#apertus IRC Channel Logs

that's not really the problem, and I think, for now, it can be done in the post processing (which wasn't done with this image)

guesst: agreed, btw, in the image you referred to with sebastian, I see a similar horizontal and vertical pattern in the almost black background as we see with the CMV12k

sounds good, there is quite a lot to do, where do you see your specialities?

Bertl: there is just noise from inadequate lighting

guesst: I clearly see 'fabric like' noise, i.e. row and column specific noise which isn't directly related to the lighting (I hope :)

I would not say that I am comfortable, at least not at your level. I did a fair bit with some Altera parts a few years ago.

it is a steep learning curve in any case, but IMHO that's part of the fun :)

danhanes: so, on the harware side, there is not much left for the axiom alpha (the prototype), i.e. we are fairly complete there

guesst: do you mind if I mark the issues I'm seeing and upload that somewhere?

because I'm curious where those artefacts come from, as we see them as well

probably best for me to start with some tools that are not on the critical path. I am sure there will be plenty to do on the FPGA side once I come up to speed.

Yes, I was going to ask if you were using the ISE or had switched entirely to Vivado.

do you have access to a zedboard or anything using a zynq?

I seem to remember a discussion on the '30 vs the '20 xilinx parts. Are you going to switch, and does the (affordable) vivado suppor it ??

great!

because they both support 4 high speed serial tranceivers

Haven't looked at the 7015 - i'll check it out. 6G per channel is impressive. 6G SDI ?

where you would connect a 6G SDI ? :) there is barely 3G capable equipment

yes, if we forget the price :)

which is not so much if you want to record 4k at higher frame rates

for example the Odyssey7 starts at 1300 USD

it hardly does 4k at that price :)

not saying those are the only ones, just as example

danhanes: so, in general, if something is unclear, please ask, I might not always answering immediately but I'll make sure to get back to you

Thanks Bertl - I'm downloading Vivado now - slooooow. I look forward to getting to the point that I can help. I'll try to minimize the questions.

this is especially true for FPGA related stuff, I've spent a lot of time already figuring out trivial things, no need to repeat that without good reason

Seems the Zedboards are in short supply - 3 weeks.

I don't think that it really matter much, unless you plan to build/buy a sensor frontend as well

for example the zybo looks quite interesting

not sure what documentation is provided though, as it isn't open as the zedboard

I suspect I will need a little hand holding getting the platform up at first. I seems that is always the hardest part of any development - getting the tools working.

HDMI seems critical

That's encouraging

the hardware driver is a little more tricky (for jtag) to get working first, but that is documented on the wiki (now)

mageia, but folks reported ubuntu and fedora as working as well

they are quite useful in my experience, and they should for example work just fine on the zybo

it might be a good idea, not sure, maybe the tools work on OSX as well

yes, I've used them in countless cases to debug/visualize/check stuff

OLS ?

they are cheap and not very powerfull, but easy to handle and they work just fine for simple stuff

Haven't seen that site before

btw, the PMOD boards are available via OSHpark (the PCBs, you still have to solder on the parts)

yes, you can directly order a pack of three (or multiples) for a really good price

0603 LEDs ?

Thanks

http://vserver.13thfloor.at/Stuff/AXIOM/pmod_assembled_top_600dpi.png

Good stuff. I'll start setting up a dev environment.

yes, KiCad looks like it is getting somewhere

I would have used it, but I've been already working with eagle for quite a while now, so it was my first choice to get things done quickly

Probably best to use something you are familiar with. If I remember Eagle is only $US 200 or so for the non-pro version. Is that what you are using and is it adequate? 120mm x 100mm boards ?

Better yet.

but yes, the simple commercial version is quite cheap, you can get one for less than 100 EUR IIRC

Altium is nice, but overrated.

Wasn't aware of CERN involvement. That would be a big plus.

http://www.ohwr.org/projects/cern-kicad/wiki

you're welcome! :)

yes, I sleep occasionally, but I'm not in sync with the local timezone

Just as well, I suppose. Thank you for the introduction to the project. I look forward to working with you - hope I can help.

hey troy_s! how's going?

\o/

quite well actually, I've revised the LVDS part, and should be done with that in the next days

Voltage or otherwise?

Bertl: Hrm. I have seen the columns...

the row pattern will probably be easy to fix with the black columns, unless cmosis screwed up big time

the column pattern I'm not sure, but we will be able to address/compensate that at least in post processing

the 'row noise' is mostly caused by bad design on the cmosis side, there are known fixes (i.e. so that you do not need to compensate them)

Meaning?

according to cmosis, the sensors internal voltage reference is insufficient

an external reference can be applied, compensating for most of the row noise

(note: there are two kinds of row noises)

I don't think they will change the sensor as it can be worked around in external hardware

Is there a method to fake this fix in the prototype imaging pipe?

note that we haven't done any post processing in this regard yet, but most of the noise is either systematic or static

so I'm confident we will get 'clean' pictures soon

Which is?

this is with different exposure times (just a simple histogram)

Bertl: Which hump scare you? The middle?

you see that there are actually two maxima

but looking at the samples separated by color in a closeup:

shows that it is some kind of sensel effect

we haven't done any analysis what happens if we remove the 'hot' pixels yet

maybe they cause the effect

I would expect an almost gaussian distribution on a 'dark frame' as was used for this

and I definitely wouldn't expect it to show values aroun half the brightness after 10s dark exposure

so, IMHO there is some stray light involved

And do the results compare on the perfect sensor versus the imperfect?

