Improving the Camera: The Image

[Sabovic Adis]

I don't see how a LUT which works outside of the SDK and debayer fits in this particular scenario - Improving the camera's image.

Besides being a RWG/Log3G10 to Rec.709 transform (in a LUT), it's a correctional tool based on - ask Hrvoje what.

Don't lose the actual goal out of sight: artistic intent. Everything is just aids of getting us there. Some cameras may give you more trouble in fully achieving the desired artistic intent, because of certain inherent characteristics or limitations or because of specific lighting situations .

Hm, I don't mind artistic intents (there are many different ones as there are many different artists) , but there has to be some techical correctness too. Right?

Camera optimization can generally be done to some degree. A LUT is limiting (but better than nothing) compared to full control (--> new RED SDK with new color science, optimized at debayer stage). This right here is static adjustments, ergo various versions of the LUT solution. And naturally, it can never compensate for ur particular unit variation of: sensor, OLPF etc etc etc

Like Aharon said: this here is one out of many tools - and many of these other tools will do a much more suitable job for a given shot. Just look at all the example shots in this thread. Very easy to see. And those are handpicked examples.

Try it out, see if u like the result. If u wanna see what the LUT does simply run it through a LUT analyzer. But all not needed, just look at the image, and see if the solution aids u in achieving ur artistic intent.

Funny, I always thought that the more "controls" You have the more You can express Yourself - artistically. And, that there's only one, static, technically corect, setting that doesn't need any "controls" in order to work as it's supposed to.
:biggrin:

 

[Mike Nagel]

Hm, I don't mind artistic intents (there are many different ones as there are many different artists) , but there has to be some techical correctness too. Right?

Funny, I always thought that the more "controls" You have the more You can express Yourself - artistically. And, that there's only one, static, technically corect, setting that doesn't need any "controls" in order to work as it's supposed to.
:biggrin:

In my last post, YOU are the artist. You can do whatever you feel pleases you. ;-)

There is nothing "technically correct" here. Bayer data. LUT cannot know the original scene, nor can it dynamically adjust to specific lighting scenarios. It tries to subjectively compensate for whatever the dev wanted it to.

Ultimately - at the end of the signal chain - only one thing counts: how does it look to your eyes.

 

[Sabovic Adis]

In my last post, YOU are the artist. You can do whatever you feel pleases you. ;-)

Ha, artistry is what I've been running away from. :rofl:

But I get You, Mike. :thumbsup:

There is nothing "technically correct" here. Bayer data. LUT cannot know the original scene, nor can it dynamically adjust to specific lighting scenarios.

Not by itself, no.

It tries to subjectively compensate for whatever the dev wanted it to.

That's where I think You're wrong, Mike. It's not the Lut that compensates, it's Hrvoje and not subjectively but objectively - based on his findings. :biggrin:

Ultimately - at the end of the signal chain - only one thing counts: how does it look to your eyes.

It depends.
It depends on what I'm comparing it against. Right?

 

[Mike Nagel]

That's where I think You're wrong, Mike. It's not the Lut that compensates, it's Hrvoje and not subjectively but objectively - based on his findings. :biggrin:

;-)))) hahahaha yes, this is what Hrvoje is trying to sell to people

That, and that he is the only one who understands the HVS... ;-)))))

It depends. It depends on what I'm comparing it against. Right?

It does not (Unless you're trying to match a reference). YOU are the artist. You decide what u want to do with a given shot.

 

[Sabovic Adis]

;-)))) hahahaha yes, this is what Hrvoje is trying to sell to people

Well, I have no reason not to believe what he said. :rofl:

Or I'll have to redo the whole thing and see for myself...by myself. :biggrin:

That, and that he is the only one who understands the HVS... ;-)))))

I wouldn't go that far. :rofl:

It does not (Unless you're trying to match a reference). YOU are the artist. You decide what u want to do with a given shot.

Yesss! The reference!

I mean, how do You calibrate Your monitors, Mike? By eye-ballin?
"All" You have to do is give the "artist" a flat, white piece of paper or canvas (which is not easy achievable at all, You know that) and the rest is his "artistry". Right? :biggrin:

 

[Ryan Sauve]

My results tonight after buying the primer...

