#apertus IRC Channel Logs

wescotte: Per well would be temporal nasty.

comradekingu: Global voltage. Not sure how it would work otherwise.

but as the shutter is a global shutter and a pipelined architecture, the voltages resulting from the exposure are 'copied' into a sample and hold buffer to make room for the next exposure

and just bite the bullet on increased noise

comradekingu: what does that mean?

hi troy_s

A regular webcam has that effect, albeit not very controllable. At least thats what im thinking. And also i think you could do something more than adjust it on a per-scene basis

comradekingu: page 33-34

This can be done 2 times within one exposure time to achieve a maximum of 3 exposure slopes.

well maybe not :)

so each frame of video is split into 2 separate exposures. at the end of each exposure, individual wells that have a voltage above a certain level are partially discharged

(AIUI)

intracube: what benefit do you get by not discharging all the wells?

it sounds like the result of that might be less noise/banding in the dark parts of the image

troy_s: Bertl: is that the gist of it?

intracube: my reading of the PDF was - if well number x,y reaches VTFL3 earlier than a predefined moment in time then it does not accumulate any extra charge until that moment arrives.

wescotte: and the eye is more sensitive in darker areas of the image (and more likely to see banding there) if the image is stored as true linear values -if- the bitdepth isn't high enough

If we have a sensor that is 1:1 linear (one photon, one recording of integer)

We may only store a maximum of say, 100 levels.

which is why most 8bit image/video formats jpeg, bmp, tga, tiff etc store values already gamma encoded (not true linear)

(2:1)

If we put another knee in at 75

Bertl: yup, just saying what might happen if you mixed linear values + 8bit :)

But our granularity (the steps) are fewer... So in highlights, we have less gradation

what if you had a ram-drive accessory?

We can't see into that ratio anyways.

(Our eyes are terribly insensitive at the higher iris stop range)

So if we nail exposure, the reduced steps in the high end are not noticeable.

So is it only a benefit to set exposure at record time due to storage bottleneck for doing raw?

troy_s: okay, I understand now.

comradekingu: The sensors are excitable surfaces, and they are display referred.

what is a well?

Micro site

A sensel

a dip in the histogram?

No... Per sensel / photo site on a sensor; it is akin to a light well.

so range

On the video it looked to me like someone was doing ISO adjustments over two sets of variances in light input

Primer on Film and Digital Capture by Rob Hummel …: http://youtu.be/98FZ8C6HneE

The well analogy works well to describe a few other phenomena

comradekingu: a well is a single pixel I think

wescotte: Phew!

troy_s: thanks for sticking with me :)

wescotte: Happy you found some sense in my dribbling.

I guess that is more to fight aliasing issues

And oscillating sensors have a long lineage... See film.

God no.

also mounting the screws horizontally is

intracube: No.

intracube: It is all part of the same exposure frame.

and that would also be a minimal warranty against impact damage

comradekingu: do you have experience in camera design or other mechanical appliances?

Bertl: how so, i visualise things in my head

I do make stuff, all things audio, some electronics and i try to help out where i can.

But someone above already gave away that aligning the lensmount with the sensor was the hard bit, or the critical part anyhow

troy_s: so Red's HDRx mode =/= PLR technique? and might have issues in the highlights

like if a spark from a grinding wheel clipped the main exposure but didn't on the second highlights exposure

From my understanding you get uneven chromatic abbreviation if there is a mismatch

i think abbveration in plane, and different focal length if tilted

intracube: Hence temporal nightmare

Process

(Log is most logical.)

intracube: So a highlight in HDRX is what you would use from the stopped down (under gained) frame, which means it lives in an alternate time slice

intracube: Hence temporal effects; it isn't part of the same frame, subject to nasty other things such as blur, position, etc.

troy_s: they've made a feature out of it: http://www.red.com/learn/red-101/hdrx-high-dynamic-range-video (controlling motion blur section)

LOL

Ugh

comradekingu: Who? Bertl and se6astian|away ? Yep.

but the lens and sensor with respect to the body

If the lens-mount and body are mechanically fixed together along with the PCB also being screwed to the body its easy to get warping from either not good enough clearances or screwing it together wrong

