#apertus IRC Channel Logs

Sorry for the lab bot spam. Getting the ideas simply out helps free up my mind of thoughts.

Clarification on the port

Or it could be some sort of colour range issue, or encoding/transcoding issue, since most films i tend to see of it is presented online in one form or the other

But yes, i think the bmcc is good at making amateur film look amateur, which is a good thing in many ways

morning folks!

off for now ... bbl

open sauce

hello

hi Bertl :)

AFAIK, not yet

like fairly linear for shadows and mid-tones, but an extreme slope to capture highlights

it's always linear (piecewise linear :)

well, the uppermost piecewise linear slope! :P

like do decent physically correct lens flares

intracube: It makes exposure a little more fussy, but it is also precisely why the existing high-end cameras record to log.

troy_s: what about flares?

oh, meaning in post you'll have access to a wider dynamic range compared to if it was recorded fully linear

"fully linear" makes very little sense with current technology levels in cameras.

And you still have a large contingent of bumblefsckery that is trapped in display referred imaging apps thinking it is the only way.

There's no way to do that with PLS either.

which I think is why flares done in post on 8bit camcorder footage usually look poor

troy_s: right, but there will be -more- range than otherwise

You can do a luminance key to yank values and log them up to relative ranges.

Really _no clue_. Nor why it is important.

The way a flare interacts with the existing image is very similar to the way that fire adds.

In particular, the values have got to be composed in a scene-linear format, and you can't use a blind clip on the view transform.

As in, I could assure you that a 20 stop PLS curve will still result in crap work if artists are happily churning away on AE.

(and even when you do flip to it for imaging, controlling that transform back to display referred land is rather important for that roll-off you were speaking of.)

wouldn't you just have an inverse transform to turn the PLS curve to as close to linear as possible

then in blender's comp (for example) apply whatever flare effect you want, then finally grade to taste

right

So you have an issue here... in particular two issues:

B) How you are going to deal with the roll off to display referred white.

B is subtle and more nuanced.

You can see why B is a complex issue?

B is an artistic choice

(what happens close to clipping point in display referred land)

As in

Okie?

Now let's say intracube the grader decides that he wants his white clip to live at something around say.... 35.

The 103 is a unique problem...

where we might want a roll off

Not always.

You might want that 35 level to be blooming to white

which is actually a very unique intensity transform

Something like "keep values at their value up to display referred 0.9

we actually want _all_ the wells to saturate

If we pretend we shot that green barn on film... and let's pretend it is a mythical narrow band green... the value that perfectly saturates the G register in digital.

Which is _not_ what you get with a hard digital clip, curves or otherwise.

but when green steps over 0.9 to 0.91 perhaps, then we want to also rapidly bring up blue and red.

You can demo this very easily in Blender by creating a fully saturated R G or B face or use a neutral white and use a fully saturated R G or B light

troy_s: of course. this is something that a grader/colourist would do instinctively

Not really.

(most folks around Blender for example, think the default LUT is "blowing out" values above the display referred land. This is not the case. It just happens that very hot intensities happen to push even moderately saturated R G B to clip.)

Not really.

Highlight rolloff is one thing.

Where display referred one can almost always assume (except for intermediate protocols like broadcast YCbCr with scale and offsets) that display referred 0 and 1 are black and white respectively.

Except many folks screw around in half assed display referred models.

And you certainly can't just do uniform curves

because that would completely squonk your looks.

Except more pronounced.

It makes the issue much more complex.

oh ok

yep

Not the most easy thing to deal with curves

You need sort of a converging film-ish 3D LUT (and it must be 3D because the single channel must slide the other channels)

You _could_ probably manually mangle the crap out of curves per shot

intracube: I just emailed a buddy on a tool to generate those sorts of LUTs.

intracube: Did you know that the 7D can record raw now offspeed via ML?

^ +1 offspeed?

when is ML going to hack Nikon FW?

sure about that?

Bertl: I mean uncompressed, not pre-debayered

no

but HDMI to an Atomos is obviously going to be much better

the BBC don't transmit 'HD' at all (by their own internal definition)

Bertl: On the 7D.

troy_s: yeah, but unlikely via HDMI

so 2.5x for 60

troy_s: do you mean native 60fps?

With aspect ratio native 2.67:1

2.67:1? eww!

Figure why waste pixels really

but it has some fine granularity actually

alexML_: Ping.

How come?

troy_s: because for the last 10-15 years widescreen has been pushed hugely

More 21:9 displays out there now than there ever has been.

'same height as 4:3 but more at the sides' - so wrong

troy_s: 21:9 is still quite niche

21:9 is wider. So I'm a little confused as to your meaning (I know you know what you mean, but I just lost the message)

Than 16:9

Erm.

you could argue 16:9 is wider than 12:9 (4:3)

because of the implied X

as in 21:9(x) is wider than 16:9(x)

10 seconds... 200 megs!

common height is an assumption

try common width

Hrm.

It isn't the case.

as there is a squish in there to gain quality

troy_s: blu-ray is a poor format. 16:9 all the way. no proper provision for 2.39 or 1.33

= wasted pixels

No wasteds.

afaik blu-rays PAR is always 1:1

blu ray??

Yes

But no real clue on Blu-Ray.

I think there is even a push for 21:9 enhanced

(Pretty sure it isn't there yet)

intracube: Sure. I'm sure Google is wiser than I though.

