#apertus IRC Channel Logs

the axiom only has one hdmi currently

there are two plugin slots on the AXIOM Beta each with 6 LVDS pairs (potential TMDS channels)

so you can use two 'Single HDMI' plugin modules or you can use the triple display port plugin, a 'dual plugin module' and LVDS - TMDS converters for up to three HDMI outputs

okay, now I am getting things slowly!

unfortunately there are no 'high speed' LVDS to USB 3.x bridge solutions readily available only 16/24/32 bit parallel to USB 3.x

You make things so clear :D

of course, as mentioned before, it could also work the other way round, because the basis for the gearwork is the MachXO2 FPGA on the plugin

Okay I will think of complete mechanism and will discuss with you

you're welcome!

another question : I saw that the task updated and "pll" is added

I am now confused a little for "You can assume a clock with 100MHz and a fixed phase relation."

i.e. you usually use a PLL to generate the required frequencies, but it is also fine to 'assume' other clocks if it makes sense

I am assuming the clk is an input for my module

thanks Bertl

hi saurabh_raj, would go with 4 spaces for now

ok,, I tried to make modifications to my code as per the suggestions you provided, but there seems to be some issue with byte swapping

yes, even I faced this

then the cause is wrong splitting of sensel data, there are just 2 causes at this

its a silly question, I know but most of the documentation I could find about endiness dealt with 16, 32 bit numbers

but first suggestion is to check the splitting

yes, I have turned my github repo private. May I have your github id so I may add you as a collab?

nevertheless, off to work, will be back on later

se6astian: I was doing the C++ challenge and I had some doubt when with endiness, I was hoping you could clarify that

morning Bertl

so you have a question to one of the qualification tasks?

let's hear then

the raw12 file is in little endian, so from what I know a 12 bit number like 0x123 will be 0x2301 in little endian, but since we are storing two sensels in 3 bytes I figured we must be ignoring the leading zeros

so is 0x123 stored as 0x231 in the file?

Bertl: is that correct?

there are no 'leading' zeros in the data stored in the raw12

so the number 0x123 will be as 0x231 in the file?

it depends on the position in the datastream

if I read the entire data into a char array then the data is read as is

then the stream will consist of bytes

B0 and half of <B1> will make up the first 12bit value

that is what little endian means, yes

that depends on what you are doing with the data

do you want me to solve the challenge task for you?

it is rather simple to check whether you need to swap anything by separating out the channels

then look at the images and combine the channels to an RGB image

hi Bertl, i want to ask if i can do code challenges using verilog not vhdl? as i am interested in Pixel Re-mapper task.

hey BAndiT1983, se6astian

guess you guys are away right now. will come back later

I found T1121 to be interesting, so was learning shell scripting when I got time

sorry about to leave the axiom office now

sure

https://www.apertus.org/GSoC-2019-Mentor-Contact-List

could you explain what is meant by "add the data as VCD file?" for me !

"(16:38:19) rohan_: guess you guys are away right now. will come back later" < Even if this seems to be the case, still post your questions here as nothing is lost and they'll be answered when someone comes back online.

aSobhy: 100ps ?

VCD = value change dump https://en.wikipedia.org/wiki/Value_change_dump

yeah its a simulation !

aSobhy: there is no point in running functional simulation in the GHz range, it will always be just fine

- synthesis and routing

- simulate it on model-sim

post-implementation (timing) is what you want

hi guys, is there anyone around knowing which 1/2.3 and smaller sensors are FLOSS-friendly?

we are quite happy with the ones we investigated (see AXIOM Beta page)

we also know that OnSemi is quite FLOSS friendly

ah, good to know

i would not call them FLOSS friendly by any means

we went for the ar0330 and it is _possible_ to interface them with the existing documentation

dcz: if you could provide a link, that would be great

many sensors have a high moq, so it is difficult to design anything with them

I'm looking for it rn

interesting in what sense?

mostly bigger sensor size or more resolution then the ar0330

add http:// in the front

do you know how much it is and how easy it is to source?

no, definitely not

what's your address?

did you find any other interesting sensors?

that's a bit of a problem, because I'm not entirely sure what to look for

I'm trying to get something good for https://puri.sm/products/librem-5/

so you are probably looking for a rather high res cmos?

the big worry of mine is writing the software if the sensor doesn't come with a DSP\

you're not interfacing with the sensor using Linux, are you?

dcz: well... we are interfacing it with an fpga

then the raw image data is in ram which is shared between arm and fpga but the main idea is to explicitly not move all the data through the arm core

does it have a mipi interface

yes

hm...

anyways, battery performance will probaby get really poor

hm... what do you have in mind there?