Beefing up the PCB and letting that carry out to the body via the screws and having some clearance/padding between lensmount and body is the silly-simple solution

Think of it like what people do when they gestalt an elephant

and we appreciate the input

that's what is currently planned

well why not say that instead of making me suggest silly things ;)

Ok, im onboard with that, just the fan that irks me now

Are there ventilation holes to push air out?

the fan is an exhaust

there are holes at the front and bottom which will allow fresh air to enter the body

if the holes dont have dust-covers that seems like a disaster if it draws dust into the same compartment the sensor is in

we had the same setup in the Alpha prototype, and no dust on the sensor, except when you opened up the lens system/mount

troy_s: have any more videos like that "Primer on Film and Digital Capture" you just linked?

(given that the copyrightholder allows it)

taught me what FPN actually is. I thought it was a result of something else

wescotte: which bit of the video covered FPN?

FPN btw, consists of two components, DSNU and PRNU

DSNU is the offset, PRNU is the gain related pattern

which for some reason happens mostly across line/column boundaries

wescotte: Someone should add it to the wiki.

wescotte: It segues well into highlight recovery... Understanding that a well might be full, the typical vendors discard the entire pixel result.

If you do not, you end up with a filled sensel well in a sky region for example, and then semi-full R and B. R + B will end up pink.

You can however, do highlight recovery that will “see� those sensels. The color will be pink or cyan or yellow, but still data.

(For those interested, I believe Mr. Savedes has serious highlight issues on Zodiac and they had to hire a firm to perform highlight recovery on a good deal of footage.)

s/Savedes/Savides/

do you have an overview of those algorithms somewhere?

Or was that still Red One time?

Bertl: Jerk.

Bertl: Dcraw explains them rather well in the docs. Code available.

Bertl: It throws out the citing pixel

wescotte: No what happens on branches is darkness against pinned regions.

Wow, forgot Red One came out in like 2007

wescotte: The green sensel fills and voltage spills to adjacent sensels...

Red and blue on near black looks magenta.

Are there any methods to detect overflow or prevent it?

there's another theory here: http://en.wikipedia.org/wiki/Purple_fringing second and third para

The most common spill colors magenta and cyan

Almost always spill.

Easy to diagnose.

What if you use really high frame rates and then sum them together to make your final exposure?

wescotte: Temporal nightmare.

Hence the rule (as Savides likely discovered, always underexpose)

Digital. Film was opposite.

Digital is nightmare on high end.

troy_s: the only time I've seen purple fringing on my DSLRs to any noticable degree is with a zoom lens with a wide range

intracube: Try the tree test.

intracube: I can assure you L series glass in the sweet spot will do it. As will master primes.

intracube: Just look at the color; if it appears subtractive (CMY) it is either sensel well spill or microlenses.

Slightly offset.

You can probably speculate how difficult it is to blend different temporal images. If not, try your hand at an HDR bracket in Luminance with your DSLR. With a tree with leaves for example.

Try it. ;)

Try a tree with leaves.

but afaik I've seen hardly any fringing on either a 24mm prime or a sigma 10-20

intracube: Let your sky blow out against a tree with no fill... Thin branches are best because it introduces aliasing.

intracube: If you understand the sensor layout as you do, you can easily find ways to trigger it.

troy_s: you talking for video or just stills?

wescotte: Same issues.

If you have a series of images and the exposure time doesn't sum up greater than 1 frame length then it should just appear as natural motion blur should it not?

You have to remember that the blend is not unifo

troy_s: also ENG zoom lenses have a HUGE range it seems

isn't it taking non uniform length exposures and combining them?

say field 1 at 1/48th a second and field 2 at 1/12th a second to capture highlights?

Bingo

.and try the HDR with Luminance.

Etc.

But if you took 5 fields at 1/240 of a second and them just summed them. it would be the same as taking 1/48th exposure except if the highlights didn't clip in the underexposed ones now you can recover them during the sum

PLS is ultimately the highest quality approach

PLS is lossy :)

just because my eyes can't see it doesn't mean I like it :)

Well the Alexa and Sony both do it with SLog and AlexaLog

;)

Because the Alexa is a solid camera. As is the F65.

