#apertus IRC Channel Logs

back now ...

it could be interpreted by people not familiar with indiegogo/crowdsourcing that they're obligated to buy the camera when they choose one of the €350 perks

I was a bit unsure about that when I first read it

and there have been comments from at least one other person (on magiclantern forum) who also seems to have misunderstood

'payment due' in english usually means some form of a bill

maybe the number of people actually confused is very low, idk

but it isnt actually when it ships, is it?

intracube: the problem is, the perks become immutable once they are chosen by at least one contributor

comradekingu: not really, it still doesn't make clear that:

but I agree, wording could have been better ... then on the other hand, we are not native English speakers, and we try to ask native speakers for help on every occasion, it probably just wasn't available when it was written

but as Bertl says, it can't be changed now

+d

Bertl: yup, no worries :)

figure of speech :P

btw, can anybody help me understand what the idea of 'will sandenbergh' is? (comments section)

I do understand the feature list and vote on part

hmm ...

which might be the basis for anther funding campaign

we already do regular feature voting (although it was a little slow/lame the last month)

right now I guess the focus for you is on the fundamental 'must have' features

yes, I'm constantly urging folks to put features they would like to see at some point on the wiki

I'm surprised there are three sensor options already

and we do not want to finish the Beta in five years :)

yes, precisely

what's the problem with that?

I mean, what is the problem with simply adding one?

no problem

except finding a supplier and getting a filter that closely matches the sensor resolution

and have it physically fit in front of the sensor

Bertl: agreed

it is a quite expensive part nevertheless, so nothing everybody will want to have integrated or so

and we won't be able to test a variety of solutions, otherwise we would spend all our money on OLPF samples :)

maybe you could get free samples if OLPF manufacturers think they'll get more business?

something like the Beta can only scale if we actually manage to leverage to potential of the community here

have everybody interested acquire a specific solution (coordinated efford) and test it with the devised test system for classification

yep

sweet dreams!

90k \o/

morning se6astian!

still working away on the AXIOM Beta? :)

great!

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

Heya Bertl!

off for now ... bbl

almost, yes :)

5 days and nearly 8k€

:)

I would donate, but still confused

maybe we can clear that up once for all

oookay?

I'm reading now the axiom beta

anyway, if you have questions regarding the AXIOM Beta, your potential contribution or anything related to the development, please do not hesitate to ask

I will be probably Mega supporter in a few days :P

are there engineering grade cmosis CMV12000 available? Thinking of a long therm "burn in" test as soon as possible. So we have maybe 6 month test when the first batch is ready. We could put the latest image on the webpage...

comradekingu: german, right. magic lantern dev.

I think i tried that on my dads point and shoot canon

g3gg0: btw, has there been some support from Canon for ML?

not even a word about us

the chips share the same architecture

at least the DIGIC-chip internals are related

http://magiclantern.wikia.com/wiki/Register_Map

ADCs are external and we are not sure how to program them. we learnt a lot from experiments too, but a datasheet would help quite a lot more :)

Bertl: its complicated for a company like canon to give a statement about us... they dont control us. so they cannot be sure that we will not act against their plans

see the 1DC 1DX discussion..

so how is ML designed ATM, can you give me a quick intro?

Q_: http://magiclantern.wikia.com/wiki/ADTG one of them in a 600D

Oh.

Bertl: stock canon firmware has a flag in its configuration area that tells the bootloader to check the memory card for a binary AUTOEXEC.BIN that is being loaded to a predefined memory address and executed

Bertl: now using a forged firmware update file we rewrite this flag. (firmware update does not contain a new firmware, but just custom flashloaders that write the flag)

then we have two different methods of hooking into boot: a) copying firmware parts into RAM, patching it, hooking a task and continue execution

flashloaders are signed

okay :)

so basically you get control early and patch up the existing firmware with custom parts, yes?

yeah

normal magic lantern only needs startup patches

sometimes we have to patch RAM pointers, but normally no firmware content has to be modified

the ARM's CP15 interface allows arbitrary access to cache content

understood, and development tools for the cpus (what do they use) are available somewhere?

is there a danger/potential for canon to lock you out?

they never did :)

you mean A20 ?

Afaik, canon known about it and doesn't care.

they dont love us, they dont hate us

we dont cause trouble, people like ML, so they stay nice and dont actively suppress ML

Bertl: who knows what happens in the future, yes

reworking all the sensor IO driver stuff will give us perhaps 100 fps without trouble.

toooo much reinitializations. every frame is set up like a single frame shot. all settings written again etc..

comradekingu: not sure... dont think they ever noticed our work in their money pockets

the coolness factor of a custom firmware hack like ML is absolutely high. but this affects maybe 2% of their customers

this is the paradox though, i value magic lantern, but i dont want to support canon

Bertl: both. its our hobby. if we find things out that look cool, we try to make it a feature that users can use. if users ask for addition or changes, we try to add them.

e.g. raw video. people then asked for raw(-data based) histograms and alex added it

4k raw video

and that works at a reasonable frame rate (i.e. 30 FPS or so)?

24fps or a multiple

in ML?

yeah

a good CF in canon cameras can get up to 110 MB/s

we have a few hundred megabytes of RAM, so we can buffer a few seconds

some 1080 resolution recording as RAW without limit. so you can record minutes

others do only 5 seconds recordings with max resolution

depending on your storyline this is possible.

Or did they fix that?