assuming it's not moving the data that's the most expensive, then encoding a downsampled stream is going to be less intensive

not of the camera performance

but did you try simple encoding tests with for example ffmpeg?

hm k

but idk anything about the project in depth so my comments are probably not really helpful

we're running out of space and costs

but we have had performance troubles with just coppying the raw data to the ethernet with the arm

saurabh_raj, yes

there are fewer new sensors to my finding

I have turned my code back to public, maybe you could have a look, n let me know there is more to be fixed

skipping the video topic, what do you think is worth to look for in sensors suitable for stills, outside of price?

saurabh_raj, should i reply by mail, so it's only visible to you and some mentors?

definitely raw access to the data

that's going to be easy

do they usually come with lens assemblies?

no

that was one of the questions i wanted to ask :) what kind of optics do you plan?

I don't actually have the faintest clue

BandiT1983: not an issue,

i would be interested in fast optics, as this directly infulences image quality drastically

yup

I'm afraid that would be difficult to design

how do you measure how fast the optics are? bigger f/ numbers? more transparent materials?

smaller f numbers mean geometrically brighter lenses

right, the scale is reversed

so f1.0 is about double the light of f1.3

t stops measures the amount of light, that makes it actually through the lens

(in inverse)

yeah, but coincidentally ;)

also, the f number impacts the depth of field (together with the sensor size)

so if you have a really small sensor (and therefore a short focal length for a "normal" zoom) and a not-too-bright lens, you can probably get away without autofocus at all

f/3.0, 1/3.2, 2m+?

that's software though

focal length is normally given in mm and should be in the range of ~5mm for your usecase (i guess)

http://www.dofmaster.com/dofjs.html

seems so

yes, that's going to be something smaller and I think with a DSP

ah ok

don't think so

I'm sure we have a couple of candidates

you might also want to buy a ready made optical assembly for the back camera

I think that's what's going to happen actually

danieel: can you post a link to SMIF?

SMIA ! mostly nokia phones used that, even up to the 41MP sensor in the lumia

thanks

the other way is just snoop the i2c of a existing camera

there are a couple of Samsung drivers in the Linux kernel thanks to Android

is phase-shift autofocus worth looking for?

and other usage of the sensors, the kernel part could be just a glue code

those shall have full source in kernel

it can have the magic missing, especially on android

dcz: looking at the imx219 camera driver that is used in RPi, it seems a complete driver (but the camera is rather basic perf, no wonders)

dcz: some of the linux drivers are rather bad

they only send a bunch of undocumented sequences to the sensor

well we looked specifically for the ar0330 ones

but if you have ndaed datasheets and register reference that is not much fun

(as in wired; the ar0330 has some magic init sequences and some binary patches that you are supposed to upload to the sensor via i2c depending on the hardware revision)

that's not encouraging

I heard @whitequark has a library for that kind of stuff that might be worth looking at ;)

do you have any datasheet for that?

dcz: http://www.gizbeat.com/wp-content/uploads/AR1335-D-PDF-framos.pdf this one has atleast a datasheet

OnSemi does seem open

you can find some datasheets on some websites

but atlesat for the register reference of this sensor there might be some possibilities

idk, I think se6astian is the person to ask there

https://wiki.apertus.org/index.php/Image_Sensor_Table

Bertl: which connectors does the beta use for connection between the sensor board and mainboard / interface board?

Hirose FX10 series IIRC

more pins between sensor board and interface board

Hello , Can i submit draft proposal for review on GSoC website, so i can change it accordingly or is it too early

No i am here for a long time

OK. Best thing to do would be to put the text into a Google Doc and I'll take a look at it.

okay, i will do it

also BAndiT1983 , Can u review the challenge code once agian, i tried to change AVI generation , hope it is okay now

hi Dev_, where is your code residing?

ok, will check when i have some time

thanks

address*

Found it?

How should i share it u, PM??

se6astian: thanks

to Team emial address ?, okay

Hi ,in the https://github.com/apertus-open-source-cinema/axiom-beta-firmware/blob/master/software/scripts/set_wb.py how are we determining the value of color_matrix?

Y_G: yes

it could be done empirically

good night

Hello, Bertl !!

Thanks se6astian

So we are focusing on semi automatic white balancing here ,where we can set various illuminations like sunny ,fluorescent and get the respective white balanced/color corrected image ,right?

we can re-use the same fixed-size image (basic graphical cordinates and scales) and re-write it over and over everytime we need new graph using SetImagePixels() .

for this we can use Wand API, which is prefered for drawing graphical primitives.

please let me know if there are some changes needed . I will be reading logs ;) .

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