Log footage is a typical entry point for display referred grading

PLS sounds like you could get artifacts if you push too hard with your contrast

It's interesting stuff but I almost never do it

Or on a heavy secondary on a sky etc.

Just enough for a decent roll off.

Been around film for a while.

Depends. If it is one of my projects, everything.

pretty much the same here. Done pretty much everything except write/direct.

Which I guess I have done too but it was class projects I didn't care about :)

heh

Everything begins to make sense once down that path.

Learned even more dickering around Bertl - the raw sensor data was an education

I have yet to really touch anything raw. Shot a film on a Red One/Epic once but I didn't grade it so I never really touched the r3d files except to make proxies

otherwise it's been DV/HDV and AVCHD stuff

heh, what's your beef with Red? :)

Someone that has worshipped at the shrine of Jobs too long.

That is fundamentally problematic to me.

They encrypt the raw format.

And you must use their tool to decode it.

RedCine-X is a pretty decent tool though :)

However, I agree with you that openness is key. However, I wouldn't say any other companies are any better in that regard

(All the magic was / is JPEG2000 frames)

Anyways... My personal feeling is that the backroom politicking in the name of selling a camera toy has done nothing for independent filmmakers and artists.

Arri and Sony are both “better� actually

And Arri in that they provide very good specifications and transforms on their raw data etc.

I think Arri is in a completely different market than Red

(Tells you much about their desired audience)

dunno what ACES is

http://en.wikipedia.org/wiki/Academy_Color_Encoding_System

Odd... Just noticed they say 16 bit half floats for -65k to +65 k. Something wrong there. ;)

they like DEEP blacks :)

No such thing as black. ;)

That is actually for upper end latitude largely.

0.18

that kinda makes sense

So your upper bound high value in scene referred is about 1360.0

floats aren't uniformly distributed either I think

Float data is if scene linear.

that sorta thing

I dunno it's been a long time since I had to deal with them but I could have sworn I remember reading a few articles talking about that sorta thing

Sure Wikipedia would have more.

Maybe that's why they limit it to +- 65k cuz it looks like the distrubition takes off around there or maybe that's even the limit of a 16bit float

16 bit float seems odd to be referenced in integer.

What mantissa on 16 bit float...

5bit and 10bit

5.96 x 10^-8

Anyways... Should head off.

I'll be hanging around :)

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

i think the non-uniform distribution of values is not a problem- its sort of a lut/log type of mapping

but the sign is unnecessary :)

and there the sign makes sense

the variation in highlights is much less visible... well, unless you shoot snow or weddings :)

I'm trying to picture exactly how it is mapped to each increase in exponent is a stop.

The 1024 values per f-stop makes sense wit the 10 bit mantissa.

ehm, exponent

Yeah. It is easier that I thought. I forgot f-stop derives from linearizing the ratio of the diameter of the hole to the area.

I recall some old caeras had "T-Stops" to get rid of the 3.5, 4.8, and all that. Just 1, 2, 3 ...

So, do you know, is this camera going to produce scene-linear with some meta-data about the sensor's primary coords? Maybe many optons given open source.

good morning

good morning!

Sebastian: for the active Canon EF Mount you are thinking of the IMS from P+S technik? Their mechanical adapter costs 600EUR without electronics, so I think it's not this one...

but it's not an open standard ... :-((

we will create the mount ourself if we reach the stretch goal

for 2200€ :)

if you by two mounts you get one beta for free...

Has anybody thougt about the licencing for the lens mount? Is there a open standard mount?

if you find information please add it to the wiki

yes

maybe the axiom should upscale to 8k :P

axiom will upscale at 8k externally

8k resolution (upscaled externally)

upscale to any resolution you like - thats the ultimate freedom

Hi guys, could you pls remind me which tools you use for 3d modelling? I'd like to open the .idw and .ipt models on GitHub and possibly create a couple of rough sketches to see what you think of a couple of design ideas

.iam .ipr .ipt are Inventor files

yes inventor

perhaps it would be good to have conversions of the inventor files at the least se6astian

morning folks!

checking it out

maybe FreeCAD?

hello Bertl, i've got here from the magic lantern ML, i hope the indiegogo campaign will meet the target, apertus deserves it.

