#apertus IRC Channel Logs

How are you going tonight dmj_nova?

mulling over cinema camera forms

not yet

been thinking about how to incorporate those thoughts into an AXIOM camera

I've worked with Canon and Pentax Dslr's, the H16 Bolex (with reflex viewfinder) and panasonic HVX200

Okay

It seems that there's a bit of a divide between "box" (DSLR) camera ergonomics and "balance" (shoulder) camera ergonomics.

this: http://www.aaton.com/products/film/delta/index.php

In that last one, the rig is probably more expensive than the camera?

Sasha_C: yes it is

I think I should come up with a concept focused on ergonomics and simplicity of use

do they generally operate cameras right-handed?

rexbron: thoughts on this camera: http://www.ikonoskop.com/begood/image_db.php?id=224&w=700&ne=1

(for left 'handed' operators)

Bertl: thanks

So it seems okay to design a control scheme focused around right-handed operation

I was told, that in professional environments, the cameraperson doesn't do anything but aiming the camera and controlling the focus

that everything else being exposure, exchanging the magazine, and what else?

I think almost everything which can be controlled on the camera is usually done by the assistant

Well, I suppose a shoulder-mount system can have two methods of interface:

I'm pretty sure that is where indie differs from pro

I don't think that 'pro' camerawomen want to bother with anything in the viewfinder except sharpness and content (i.e. target)

The same viewfinder hardware would be able to accommodate either HUD+content or simply content though

What we do need is specific modes of operation with sane defaults

2) "pro" mode which is tailored to 2-3 person teams

in what regard?

They couldn't even bother to conform the grip to one's hand

maybe they had just a person with unusual hands :)

yeah, maybe a stone golem :)

Which one do you want to hold

the DSLR, because it seems a lot less weight?

well, probably depends on the lens system

ah, yeah it is heavier than most dslrs

It makes sense in the case of eyepieces because some cameras don't give you an option, so you learn right.

Further, if operating right, the idea is to be able to open your left eye to see, so again, left eyed operators train right as convention.

btw, what is nowadays the political correct name? cameraperson or camerawoman/men or still cameraman/men?

troy_s: as far as pulling focus, how critical is latency?

ah, nice, tx

dmj_nova: There is zero latency. If there is latency, the system is generally avoided like the plague.

zero noticeable latency that is

I assumed that it would be a big deal but had some implementation thoughts

General rule - there simply is zero latency.

that means you need the focus puller physically tethered and viewing a direct feet from a physical cable

But it is a defacto norm for even the lowest budget productions to have a 1st to pull focus.

yes, we'll always be reading the previous frame

because we can't get sensor data before it has been collected and read

dmj_nova: The latency is about zero. Even a frame can screw you up.

troy_s: it is technically impossible to get below one frame in a global shutter setup

Bertl: Just saying that the systems in use are effectively perfect frame sync

perfect one-off frame sync systems, yes :)

two lenses has obvious implications for framing and focus

Bertl: Not sure how the Alexa or Sony's deal with it, but it is certainly not a frame out. Likely due to 180 shutter

well, technically I suppose one could shoot 48fps

and toss half the frames

No. That would bugger your motion blur.

Aesthetic dictates that 24/25 is the learned baseline.

180 shutter at 24.

the funny thing is, human vision cannot 'react' faster than 150ms

180 shutter means you capture for 1/2 of each 1/24 second period

You realize I know this very well right?

during that "dark" time, you could sneak a frame in that you don't actually keep but allow the operator to see for focus

which actually means that at 25FPS, we still can skip 2 frames and 'get away' with it :)

dmj_nova: That is sort of what I am suggesting. That is in fact precisely how older mirror systems worked.

Bertl: troy_s has a really good idea here

a single _additional_ frame, yes I buy that

180 degree shutter means that the sensor is only being used for 1/48th of a second to capture a frame every 1/24th of a second

dmj_nova: Only issue is variable shutter - where you may be at a wider angle.

that gives us about 1/48th of a second to capture *another* frame in between the ones we actually keep

So it isn't much of a solution.

doubling our framerate for focus and halving our latency until display

Extremely common

i.e. what is the point in taking a second frame?

the 180 degree shutter can actually be very important for 2 reasons

Wider is an issue if you hope to alternate for playback.

That is the technical reason. You may need a 144 etc.

*the between frames

Regardless of shutter.

this will *not* work with 360 degree shutter

I would think the image might be suck if at a 270 shutter and gained :)

That is a whole stop of light

Anyways... I am sure Bertl can figure out some magic.

aside from the clever dark-time second exposure, we can't really do more magic

As long as the refresh hz is faster than the dumping all is fine.