R3D Download Link: A001_C004_0208G9.0002349F.R3D

 

[Mike Nagel]

"All" You have to do is give the "artist" a flat, white piece of paper or canvas (which is not easy achievable at all, You know that) and the rest is his "artistry". Right? :biggrin:

the calibration reference does not work here, as those are standardized targets. There is no standardized color grading target for a given shot. so unless u have to dead on match an existing reference shot - on any new scene, a reference/direction to the colorist/artist is usually to invoke a certain "feel"... it's up to the colorist to translate that into the image...

this has been discussed in another thread, but re ur point: I personally prefer a rather neutral starting ground, although that is not necessarily required depending on where we're going in the shot. More importantly, even with linear RAW capture, there will be a penalty after so many transforms and u will thin and thin the RAW, with negative consequences... there's many parameters involved in the equation, and I would advise to understand what destructive manipulations vs. non destructive are etc

 

[Wayne Morellini]

There is a bunch of stuff in, which fixes the image to a possible extent, while mapping it to a standard.



Gamma/gamut conversion tools are something else and do not correct camera-specific image offsets and "auto" chart based algorithms are rudimentary, dumb, borderline usable and fundamentally flawed.



Thanks Fabricio.

Thanks Hrvoje, I think I get what you mean. Your explanations are just a bit dissociated at times, jumping from one thing to another without linking the association in language.

I remember, I forgot to ask if you are doing the conversion more accurately in 3D color space? Which I guess is your meaning with the gamma etc conversion, that you are doing them all based on 3D conversion space for more accuracy. However, I was asking about the conversion between different standards, wherever your tool does more than other tools export conversion process between standards?

Thanks again.

 

[Sabovic Adis]

the calibration reference does not work here, as those are standardized targets. There is no standardized color grading target for a given shot. so unless u have to dead on match an existing reference shot - on any new scene, a reference/direction to the colorist/artist is usually to invoke a certain "feel"... it's up to the colorist to translate that into the image...

Oh, think that only the monitors get calibrated (and recalibrated, for that matter)? For monitors we have test signals, and for cameras we have charts. Those are pretty standard too.
(I wasn't talking about "standardized color grading target" because this isn't about grading at all...)
:biggrin:

this has been discussed in another thread, but re ur point: I personally prefer a rather neutral starting ground, although that is not necessarily required depending on where we're going in the shot. More importantly, even with linear RAW capture, there will be a penalty after so many transforms and u will thin and thin the RAW, with negative consequences... there's many parameters involved in the equation, and I would advise to understand what destructive manipulations vs. non destructive are etc

With this I agree, 100%! :biggrin:

 

[Hrvoje Simic]

This is a 3dLUT. There are many of these out there and most are film emulations. You can create your own 3DLUT is resolve by grading a shot and right clicking on it in the color page and exporting.
These Omeneo LUTS are LOG to REC709 and are tweaked for subjectively better color and don't seem to fully expand contrast to REC709 so they could be useful to color on top of provided they are not overly destructive causing banding or breakup etc. For anyone looking to build something similar in resolve I would recommend building a look by rolling back highlights and tweaking color with hue vs hue and hue vs saturation, setting up a mild s curve for a pleasing correction that can be exported as a LUT and used on set etc.

In this thread there is sufficient information telling you otherwise so you don't need to guess and for some reason make statements about something you are unfamiliar about.

Good luck finding something like this. Or getting this from Resolve. Or anything.

I don't see how a LUT which works outside of the SDK and debayer fits in this particular scenario - Improving the camera's image.

You can always try reading the thread.

Average online attention span is nowhere near sufficient and although guesswork takes less time and mental resources and sometimes goes on "auto" to save them, it often leads to fantasias.