*makes

no idea, se6astian will know

hi

I'd say from the features planned so far it just needs to be bigger :)

intracube: I might imagine the sensor head/main unit not necessarily being bigger

a better comparison there would be the beta plus atomos recorder

even if it's actually several different modules (head end/recording + io/battery)

one advantage of a bigger basic camera is the option of using larger fans, more internal volume for passive heatsinks and addon modules later down the road

dmjnova: yeah, but cooling it might be difficult

you don't necessarily want to conduct the heat directly to the external case

no I mean there will be a "head module" inside the larger camera body

did this quick mockup: http://www.pasteall.org/pic/show.php?id=78031

larger alexa-style form with 4inch/10cm fan at the bottom pushing air up, helped by convection

That said having one is probably a good idea

and coupled with decent passive heatsinks like: http://www.pcper.com/images/reviews/286/4-Trio-of-Coolers.jpg

especially if we can still make the electronics air/watertight

I went to some lengths to make my new computer quiet. the cpu fan is oversized and conneced to a jumbo heatsink

for one of my purposes I'll just mount 6 large fans spinning really fast over it :P

?

cooling would be a side effect of flying

lol

intracube: if you actually do some cooling tests, we'll promote you to chief of cooling :)

while it was a joke, it has a serious core, we will need to identify folks who are interested in helping with various tasks, and I think those will be the ones getting early access to Beta hardware as well

troy_s: Unfortunately I did not save how I set the kneepoints for this particular shot

if you use a proper fan, you can just turn it up to 100% and it will blow out all the dust :)

compressed air can or bicycle pump do a good job

Graphene might have a use in this sort of application

i think its samsung that produces them

http://semiaccurate.com/2013/08/15/panasonic-make-a-10-micron-thick-carbon-phone-heatsink/

http://www.digikey.de/product-detail/de/T68-1A-120-90-0.04/1168-1753-ND/3460910

panasonic, sorry

has it? the logs should be instantaneous?

you usually do not need/want to cool the front of the sensor

So more like an insulator there

i think the best way is the try and error concept. get a package and test it out :-)

they do not produce any power, they consume power :)

most of it as with all electronics

the sensor can take up to 2W, the FPGA up to 20W IIRC

http://www.altera.com/devices/fpga/cyclone3/overview/power/cy3-power.html

assume the MicroZed for now, the Parallella gets quite hot

in the Beta, the FPGA has to move and process huge amounts of data, so there will be significantly more power consumption

Hello, how could I contact Bertl?

welcome gdims!

Im thinking the FPGA could possibly be mounted with the chip coupled to the back of the camera metal body

hi there

comradekingu: read up on the irc log, we had this discussion last night, even with some drawings

alpha/beta shots

g3gg0: good! comradekingu: np

you are doing all the hw development?

Instead facilitate mounting a 80x80mm fan by putting some screwholes in. That way its optional, and you dont have to carry negative fanspace around if you opt out

intracube: the ambient temperature inside a device is always higher than the outside, that holds true even in hotter countries. It would be less effective, but effective nontheless to couple to the case

ambient air temperature, yep. but the temperature of the case might be significantly hotter than the air temperature if it's in direct sunlight

still cooler than the chips though

You would be hard pressed getting a white metal body over 60 degrees

comradekingu: maybe not. example; some light coloured concrete in my garden showed as (IIRC) 40°c this summer

(with a laser temperature gun)

and I live in the UK. a black camera somewhere like the Mediterranean or Africa would likely get -really- hot

intracube: you wouldnt get the black model in africa. Meditarranian tops out at 50

intracube: yes, and additional cooling on the hot side

bertl: regarding the possible raw recording - if someone designed a storage module, with e.g. space for two CF cards, how much dedicated bandwidth from the zynq to this module will be possible?

comradekingu: so wouldn't any peltier cooling be impractical running of the camera battery?

