#apertus IRC Channel Logs

hello sir

hello sir

did you choose a task and start working on one of the challenges?

Hi. I am doing the C++ challenge, can you please clarify if the input image is RGGB or GBRG?

sir i went through the link you shared

the green will end up on the red and blue channel ... you can also assume that the 'two greens' will be reasonably similar

@Bertl_o0 I think that even if we interpret it the opposite way the output image will stay the same? Is that correct?

humbe_coder: let's assume this is true, what conditions would the image need to satisfy?

well, in that case you might want to check out for easier projects to work on

no problem

@Bertl On second thought I think that the image will nog stay the same, because if green is interpreted as red n blue then maybe the insity of unknown pixels remain unaffected after demosaicing but the colours of the output image will change, also since there is no header in RAW12 images, how do I find if the image given is RGGB or GBRG

so the image would be a pure gray image

now assuming the image is not a gray image, then when you have the channel order wrong, you will still end up with basically R==B because the greens are similar

you result image if you put the channels together will be all purple - green

but that could only be verified with visual inspection of the image, right? everything will seem okey in code.

So I might be testing my code on RGGB image and someone else might test it on GBRG image and get a grey purpleish output. So how could I figure out which type the input image is in the code itself?

if you want to add a feature, add an option/switch to select the order

you can do that, yes

I don't know, you have to figure that out yourself

sounds good to me

that would be my interpretation of 'output'

so before debayering there will be 9 red pixels amoung the 25 so do I print the insity values of those 9 pixels or the red insity values of all 25 pixels after debayering? also I think that we will get insity values in 12 bit, so do I print the 12 bit values or convert it to 8 bits then print?

ok.. thanks a lot for helping me out.

Hi,Bertl

hey

Snap is for click a pic.

Question ; )

I tried to find on wiki but didn't get much

and train does train the LVDS channels (coming from the sensor)

Ok and in this sections7.1 https://wiki.apertus.org/index.php/AXIOM_Beta/Manual#raw_still_image_capture

In the fabric part whats need to change 12 bit to 18

And we work directly on the contents we get in memory ,right?

https://github.com/apertus-open-source-cinema/axiom-beta-firmware/blob/master/peripherals/soc_main/top.vhd (line 715-728)

and yes, T734 works on the data written to DDR memory

the input comes from the sensor, the output goes to the HDMI (typically)

Can we use ODDR and PLLE2_BASE in the serial to parallel converter in T871 ?

Amaterasu: what does the task description say?

just changed that to: You may use existing hardened units like PLL, DDR or SERDES to simplify the task.

Okay bertl thanks, i will read more and come back if have any doubt.

s/doubt/question/ :)

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

As mentioned in T734 , why isn't have to be real time capable everytime??

Alan_: if you manage to _make_ it fast enough to provide a full histogram for every frame that would be awesome!

Hello.

Silly question. Can I actually get one of these cameras yet?

I notice things like dead pixel correction and colour calibration aren't marked complete yet, though.

depends on what you want to do ... it is still a developer kit for now, so you have to be prepared for diving in and taking control yourself :)

never

Well, OK, but it has to hit a 1.0 at some point

Well, sure, but at some point it has to be possible to take it out on set and have it be usable and reliable.

How long has it been in progress?

the basic development has taken less than a year for the dev kit

you can also check the development status on the web site

I was looking on the website for some history info.

https://www.apertus.org/axiom-saga-history-article-november-2014

what exactly are you looking for?

we definitely did that with the AXIOM Beta dev kit (as you can see with the footage)

something you can easily do in post ... actually it is better to be done in post as you have more control over 'how' to correct dead pixels if there are any

well, that's completely up to people (you) to decide for yourself

there are many folks out there who do not like the 'papering over' basically all proprietary cameras do before they generate 'raw' data

modifying the data to make it 'nicer', hiding the imperfections

if you do not see any advantage of a completely FOSS/OH camera for you, then by all means, please by a cheap proprietary camera

I'm trying to understand what the advantage would be for anyone.

you can also adapt it to your needs and extend it where needed

Like Kinetta was supposed to have.

just attach a rotary encoder to it and have it control the shutter

not really, for this you are probably fine with a little python

Did you look at the Fairchild LTN4625

I think the bigger problem may simply be obtaining one.

