#apertus IRC Channel Logs

Hi Bertl.

Next step, I will make some codes .A test platform needed for software debug.Have you completed SD cards PCB? Can this PCB be used of ZEDboard? I only have a ZEDboard.

there is socket for CMV12000 on FMC card. What is the part number?

regarding the bayonet, we got the last from a nikon F65 body, IIRC

http://vserver.13thfloor.at/Stuff/AXIOM/ALPHA/cmv12k-adapter-v1.1.bom

I am back from amsterdam!

workshop?

"Meet us in Amsterdam at IBC 2014" <- this one

no workshop

btw, do you mind taking a few test images with the axiom alpha, so I can check the SNR curves and the dynamic range?

so, just point the camera at some light bulb, make sure it's really out of focus and also make sure it has something really dark in the frame

I'd like such an image pair at each gain setting (1x, 2x, 3x, 4x), and also with HDR mode enabled if possible (I guess you only have it calibrated for one gain)

thanks!

they are scene-dependent? I guess you can use the parameters you have used in your test video

right, so you have a lot of room for finding the optimal settings

exactly, and the "optimals" are scene dependent

right... the method is promising though, I expect this mode to go beyond 15 stops (might require an external timer, not sure yet)

the HDR will capture more highlights, but I don't expect it to change the shadow noise

how was IBC?

Hi dmj7261, great!

alexML_: I mean combining exposures of 1/FPS plus the lowest possible exposure time the sensor supports

se6astian: any cool details?

he said "sure, ah that prototype you are holding, thats the open source camera right?"

sounds like you generated a bit of buzz

morning folks!

hey Umibuta!

Hi umibuta ... errr... Seapig? :D

Hi Seku :) Yes seapig. Or Dolphin :)

with reuse of the sensor (the most expensive part)

i thought they were called Iruka

mh,, interesting. (my nick is short for sekuhara ... a word i guess you know) ... felt i needed to change it when a friend of mine who works at the embassy introduced me to the japanese embassador : "Konno kata ha ... sekuhara desu"

Hahahaha. Sexual harrassment ? You are not by the way in Japan, are you?

to be honest, we have no clue at the moment _how_ the upgrade path will look like

No worries, Bert1 :) My other question is should I include shipping cost of 3Euro to 350Euro?

Yes. Regardless of Gamma. Beta as it is is already enticing. Even for a soccer dad like me....

I know I read somewhere that Beta is meant more for people in the industry. And you would benefit more from their feedback. But still I want to try my luck.

Magic Lantern raw video hack for canon DSLR

the main purpose of the Beta is for development

Yes. I that was what I read a while back before the funding was launched.

I have an appointment with my dentist now, so I'm off, but I'll be back in a few hours to answer questions ...

haha i just came back from my dentist

no

Bertl, se6astian|away and philippej are

we all are :)

:)

Sure. I might have missed some stuff.

hey se6astian tell us

hey aombk

about ibc

were people coming at your booth or i dont know what you had there to ask or were they ignoring you?

blogs/news sites were there?

we will post a news update about it all soon

alexML_: PLR knees to a logish curve is probably the only real way to make the latitude useful.

all the gory details :)

alexML: mucho gracias :)

troy_s: I think it may be possible to minimize the artifacts in the third HDR mode, with heavy postprocessing

https://apertus.org/axiom_imagesensor - there are 3 modes described

okay, so PLR HDR is like taking 3 shots sequentially with different shutter speeds and superimposing them

if my reading of the spec is correct we can start exposing next frame as soon as 70microsec (FOT)

it will be like starting next frame immediately?

wouldn't it be like one continuous shot with extra DR?

but if one wants to economise he ca increase storage format bit depth by several bits and do this addition between frames before storing to the ssd?

sorry about typos I'm typing on a small phone..

1/4 of the desired shutter speed but 4x framerate

that is when you use same shutter speed for all 3 frames which you later combine into 1

anton__: how did you get 2.5 stops? log2(sqrt(6)) = 1.3 stops of DR

or if you shoot in triplets of frames say 1/100, 1/1000, 1/10000 and store all that info on ssd then software on a pc can combine each triplet into 1 frame similar to in-sensor hdr but the software has much more info to play with to remove artifacts

in post, you only need a LUT to get linear data

but if you have the processing power to do the merging in camera, that would be cool

in highlights yes, but in noise, no

if you average 4 images, the stdev of (a+b+c+d)/4 will be 1/2 * stdev(a) - assuming all 4 frames have the same amount of noise

yeah, if you average 4 images taken at identical exposure, you will get 1 stop less noise

you still need to do the choice in the camera

quick question: is "look-around" the official name of the feature that would suggest to you to show you an image area around the the area thats actually being recorded?