I only suggested that a frame of latency would be a nightmare to operate from

not necessarily

entirely depends on the exposure and transfer times

Must sleep. Ciao friends.

with 0 transfertime, yes

that extra frame would help with latency *between* frames, but not between exposure and display

which may be helpful but certainly at a cost in battery life

yes, exactly

never claimed that :)

you could at least theoretically start displaying the frame immediately after exposure ends

you cannot possibly get better than that

Bertl: you're right, but you're using the wrong terms

not one *frame* delay

exposure duration and frame duration usually aren't synonymous

unless you actually increase the frame rate, at which point you're back to 1 frame delay :)

60, 72, or 120 are common rates

yup, so if your exposure time is 1/60th second

then you can get away with 1/60th of a second delay (at best)

which is still higher than the 24FPS you might be recording

2) frame rate = number of frames per second that are captured and permanently stored (1/24th)

Bertl: yes

also: you can totally manipulate such a system to get you closer to 1/60th of a second

Bertl: true

has nothing to do with dumb actually

didn't mean to insult :)

but yes, "display output rate" might have been better

Bertl: that's only with motion tweening

which incidentally is done already with modern screens and projectors to compensate for flicker headaches even with 24 fps

it doesn't change the motion blur properties, but it makes it so you can't "feel" the flicker

or that could be a coincidence

well, modern screens/projectors interpolate the missing frames

i.e. motion estimation and similar stuff

you can compensate for those by e.g. taking two frames and combining them

which would allow you to display each frame in a view finder 'earlier' than the combined frame

Bertl: it will be noisier

not necessarily by a lot, but it will be

okay, yes, if you consider the bad quality of sensors, then yeah

well, you have double readout noise

i.e. if the readout causes a 5% noise, the combination will have 10% noise of the original value, but twice the intensity

no, not twice the intensity

in which case, the noise will actually get less by averaging

well, how does the readout noise figure into that then?

it is hard to quantify and generalize

I'm pretty sure in certain cases the noise will be better and in other cases worse, around the typical cases it will be about the same

= 2*noise + light

i.e. sample a 'darker' picture?

whereas with one exposure you get: noise + light

and on the other hand, the termal noise gets less when reduced exposure

so the averaging will negate the increased noise from the gain, but not from the sampling process itself

it will get larger with longer exposure times

but with resets inbetween

i.e. the photosites will drift away from the optimal discharge path

Bertl: you could test this

how many exposures can you capture now?

try your trick with as many consecutive exposures as possible and an equivalent single exposure

as I said, a lot will be tested, currently the focus is on getting the alpha prototype done

I've also done some work on temporal denoising

also: dark noise always adds, never subtracts

interesting that you say that, do you know how a CMOS sensor works?

the photosites are precharged to a certain voltage, and light 'discharges' them over time

you mean wrt the band gap?

here is a nice read if you're interested in

hmm...it occurs to me that I'd actually never looked up what a full CMOS pixel looked like

Bertl: okay, this is a lot to digest

I'm off to bed now ... have a good one everyone!

Bertl, troy_s, dmj_nova just read the backlog about the delay in the pipeline. There *is* some delay between reality and what is shown on the display, in any camera. Altough this delay is as minimal as possible. It is hard to say how much is too much, less than a frame means less than 41 milliseconds. I guess we can achieve much better, like 10 msec delay. Which I'd translate as humanly not perceivable.

Interesting summary : http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=5540888&url=http%3A%2F%2Fieeexplore.ieee.org%2Fxpls%2Fabs_all.jsp%3Farnumber%3D5540888

experimental results : 50-60ms - absolute limit of delay detection - small object tracking - mouse cursors

Can you add that to the wiki

yep

hmm, interesting

I didn't have any captha to filll

ok, let me take a look

I'm member of Autoconfirmed users, Users

still no edit tab :-(

can you edit the sandbox?

I think the main page is protected

well in fact

but it's getting long, and the info I'm going to put is more generic

works

lets see if spam increases as well now

even anonymous can edit

I read up on it

because bots are used to registring user names

posted here : https://wiki.apertus.org/index.php?title=Image_latency

great, thanks

when you say less than 41ms

41.666ms ?

should be filled soon :)

hello dmj_nova1, feel free to fill this with more info : https://wiki.apertus.org/index.php?title=Image_latency

hello Bertl!

look at?

the website page explaining new members the working conditions maybe?

maybe here: https://www.apertus.org/node/120

but we might need to make it more prominent

we should add your now stadard phrase of "join us on irc, introduce yourself and ask for work :-)"

good idea

let me create it

I cant edit yet

(now)

Bertl, always welcome, either here or directly on the doc

it should be fixed, maybe reload the page

Hello all.

dmj_nova1: “realistic motion blur” = misnomer. All aesthetics are emergent phenomena. Even a basic photograph is a convention. Net sum is that 24fps merely is the learned aesthetic baseline.

troy_s, I was thinking that to measure the delay between reality and what's on the viewfinder, we could film existing cameras (at somehow high framerate) with both a clap and the viewfinder in frame, to guesstimate the delay they have.

philippej: I am sure that documentation must be online somewhere.

philippej: Certainly for the Alexa or F65 etc.

philippej: I suspect it is in your estimated ballpark. 50ms or so?

ok, google doc looks fine

I will ask sasha to look over it and then edit the current page content to replace it with the new

philippej: EG http://www.teradek.com/pages/bolt

I think we met those guys at ibc

as I see it, in axiom, all the processing from sensor to sdi (for example) will be done in the fpga, so it will be as good as it gets

philippej: I only voiced concern when I was reading "only a frame of latency" which in camera work, is unacceptable.

I guess Bertl have some idea of the expected latency between reality and sdi out

The SDI out "raw" (debayered) would likely need to be higher quality, and "zero latency" goals are less important.

well, in the optimal case we have the exposure time + FOT + readout time

I wonder if gcolburn made way with the profiling.

i.e. the time it takes to copy and digitize the samples

which will be at least 1, probably more like 2 'exposure times' as we decided to call it

time to leave the office

more progress with irc channel logs: https://www.apertus.org/irc/index.php

nighty