g3gg0: on the Beta, you basically have 20 differential pairs (40 IOs) between the zynq and the high speed shield

the HDMI shield will be one of them, yes

comradekingu: another option might be to accept the external surface of the camera -will- get hot and instead work at insulating the internals from the case somehow

intracube: that sounds a lot like an oven

let's build an AXIOM furnace :)

inside of camera stays cool, but heats up the whole world :)

https://en.wikipedia.org/wiki/Thermal_conductivity for reading

(cine film cameras).

in the end he relied on his own (old) cine camera that was decades old

Fans arent really ideal for the application, since they produce noise.

a lot of equipment cold-bugs, where it doesnt work reliably if introduced to sub zero temperatures

If we follow the principle of "the bigger the lens, the more pro the person behind it" it makes sense to couple the rectangular sensor area to the circular lens mount

Also im thinking some knurling of the outer metal frame will double as increased surface area to dissapate heat

!

its not really a lot of heat that needs to move, and a waterblock coupled indirectly to the heat is less effective than getting it away with graphene

yeah

uh :)

as my main job is in the automotive domain, especially ECU development, i am used to stick to passive-cooling only :)

https://en.wikipedia.org/wiki/Computer_cooling#Liquid_submersion_cooling

Learn everything* you need to know about the AXIOM Beta in under 1 minute: http://www.youtube.com/watch?v=uyFRMgi6CUQ

comradekingu: yep, its still passive. but in automotive domain passive means not even a liquid is moving :)

but thats mainly due to the price

se6astian: isnt that video very bad for people with epilepsy

thanks for making my point regarding 'featured'

comradekingu: maybe, but I guess we can put a disclaimer somewhere in the description

I think more information about what a osi open source camera is, and the people behind it, would be a minute better spent in terms of explaining the axiom beta

https://apertus.org/opensource

im only adressing the video

i can help with a rewrite if you want?

feel free to work on that, but I think we should not change too much at once, so no guarantees that se6astian will adapt it right away

Bertl: it says open source doesnt tie in with production and sustainability, which is true, but free software certainly does.

And the ability to check and do something about it is the difference

I totally agree there

good night!

thanks

intracube: digital cameras bring a whole new set of problems compared to film. Rain bags overheat the things quickly etc.

Just huge vents

Not an easy set of design constraints. I would bet on liquid cooling to arrive very soon. Quiet pumps, and very effective.

They try

troy_s: I sees :)

Arguably Arri is the most experienced imager in the market currently, until Panavision gets back in (if)

was the last digital camera from Panavision the Genesis?

I think we need the Super35 batch in the hot phase...

Panavision paid the price for the shortcut, while Arri was taking their knocks learning their tech.

Arri played the long game. Panavision played short.

and now ARRI have this: http://www.arrirentalgroup.com/alexa65/

o.O

intracube: Isn't it basically vistavision gate?

giving 2.11 AR

(open gate is 52.5 x 23)

hmm?

but the 1.78 dimensions are 42.2x23.8

their 1.5:1 mode is VistaVision

yep, as you have to crop the width to get 1.78 or 1.85

but 2.39:1 can make use full width

Guranteed.

2.11

VistaVision for the record is 37.7x25.0

even if they capture full aperture

you think 2.39's days are numbered?!

but, but all the hipsters LOVE it!

I love it.

But I love it because of the depth of field, and 65 sort of outflanks that issue.

troy_s: you like anamorphic lenses?

Yes I love them. Aside from the physical side that sort of sucks (the image quality is rather crap on most breeds until you hit 5.6)

they can be tricky to shoot with.....

weight, minimum focus distance, barrel distortion on many lenses wider than 50mm :)

never held an anamorphic lens

:)

troy_s: I've never seen a definitive answer to why people have historically shot anamorphic 35mm

intracube: More resolution and aesthetic difference.

but is there more resolution when you factor in the lens issues?

Rather like "Extended" DVDs.

looking at bluray remasters of anamorphic films there seem to be rather a lot of soft lenses out there

Hell, one of the greatest shooters of all time, Matt Mahurin, used broken lens elements.