?

cool !

airblower

welcome neverlis!

morning Sasha_C!

thanks!

yeah, not sure that is the noise you were seeing though

I think there is a reason why the sensor has a mode where the first and last 8 pixels in each row are replaced by so called 'black pixels'

so that you can basically calibrate the noise floor/black level for each row

that sounds pretty useful, no?

you probably won't miss the 16 pixels out of 4096

I also suspect some to-be-calibrated differences in columns, as they are grouped in 128 col groups

So we'll have our work cut out for us trying to calibrate this sensor for a cinematic application

while it will make sense to do a really good calibration and pre processing inside the camera, I don't think that it needs to be done (right now) in the camera

ok, i understand

(and some kind of reference like the black columns)

btw, I think our web presence is rather bad, no offence to you and sebastian, but I'm worried that we are losing momentum ...

I agree, we should be more active

and all pages I get are either outdated or obsolete :)