#apertus IRC Channel Logs

Good afternoon everyone

It's a sweltering hot day here in Sydney. Apart from that, I'm well. I've added some more text to the 'HDMI now streaming' google doc, answering Sebastian's question regarding the Zedboard's clock speed, etc. If you have the time, could you have a look?

I'm fine, thanks, working on finalizing the hdmi lut mapping

ah, I see

the HDMI encoder used on the Zedboard (ADV7511) is limited to 225MHz

so, in theory, we could go up to 225MHz given that we have a HDMI receiver which can handle this

the rest is used for sync signals, back and fron porch

2200*1126*60 = 148632000

good morning

Bertl: Pong

philippej: I appreciate it when people actually do leg work and record things. It helps for others to pick up where things were left off often when an individual departs etc. It's a great idea.

philippej: I like to give a breakdown when discussing things like this as it helps to frame / contextualize the otherwise worthless anecdotal opinions.

I guess my role is to keep track of this, and put it in a between technical and production wording

philippej: You seem to have a huge amount of knowledge beyond that. Useful stuffs to have such a mixed bag of folks.

it was easier with film, since there was the proxy thing, edl, and as I understand it, a machine that would slice final footage together. Now the lines get blurred when you can edit r3d files directly

philippej: FCPX / Avid / Lightworks are all very capable _editors_, but some folks mistake the role of editor (NLE) in a more nuanced / complex pipeline.

philippej: Negative cutter in the credits.

philippej: But the _reasons_ _still_ hold true.

It remains the only forward-looking process.

(And a process that permits cross boundary interaction. For example, tracking a falling glass and rendering it in a scene _cannot_ work in a kitchen-sink styled application. At the very least, even if an app aspired to be an NLE / 3D renderer / Tracker / Compositor, it would likely suffer on at least one of those areas for integrity of data / qualitative issues.

The idea of creating an editor that will spit out final frames is the path of pure ludicrous folly if you start trying to visualize a project.

philippej: It all comes down to buyng fully into the offline system with interchanges or rejecting it. And if one rejects it, they will have a helluva time accomodating the nuances of each unique need.

it all depends on the scale of production. i guess most people who don't get it are looking at a one man band short. Where a feature film involves so many different people with different jobs, tools and file formats

philippej: Even one-person projects require it.

philippej: The exception being newsy styled things. Although even then, such as in the case of creative documentaries, often visual effects lend a huge bit of 'connective tissue' (think visual effect motif through a short piece) that cannot be ignored.

agreed

LOL

morning everyone!

morning Bertl

hey troy_s, I was wondering, what would be the effect of switching color correction matrix and gamma lut?

:)

The issue is that the color can only be transformed on linearized data.

Color always must be transformed in the following order: 1) Invert any non-linearity to get to a linear ratio system 2) transform to linear XYZ 3) transform back to destination RGB 4) apply any non-linearity curves where needed (if EXR for example, don't bother)

crazy suggestion, is it possible to merge 1 & 2-3 (at the hardware processing layer)?

step 1 : is it also done with a matrix ?

I don't know if it's simpler, but doing all the processing at once might probably be easier for you? So I was suggesting to merge all the transforms to it in a single pass (if it can be done without side effects)

but I'm not so sure it is the simplest way to get something working right now

Bertl: A LUT only speeds up the matrix, and then you have to also worry about interpolation.

You could of course combine both the matrix and the 1D transfer into a 3D LUT.

Bertl: Also note that I believe Gabe's work only applied the sRGB transfer curve, which isn't obviously the 709 transfer. You'd need to get the correct 709 transfer on the end.

but there is no mechanism to do any kind of gamma correction after that

and the "gamma" is derived from signal type?

the CSC is 13bit signed plus 2 bit adjustment, and I can probably feed it with 12bit input

i would worry about that assumed de-gamma on the input - csc does 13bit after it is linearized, right

I would assume so, the documentation for the encoder is not completely clear to me

Bertl: The 3 offsets are singular though?