Alex: suppose you had a bright lamp fly very fast accross the scene. If it didn't clip any pixels in 1/100 take you don't need any artifact removal at all; if it clipped a pathe across the 1/100 frame you know that motion blur needs to be applied within that area - but nowhere else! don't you want this info for post processing?

se6astian: I think it's "action-safe area" - http://en.wikipedia.org/wiki/Safe_area_%28television%29

anton__: I'm used with memcpy not being fast enough for realtime 1080p at 24fps (the 5D3 CPU)

alexML_: the "action-safe area" is actually part of the image while the look-around is only visible to the DOP by not being recorded

alexML_: also re averaging 6 frames - it may be that noise is related to exposure time, so if you shoot 6 frames instead of 1 at a 6 times shorter shutter speed and then add the values together - add not average - you may get more noise compared to one long take - or maybe not, maybe not a little more... I believe only prwctice will tell. so perhaps just perhaps this can be a viable way to get more dr in highlights with 0 artifacts

*but

good evening everyone :) welcome back se6astian

compared to the long exposure, your noise would be log2(sqrt(N)) EV higher, assuming noise does not depend on exposure time; for N=6, that means 1.3 EV

Hi, derWalter thanks, I have seen in the backlog that you have been quite busy while I was gone :)

especially as i wanted to get close in touch with you for month, but never have had the time for it!!!

and as i am reading the actuall irc log from today: regarding different shield types of the beta and the gamma: would it be possible to create an adapter shield so one can use the gamma modules on the beta?

alexML_: got it; well I hope you also find the idea with hdr motion artifacts removal mask interesting :)

back now ...

wb bertl

that would in theory be possible, but it wouldn't really make much sense

I presume that an adapter shield would cost twice as much as the same 'module' for the Gamma

and you would have all the problems of reduced throughput, interference, etc from multiple connectors and the solution would use up twice as much space, have twice the weight, etc

well, i was thinking of ONE adapter for the beta, to attach only gamma modules behind it. my problem of understanding is the lack of information about the gamma. i would eager to read the 100+ pages you sent in to the eu (for more reasons)

yes

i would love to contribute to the gamma

i dont know the reason why the beta has its own modules

i just dont see the reason for different modules in the end

the Gamma modules will be much more advanced than the Beta Shields

they will most likely incorporate a bus system to communicate between modules and between sensor/FPGA and module

they will also feature some kind of power rail system to supply all modules with power

no, the Beta Shields are only one stack depth and basically a direct connection between FPGA and Shield

aaaaaaaha...

haven't you seen the illustrations for the Beta shields?

yes i did

here the HDMI shield

but all i could read from the graphic were the screws, the connectors and a strange shape where i didnt know how it was related to the design of the beta :D

now i realized it

well, and it didn't come to mind to simply ask? :)

you plug them directely in, or open the case, connect it to the board and close the case again? :D

not like an actual add IN

you plug them into the front of the Beta, basically right beside the lens mount (in the current design)

wow, nice.

yes yes, well, well, well... mhhh

there is a hand full of things missing for me, i was thinking about, battery, sound, internal storage and controlls/viewfinder

and i thought of modules to add those to the beta, or even a case where you plug in the beta

so you have a case, shaped like a normal camcorder/camera with all the controll knobs, xlr connectors and so forth and you just drop the beta in the whole thing

what kind of connection are the addins using to communicate with the "main"board?

sry, the beta

on the right side, they provide 10 differential pairs to the FPGA each

aa, now i know what connectors you have been talking about some days ago

many things are possible :)

*off

not because of the technical challenge but mostly because of the space

aa, okay :)

yeah, of course, if you add a PCB which is as large as the Beta itself and connect it to the Shield connectors, it becomes much more realistic

what was the reason for using metal for the case and making it so small? (i just realized HOW small the case is, when i started to calculate and skeetch around the battery case!)

we had a shoebox sized Alpha, so this time it should be rather small and elegant

reducing the shutter speed only affects the classical thermal noise

nick Kevlar

thankks

aa, cause it looks like a bare metal case :D well, the shoebox size is not what i expected either :D but like this its just tiny :D how much cost would it add to put a small dispay with some knobs on the back?

yes, it is something which was already requested several times, so we will find a solution for that

yeah, that looks like the perfect design for this purpose *G*

but I think we should go with a more traditional interface like this one: http://www.bbcamerica.com/anglophenia/files/2012/03/tardis1.jpg

It is definitely more eye catching.