But it's very capable.

Holy hell. CMV50000.

That'd put you on the map :D

Shame.

Ha. E2V. I live about a mile from that place.

the CMV5k is still available if you are willing to spend the 5-6k

Some camera manufacturer will always snipe it as an exclusive.

Oh, well. I've shot Cion. I've seen it all :)

It's a painful memory. I don't like to talk about it.

I'm not quite sure how it's possible to get such little dynamic range out of that sensor, but they managed it.

Hfuy: one specific example of where an open source camera has benefits; one of the BlackMagic cameras at launch would only record 16:9 even though the sensor was 4:3 (so the top and bottom of the frame couldn't be captured)

I *think* this may have been added sometime later? but don't quote me on that.

The 2.5K possibly.

The thing is, I like the idea of all this open source stuff.

Since most people aren't, to most people, open source is pretty much meaningless.

not necessarily, if other people can be encouraged to make a modification then everyone can benefit

My experience is that FOSS coders work on what interests them, not what interests users.

But the results for users can be... eh...

there's nothing to stop a few people creating a group separate from Apertus, crowdfund, add specific features

My go-to example is always Blender. Huge amounts of time spent on it, fantasic features, the UX is a complete war crime and I think a lot of the effort is wasted.

but I've certainly seen worse from commercial programs

By quite some considerable margin.

And I used Real 3D on the Amiga.

mainly because companies do not spend money on GUI development and testing unless it is essential to the tool

It's a big problem.

it is Cern OHL in our case

Hfuy: the solution is simple there, no? just pay the developers :)

reference: https://wiki.apertus.org/index.php/AXIOM_Beta/Case_Studies#Mavrx

It was suggested that I included Apertus

Apertus is still here

Hfuy: did you include that 8K camera (from Italy, iirc?)

I had one here. I reviewed it. Days before the company went out of business.

It was... not... very good...

But really the thing was barely functional.

Some of those did ship and then died.

Cion was stillborn.

I wondered that. Presumably they were desperate for money.

They're mainly fine.

Of course, that's just them going to a Chinese manufacturer with a spec list. But they work OK.

Ha. Bloom got it in the same shabby box I did. I wondered why they wanted it back.

I guess I could write about the axiom camera for one of the places I write for.

We do not 'send out' devices for reviews, but there might be an AXIOM Beta Dev Kit owner in your area you could contact ...

It might not be a bad idea to put together a demonstration kit that people could use for a while.

you are welcome to visit us in Vienna

where are you based?

Well. Half an hour out of London, but you know.

All I know about Vienna is that you can't get a cup of coffee without a glass of water.

I know someone who was born in Vienna, grew up in the USA and London, and now lives in New York because she says Vienna is boring.

aha, good to know that such people live vienna, so other people can enjoy the city

To be fair, I don't think she really lived in Vienna other than a year or two when she was a baby.

(Claims it's not German. Sounds like German to me.)

The language does sound quite a lot like German.

this video teaches you the only word you really need to speak Austrian: https://www.youtube.com/watch?v=iuXR53ex4iI

was already suspicious because of eastern accent

I think it's just German with a dozen extra words, all of which are different types of apple strudel.

Hi guys, anyone around?

I am debayering the CFA using bilinear approach, the red and blue channels seem fine but the green channel has dark pixels in it, also all three channels have significant artifacts

No I have not, the format is little endian I think. I am reading 8 bits at a time, three times and then performing bitwise operations on them to get intensity values

so.. should I convert the little endian input to big endian then perform binary operations on it?

maybe it's also big-endian

yes

yes

BAndiT1983: mail sent, please check

about your problems, usual way for this data is to do byte swap, then to can convert each 3 bytes to 2 sensels, as we have 12bit data stored there

if the byte swap is done and data correctly extracted, then the image shouldn't be corrupted, if it is, then check processing path again

see you