Not entirely sure.

(As compared to DVDs which had an explicit meaning at one point)

troy_s: the HDMI spec would answer this

He should skulk in here dammit.

???

I'm sure there is, but why would it answer it?

Oh. I am certain 21:9 isn't there because there is a movement to add it, but I'm not entirely sure that some don't have some form of squeeze.

Only know I have an enhanced (or a couple) of BRs.

is likely marketing guff :P

Pretty simple really.

and hi bassam / slikdigit / Blenderer :)

alexML_ / g3gg0 I am blown away. It has been so long since I futzed with ML but 60 fps raw on my older 7D is ...

heya intracube :)

troy_s, I think I'm actually contemplating reading CIE publications....

slikdigit: May I ask why?

I'll say first and outright, CIE stuffs is sort of level three.

I'd suggest solidifying your second level first, because much of the CIE stuff really needs a good carpet bombing of basis.

I'm interested in the math

What part of the math?

To be brutally honest, if you can wrap your head around the following two sets of general knowledge, the rest is dead simple.

so there's a fair bit I'm going to read before then. I'm dreading the cie stuff because I think it'll be really dry

I have a few tomes you might want to peruse.

It generally is quite dry

That is, _any_ tri color space can be changed from one to the other via a simple 3x3 matrix

1) I don't get yet (terminology lapse), but 2) I do

time is not moving as fast as i would like

XYZ is kinda cool

It is:

I wonder if renderers should start using it internally (or if some already do)

0.2126729 0.7151522 0.0721750

I'll explain that in a second.

that will blow your mind

aombk: there is a lot of stuff which needs to be done

That's the default D65 sRGB to XYZ transform. (Which also means it is totally wrong for _any_ ICC based workflow)

See the MIDDLE row?

aombk: currently most work needs to be done on designing the PCBs

ok then i better not speed up time

oh slikdigit, hi!

hia aombk

troy_s, cool stuff, and I hadn't seen that wiki page before, silly me

slikdigit: Do you see what I am getting at yet?

evening se6astian!

What you see in that matrix is actaully the XYZ positions.

So Red's XYZ is 0.4124564, 0.2126729, and 0.0193339

(display/device/output referred are all synonymous - values have 'special' meaning at 0 and 1.0, that is there is a minimum and a maximum)

oh neat, yeah, makes sense; but I don't know yet what display vs. scene referred means

slikdigit: IF your model has no special meaning to 1.0, THEN you are likely using a scene referred model.

Bertl, ahah! I was a bit confused there

Bertl, and i am not yet able to browse in the lab and understand whats what. i know i have to watch the whole lab perentation video

hi Bertl

slikdigit: An internal raytracer is scene referred, and the values extend from zero to infinity representing the scene, and obviously all radiometrically correct (aka linear model)

right, so I'm guessing that's scene referred values since you can't actually display it

The transform from scene referred to display referred is a creative choice.

Black and white are given meaning in display referred, and they don't exist in scene referred.

In Blender for example, if you flip open the color management panel in properties, you can turn on "Use Curves" and map that point light setting

right, that pretty much matches what blender does

Most colorists add a step in there before worrying about sRGB

the log is sort of a base entry point for grading

and applies a rough curve

if your scene extends rather high into values, the log will stuff those down lower, so some tweaking may be needed to find the middle grey exposure.

And _all_ of that creative stuff

You only bake to the data when you are outputting generally.

For example, say you are doing a grade in WfE

maybe 20

but.... I'm guessing you *should* be able to apply a single different transform to get an equivelent (depending on the characteristics of the displays) look on other displays (like a film print vs. computer monitor)

Yep.

B) Use more sophisticated primary and secondary and bending transforms

In the case of B, you can take your node setup, and use an RGB image with all possible colors in it, or slightly less, and do a 'before' and 'after' image.

as an entry point.

interesting

we could have had a fun talk/demo

I was _really_ pissed when I missed the last one here (was on some nightmare)

slikdigit: Looking for something for you

by a small margin of 123,986 € :)

yeah

troy_s: ping

damn

baldand: Great to see you here. Had no clue you skulked

REALLY amazing.

baldand: I was hoping I might be able to twist your arm to integrate two libraries that will make your world a hell of a lot easier.

baldand: So that we can use the plethora of LUTs and matrix transforms out there.

baldand: Familiar with those two libs (They have python bindings as well)

baldand: It has GPU previews, with CPU for higher grade.

troy_s: I've been trying to keep as much as possible on GPU, AMaZE being the ain exception

baldand: GPU is great for previews, sucks for quality generally.

But you can flip back and forth via OCIO painlessly.

Sure. But they have always had tradeoffs.

but an online? You want the highest grade no matter what the performance cost. Especially given you are only conforming / going to archival once.

slikdigit: You could generate those LUTs very easily.

dmjnova: Possibly. Not too familiar with that aspect, other than the GLSL shaders seem to work fine for some algos and others not so much.

I'm talking specifically about CUDA and OpenCL

dmjnova: Most of the accel code I see is shader

sebastian reopened T41: Create crowd funding backer FAQ article as "Open". http://lab.apertus.org/T41

sebastian-test committed rBH52cd73b4d336: added new backplate variant and top plate riser (authored by sebastian-test).