4k = 4096x3072 or something like that, at probably 10-12bit, yes?

eos has 14 bit

so roughly 16 gigabit per frame uncompressed raw

no, sorry, too fast

MiB

lets calc with 20, yeah

*roughly

you would need a few CFs in parallel

okay, so I presume you do some lossless compression on the fly?

we *can not* do 4k video yet

ah, okay, that explains a lot :)

because I think the memory doesn't help you either

for obvious reasons, as you can hardly get a second stored in a few hundred megabytes

we did some calculations for the beta, which has 1 Gigabyte of memory

and you can get away with 128M for linux, if you need to

which leaves almost 900Mbyte for buffering

but you could add four CF slots :)

just one hypothetical example:

93k \o/

four CF-slots with 100MiB/s write rate each, hardware controlled I/O from FPGA domain. this should get you 24fps with 4k stable.

barely if at all

4k,12bit,24FPS (to take your example) gives 452MByte/s

But I think non-volative RAM (the feroelectric stuff. Samsung is going to be producing - among others) will displace flash in a year or so, so plan for it?

432

but still doable with nowadays CF

I had the 30FPS in mind

but that's actually 540MByte/s

it allows you span the data across several media if needed

e.g. on 5D3 we can use CF plus SD for writing video frames

some post processing tools already know this format and the workflow is evolving

Bertl: blocks can appear in any order, in any file and PP tools have to index them.

Bertl: i had in mind that files are split at 2G / 4G, a "recording" can contain of multiple files and due to buffering the data from camera isnt necesarily in sequence

easy to add if required

we have a lot of metadata blocks that are predefined

so if you experiment with raw video, contact me, we can get a minimal solution very quickly

feed the raw stream into a shift register, 16 bit wide and store the 16 bit words as LE

http://magiclantern.fm/forum/index.php?topic=11787.msg128638#msg128638

so the first line will be RGRGRGRG and the second GBGBGBGB

e.g. swapping columns or rows

ah i see

so you would need a field that denotes the orientation of the sensor data

sure, thats easily implemented

because they send data as 0, 128, 256 ... 1, 129, 257, ...

that's columns

kind of, there are up to 64 LVDS channels on existing Cmosis sensors

the best then would be providing a pixelmap block

on the first page you see an example file layout

contains basic infos about its content

and later you will see video / audio blocks and info blocks in any order

there are only two blocks - beside video data of course - that are important. file header "MLVI" and raw data description "RAWI"

see tab "Structures"

then four bytes block size

okay, I guess it would be great to get a checksum field into most of those structures

yeah

the actual checksumming doesn't need to be fixed in advance, but a 64bit field would help a lot

we can easily do some crc polynomial on the data, which then can be combined if needed

end is fine as well I guess

or a field in MLVI

the CRCI block

it would allow to use case specific checksums, from simple to cryptographically strong

well, if authenticity is important, it might make sense

who asked if we could do encryption, so that not even he would be able to decrypt it, only the new company "at home" (mostly for war zones)

when they want to prevent tampering or similar

wait

RSA

so you can point him to us ;)

yeah, those are special cases after all, but it is a fair reason

okay, I have to take a nap, I'm rather exhausted ... thanks a lot for the very informative and pleasant chat, I'll be back later tonight ...

cya

That's damn brilliant to have a reporter sign their footage and encrypt it.

I think we can make it!

are we austrians or austricans?!

cans!

Wow! Great last two days!

re

http://crowdlogs.com/project/axiom-beta-the-first-open-digital-cinema-camera

wow!

where crowdlogs is of course not realtime, and we seem to have entered the second hot phase, probably thanks to your (ML) excellent PR

and when this was placed on the ML frontpage

wow, we broke a lot of their stats

night night!

sleep well!

impressive support through our userbase

yes, so I wouldn't worry about your Beta cameras :)

have you decided (as developers) what configurations you'd prefer?

where a1ex is looking into getting the most out of the sensor, i tend to look into raw recording

(which may later be exchanged against final versions)

so a preference for the CMV12k probably on MicroZed as backend then, yes?

yes, but not at full performance without FPGA support (which isn't there yet)

ah okay

10G would kind of work for low quality :)

dont have an atomos

if you're tricky, you can use both SD cards, internal and one on the external SD interface at up to 50M/s

that would be a solution, yeah

so i am curious how both worlds could be combined for reasonable performance

ok BTT. sensor type: i have no clue which one to start with. i guess the one which promises best quality

the KAC looks quite interesting on paper but we haven't had a chance to test it yet

anyway, nothing which needs to be decided tonight, but maybe have a chat about that in the next week or so

you make four dark shots and he tells you how good the sensor is :)

he had the KAC on his chart, but I do not remember if it was just projected from the datasheet or the result of actual testing

in any case, danieel should be able to take some samples with the KAC, as he has built a camera with that sensor, IIRC

a1ex has itnernet trouble atm

should be fixed during next week

So I wanted to look at the raw samples from the CMV12000 that someone posted on the ML forum and points to footage.apertus.org, but that requires a login?

or is it a general link to 'footage', because in this case, we should wait for se6astian

I wanted to look at those dark frames.

Bertl: Congrats.

really?

Bertl: Really

Bertl: Every single campaign successful or failure, that hits a threshold of contributions looks precisely like that curvr

Bertl: Looking for another sample

More or less.

There was an exhaustive discussion when Ubuntu edge happened

I think there is an odd correlation between entry curve and exit too

Q_: files will show up here: http://vserver.13thfloor.at/Stuff/AXIOM/ALPHA/ML/

it will take some time till all files are there