(not sure what the term offset will do to the matrix other than ... well offset it )

each channel in the CSC basically dos X*A1 + Y*A2 + Z*A3 + A4

where X, Y and Z are the input components (R/G/B, Y/Cr/Cb)

guest: That should be a scale, not a shift.

it is shift - you move the zero up to 16

defined by scale in the matrix

So overscale it then offset?

But it can't do it non-unformly and it needs to be for the difference between the 709 Y' and Cb' Cr'.

a matrix is never non-uniform :)

guest: Matrices are all uniform I believe, as in linear.

(In this case, the Y' has to be treated differently than the Cb' Cr'.)

And that is assuming we worry about broadcast YCbCr 709 and not simply dumping full range 709 to the HDMI display.

Bertl: Because if it does, then the correct transform to YCbCr would be needed (which adds another bloody layer of complexity going from RGB to Y'Cb'Cr'.)

guest: Of course there is.

Or at least as far as I understand your question.

Not necessarily the model.

Good question though.

guest: Hold tight

guest: So the only thing is the scaling as applied to the two different formats (both of which are discussed)

Curved

the final product, understand

(Which counters my point above. Viewing suggests BT.1886 which glancing at the transfer looks like sRGB)

I lied

Pure gamma.

okay, so 12bit 2:4:2 RGB (or YCrCb if you like) is working on the ADV7511 \o/

Bertl: It's not Y'Cb'Cr' - it's RGB isn't it?

hose

And if it is, the only real transform then is A) convert to 709 primaries (via the Gabe matrix as it is sRGB)

currently I'm outputing memory data, so it is whatever I write there

it can be interpreted as RGB or YCrCb from the encoder

so, assuming we have 12bit linear input, we probably could do the required transform in the CSC, as the gamma correction should be done in the receiver

Bertl: How would it do that?

Bertl: In the output device?

well, if it does YCrCb to RGB (and vice versa) it should do the job for color correcting our input as well, no?

we could even send xvYCC data

Bertl: You'd still need to do all the same things to the RGB data from the sensor to get it into a suitable Y'Cb'Cr' transformed destination.

Bertl: No clue what the CSC is or how it works.

before what?

but given that the HDMI encoder takes RGB input and can apply a 3x3 matrix (including offset) before sending out the data as RGB or YCrCb, why shouldn't we be able to use that (for now, as a hack)

note: I'm not planning to do that in the future, there is no point as we won't have the HDMI encoder anyway

Bertl: And better, using BT.1886 means that it is a pretty simple straight gamma. Although in field it would likely need to be lower as ambient light woes.

i.e. actual kind-of-calibrated values?

looks interesting, can you convert it to the values for the matrix?

I'm not sure I know how all this works, so a probably not the best candidate; From here : http://www.color.org/opensource.xalter I see there is atool to convert to xml

Bertl: Gabe has some in his software to dump a DNG

would this help : http://www.mathworks.nl/help/images/ref/iccread.html ?

Bertl: You can actually strip the marix right out of an ICC.

it's orangish

Bertl: iccdump should do it.

philippej: It does

philippej: You or Bertl probably know the math for the matrix transform.

iccdump gives me two identity matrices :)

Bertl: Then his profile isn't a matrix

my original disclaimer : I just wanted to test the tool

I had to change some parameters to have a lower error level

I think you want the first one, a2b and not b2a

(wild guess again)

Bertl, and this one : http://public.philippejadin.be/it8-tungsten-01-12ms-perfect%20bis.icm ?

Bertl: https://github.com/apertus-open-source-cinema/alpha-software/blob/master/misc_scripts_and_utilities/dng/src/utils/CMV12000_sRGB.cpp

Bertl: I apologize for not working on them more. I've been working on this though...

Bertl: Can you post an image with that transform matrix of Gabe's please?

{0.01493516, 1.87764304,-0.61132345},

gotta go, good luck !

Ok off for some shopping. Shall return soon enough. Hit me up with a link if you get it Bert.

Bertl: Why on earth did we get all those over exposed images with NO DAMN CHART

ERF

_c or _c1, _c2

The ONLY one with a recent datestamp

IIRC, sebastian has done a number of images with chart as well