I can get some raws.

But it would require at least a dcraw pass.

yeah, I know, they aren't really raw at all :)

yes, the issue is a little trickier than that but we are working on a solution

I mean who knows... it could even be some sort of interference.

I suspect a gap between sensor and lens mount or between lens mount and bajonet ring

Thinking on it

So in theory, if it is errant light, it will hold according to exposure.

the distribution gets broader

i.e. the average width will be wider (statistically/gaussian)

That is Greek damn you!

http://vserver.13thfloor.at/Stuff/AXIOM/RAW/hist.svg

I did that way back to see how much external light affects the sensor

Oh yes.

i.e. that is a mathematical function which is the result of statistical distribution as it happens for example with thermal noise

now what you already see in this simple test is, that as the top of the 'bell shape' moves to the right, it also gets slightly smaller

this is a property of statistical distribution

All I want to know is whether that cat is alive or dead.

i.e. the bell shape gets wider and less prominent

yes, the shape and the fact that they are around 1500-2000, because that is already half the range, i.e. 50% gray

(or at least it should be, not 100% sure actually, sebastian did the evaluation)

its more in the distribution

like 40% of all pixels are like 25% more sensitive than the other 60%

And is this a per channel eval?

?

Because when I look at that graph, the channels match perfectly.

Except that for some reason the voltage goes batshit in that middle exposure range?

the optic has an IR/UV cutoff, but that won't help with stray IR of course

Or temperature?

we keep about 4-5 degree above room temperature

What is your guess and have you done a distribution analysis?

but my current focus is on getting the image capture solid, which as I said should be done in the next few days

If it were errant light I wouldn't expect perfectly achromatic patterns.

in the IR range, the filters are basically all the same

so between 800nm and 1100nm you get almost the same response regardless of the channel

Right... so it could be IR from an LED perhaps?

that's why I'm disabling all leds (all but one faint green led) in the next version (from the zedboard)

hehe, yeah wouldn't be a problem, but it would require to a) get the boxed alpha prototype back to me and b) mutilation of an otherwise perfectly fine zedboard :)

and as I said, it is quite dimm and far away from the lens mount

Well fascinating.

I suspect the images he has posted will be underexposed... that said I have yet to check them.

yes, please keep up the input regarding exposure/saturation/etc

The nature of color dictates that yellow greens etc live much higher in the luminance compared to blues and reds

So for a matrix to get a decent handle on the transform, the reds / blues need to be amped up quite high to red sufficiently.

Yet he appears to trust the kinos.

hehe, yeah, well, the meter was calibrated a bunch of years ago and never checked again, as fas as I know

so there might be truth in both

So perhaps he can pop a bracketed set using a decent tungsten that has been metered.

Anyways... that Goren article is very solid.

great! so please talk to sebastian (either via email or here on IRC, he should be awake/around soon) and let him know what the best approach would be

Right now, the best thing is to get a big brainer like you to eliminate the errant noise issues

I've seen that in my scene* images

But noise is a huge issue. Then I would suspect getting a decent interpolation. Or perhaps just use some sensel skipping and use a ‘pure’ set of sensels.

yeah, don't worry, I'm confident we'll get that sorted, and for the calibration, the noise shouldn't matter at all

that's why I was so fond of daylight (suddenly :)

We can always gaussian the patches a smidge if need be to normalize the metered regions.

btw, I was wondering, from the color PoV, is there any point in using (somewhat) pure sources of color?

Bertl: Pure?

Oh... that might be interesting to experiment with. It is a little like audio though.

We want a perfectly straight line similar to what tungsten or the sun offers. If we use many ‘humps’ of light, where the crossovers land can mangle your evals.

It is all somewhat arbitrary as color is really synthetic; a psychophysical phenomenon as the color nerds call it.

Must install another kitchen strip of lights. Joy.

will do, have fun and thanks in advance!

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

hello!

how to setup a test lab for cheap : http://www.imatest.com/docs/lab/#lighting

as we know the way it makes light, burning a filament, a bit like sunlight, and provides the most efficient color rendition of articfical lighting. Even if kino tubes are rated 90% color rendition index, it's nowhere near the 99% we get with tungsteen. Imho

Bertl, will you able to provide the external vref on the current prototype's hardware?

HI here

very interesting chat last night

morning everyone!

heya

and yes, I agree, tungsten or daylight is probably our best option for the IT8 chart atm

good night!