 

[Mike Nagel]

Oh, think that only the monitors get calibrated (and recalibrated, for that matter)? For monitors we have test signals, and for cameras we have charts. Those are pretty standard too.

with monitors/screens/display sources we calibrate in the center of the screen with a probe. We do NOT take into account the uniformity issues of the screen (outside the center measurement spot) as no 3D LUT can compensate for that. This is why I use high-end grade OLED only as primary grading display.

read more here: http://displaycalibrationtools.com/display-calibration-guides-and-workflows/lcd-oled-screen-uniformity-tests-and-comparison/

test signals serve a purpose but are bare bones limited. U need a high-end probe (and other setups). We re-calibrate screens because the screen drifts over time

camera charts are good but cannot help u with mixed, non-uniform lighting conditions and u need to be very careful in those scenarios on what patch values u can trust as a reference...

 

[Hrvoje Simic]

;-)))) hahahaha yes, this is what Hrvoje is trying to sell to people

This thread is sharing the project with fellow men.

Sales pitch and profit priority would necessitate wasting time and energy on bending over backwards and convincing folks to make a buck. Who seeks, expects or projects that here, programmed by consumeristic OS, is wasting time and energy.

 

[Wayne Morellini]

with monitors/screens/display sources we calibrate in the center of the screen with a probe. We do NOT take into account the uniformity issues of the screen (outside the center measurement spot) as no 3D LUT can compensate for that. This is why I use high-end grade OLED only as primary grading display.

read more here: http://displaycalibrationtools.com/display-calibration-guides-and-workflows/lcd-oled-screen-uniformity-tests-and-comparison/

test signals serve a purpose but are bare bones limited. U need a high-end probe (and other setups). We re-calibrate screens because the screen drifts over time

camera charts are good but cannot help u with mixed, non-uniform lighting conditions and u need to be very careful in those scenarios on what patch values u can trust as a reference...

5D lut

Seriously, they don't design anything to account for uniformity? You won't get an actual impression of the whole scene without it. Isn't there a 3D plus a 2D compensation map (which I imagine can be as simple as 1 to 3 values per channel per pixel, to get reasonable results. As per pixel should have a large amount of estimatible response once we know the light level at a point from it. A lot of this should he estimatable ramping differences from point to point, so the 2D map might consists of as little as 100-1000 points, with at least points being relocatable for obstructions related to LCD backlighting, and the led matrix being auto compensated for. That is doable even for cheap display (but won't necessarily give it high grade performance).

 

[Mike Nagel]

5D lut

Seriously, they don't design anything to account for uniformity? You won't get an actual impression of the whole scene without it. Isn't there a 3D plus a 2D compensation map (which I imagine can be as simple as 1 to 3 values per channel per pixel, to get reasonable results. As per pixel should have a large amount of estimatible response once we know the light level at a point from it. A lot of this should he estimatable ramping differences from point to point, so the 2D map might consists of as little as 100-1000 points, with at least points being relocatable for obstructions related to LCD backlighting, and the led matrix being auto compensated for. That is doable even for cheap display (but won't necessarily give it high grade performance).

once u start selling ur 5D LUT box, we can all integrate that. Good OLEDs don't need it though ;-)

 

[Phil Holland]

5D lut

Seriously, they don't design anything to account for uniformity? You won't get an actual impression of the whole scene without it. Isn't there a 3D plus a 2D compensation map (which I imagine can be as simple as 1 to 3 values per channel per pixel, to get reasonable results. As per pixel should have a large amount of estimatible response once we know the light level at a point from it. A lot of this should he estimatable ramping differences from point to point, so the 2D map might consists of as little as 100-1000 points, with at least points being relocatable for obstructions related to LCD backlighting, and the led matrix being auto compensated for. That is doable even for cheap display (but won't necessarily give it high grade performance).

Depending on the manufacturer, tech, build method, and level of quality control - yes they (manufacturers) do map out and calibrate displays in "quadrants".

 

[Mike Nagel]

Depending on the manufacturer, tech, build method, and level of quality control - yes they (manufacturers) do map out and calibrate displays in "quadrants".

if I understand it correctly, the quadrant approach doesn't help and actually introduces some other possible major issues for screen performance (because of the sections). But I think Wayne was referring to a 2D map to go along side the 3D cLUT - the 2D map being the uniformity map so that the 3D cLUT could adjust accordingly.