Why did my chat say you failed to identify in time for the nickname Kevlar? Does that mean someone else has the username?

i like the selfmade lightworks controllers a lot :D http://smg.photobucket.com/user/Greg_E/media/pics/controller.jpg.html

haha

Everywhere I go

i had to think about this instantly :D

gooooooooosh... i cant post on nikonhacker, as it says my ip is blacklisted at spamhaus.org, cause a pc on my ip seems to be infected with the conficker worm...

i hate windows :(

seems like i ve to restart to install kasperky :)

evenings

btw, i noticed the beta display discussion tonight. how about going WiFi to a tablet or phone for starters?

wifi with the live image?

just an idea :)

I dont think we want to get into H264 encoding or anything of that flvour

ah true. havent checked if the arm chips got anz h.264 easing implementation

but i guess most will go Atomos Samurai or Odyssey 7Q for recording at first, so monitoring might be a non-issue

with a plastic body?

those are probably the old ones... in metal bodies... heavy

i guess

good night!

thx :)

depends on the plastic i guess :D

oddly enough, long exposures seem to have a little less noise (but the difference is minimal)

alexML_: I take it the ML means from Magic Lantern? Can you code?

se6astian|away: Agree with you on h264. It is strictly an editorial / offline format. The only codecs on the map should likely be the ones with potential viability as an (albeit crappy) intermediate (aka DNxHD and ProRes) and those likely come with licensing issues. Little egregious with current trajectory really.

(Although bringing FFv1 or HuffYUV (with raw Bayer) to the masses is an interesting thought project.)

My idiocy knows no bounds.

I'm counting the electrons

(just in case .. hope im not being obnoxious here ... alex pioneered DualIso RAW on ML, and that includes the necessary postprocessing implementation) i follow all of his posts on ML.

Very cool.

i was always dreaming of a sensor which is capable of counting photons and their wavelengh

alternating ISO lines in same exposure,. but alex can explain that better :)

something like the foveon sensor is doing

nope.

http://www.magiclantern.fm/forum/?topic=7139.0

well, if there would be such a thing, it would be something like an ultra sensitive ccd sensor, but u would get out a stream of data when what kind of photon arrived and you could render your desired fps in post if you want :)

Very cool.

alexML_: What is the TL;DR on the merge technique (looking at your PDF)

match the exposures, then interpolate in highlights and shadows; in midtones you have full resolution

alexML_: Yes.

That was my worry...

I get a linear full-res DNG, that looks just like a normal image ISO 100, but you can push the shadows to crazy levels

alexML_: Betting I can improve your interp with a cubic b prefilter instead of average (lerp)

you are welcome to give it a try

currently I'm using AMaZE for interpolation

Hrm.

1) VNG can result in non-color values, or values that are at irregular positions given the gamut of a profiled sensor. There is a way to leverage the result against the gamut and always assure the value is "in gamut" (aka more likely closer to the value at a given position)

2) The sampling I have seen has yet to use cubic b prefilter, which is an anomaly in the scaling world currently, despite being of stunning quality.

alexML_: Very cool stuff Alex.

yeah, the DNG spec can use a LUT for this (although the values are limited to 16 bits)

alexML_: To get from the merged work to a linear representation

from dual iso? the merged work is already linear

So are the samples in the PDF just random TRCs for demo?

meanwhile I've got a ton of test pictures with tricky cases, so the algorithm got a bit more complex

alexML_: Github?

this is the dual iso postprocessing tool

no, just pixel peeping

(from a 2-image bracket you can create a fake dual iso image)

alexML_: I was going to suggest a synthetic image of worst case

(Probably 75% red / blue stripes etc. Similar to chroma sampling patterns.)

?

Frequency domain fitting I suppose.

bedtime for me, work awaits ...

least mean squares... nope, can you point me to a link showing how to use it for linear regression?

alexML_: I think Argyll has an implementation for their fitting of color cards.

in post processing

at least I (0 experience in the area) can not think of how I would implement HDR artifact removal w/o these masks

back now ...

But alexML_ has the best direction

if artifact-free seamless HDR is viable then Axiom Beta looks very promising; perw/o seamless HDR - it looks like CMV12000 doesn't provide that many stops...

with seamless HDR one could even say Axiom may have an edge

alexML_: the white out of focus wall is to prevent wrong accounting caused by clipping and/or offsets

i.e. if you have areas which are clipped (too high, too low) then the difference will not show the noise you're looking for

anton__: Solving the temporal issue with HDR adjacent frame / different temporal sensel values is not an easy problem, and will almost always have some case where the approximations fail.

and the second question is for anton__: how should a snap after the HDR snap help with identifying critical parts?

Bertl: It covers most of the stuff you would likely find interesting.

Hot sun on leaves and they _all_ have shifted.

To me, it is probably the most worthless of attemptd

IIRC we could place two knee points right?

(And did Cmosis get back to you regarding the need to toss an entire bit and a half of data?

Bertl: Oh I didn't realize. Harder to profile with knees. Need to linearize diest

we haven't had much time to poke cmosis, but we will start poking again soon :)

Data even.

yeah, got it :)