There is no tech that does this and there possibly never will (man, manufacturers are 2+ years behind on 3D LUT boxes for 4K). Good OLED panels solve all of this and once the tech gets cheaper, the stellar performance of the high-end OLEDs will trickle down to the consumer models.

 

[Hrvoje Simic]

My results tonight after buying the primer...

Thanks Ryan. Those with milder knee and toe are meant for specific scenes such as higher DR and allow you to shape your own knee and toe according to need and preference.

There are few others with different responses giving a more contrasty image.
Please make sure you use 64^3 Primers from "post" folder, those labeled with "shoot" are 33^3 for monitoring purposes.

Also, take caution with RCX, not sure about the LUT interpolation at this stage. In Resolve make sure you turn on tetrahedral LUT interpolation in project color settings.

 

[Wayne Morellini]

if I understand it correctly, the quadrant approach doesn't help and actually introduces some other possible major issues for screen performance (because of the sections). But I think Wayne was referring to a 2D map to go along side the 3D cLUT - the 2D map being the uniformity map so that the 3D cLUT could adjust accordingly.

There is no tech that does this and there possibly never will (man, manufacturers are 2+ years behind on 3D LUT boxes for 4K). Good OLED panels solve all of this and once the tech gets cheaper, the stellar performance of the high-end OLEDs will trickle down to the consumer models.

I could have meant nothing else. . Good call on the quadrant thing Mike. I understand consumer OLED has issues, and a conformity map could be cheaper than getting consistent consumer oled in the short term. I wonder if Qled individual pixels will get better conformity. However, given a 16 bit display if 3D+ puts were built into GPU cards (easily) that would solve issues. If you had a display back channel to pass the manufacturer balance data back, a program could use that as a starting point. If HDMI had a balance data format and communications command, a computer could be used to change the internal balance settings. If they don't have any of these the industry is behind.

BTW Samsung has announced in built calibration support on some TV's I think.

 

[Aharon Rothschild]

In this thread there is sufficient information telling you otherwise so you don't need to guess and for some reason make statements about something you are unfamiliar about.

Good luck finding something like this. Or getting this from Resolve. Or anything.



You can always try reading the thread.

Average online attention span is nowhere near sufficient and although guesswork takes less time and mental resources and sometimes goes on "auto" to save them, it often leads to fantasias.

1. This is a subjective 3dLUT as it is not a transform to a known target like REC709. As such it's usefulness is limited by how well your subjective corrections play out on different shots and by how destructive the LUT is.
2. Having looked at the images posted here and on your site and having read through the thread, it becomes clear that what needs to be explained is that there is in fact no way to Improve the image outside of the SDK. Change perhaps, but your LUT will only manipulate what is already there and is destructive, meaning there is a penalty for every change. As such it is not a "Primer"but a Artisitc LUT similar to a film emulation or any other LUT that offers a subjectively more pleasing image.

 

[Sabovic Adis]

with monitors/screens/display sources we calibrate in the center of the screen with a probe. We do NOT take into account the uniformity issues of the screen (outside the center measurement spot) as no 3D LUT can compensate for that. This is why I use high-end grade OLED only as primary grading display.

And how do You compensate for hardware errors? With software? Just by remaping c.v. around?
Aren't You inducing banding by doing so?
My point is if it's acceptable with monitors why wouldn't be acceptable with cameras?
:biggrin:

read more here: http://displaycalibrationtools.com/display-calibration-guides-and-workflows/lcd-oled-screen-uniformity-tests-and-comparison/

I know this...

test signals serve a purpose but are bare bones limited. U need a high-end probe (and other setups). We re-calibrate screens because the screen drifts over time

Oh, not only that they drift but they already come messy from their manufacturers too.
Tbh, I don't calibrate my (new) monitors any more, I just re-watch my movie library of couple thousand movies and that's it. I recalibrate myself to the monitor, so to speak...:rofl:

camera charts are good but cannot help u with mixed, non-uniform lighting conditions and u need to be very careful in those scenarios on what patch values u can trust as a reference...

We're talking proper focus, proper lighting, proper exposure - everything proper here, right?
I'll not even try to "corect" an 3 stops unintentionally underexposed shot - GTFOOH is my gut response to such things! :rofl:
:thumbsup: