One was shot with the ND, one without.

Thing is that by definition the filter profile cannot be well behaved. Meaning muddied highlights where pixels have been clipped.

Hmm. Not sure what to do about those highlights. How on earth does one even get white clipped highlights when doing a WB correctly. By definition as soon as you start correcting, your channels will stop lining up!

@troy_s So, I just rambled, let me make this a clear question. Normally a compute the white point by converting 1,1,1 into XYZ space, and then using that value as the white point.

However, when converting from filter space, the filter will have a cast and therefore the white point must be computed by feeding it a value that should have been white (in this case the white patch on my colorchecker) through and using that value as white. This works perfectly for non clipped pixels.

However, when they are clipped, the ratio is messed up. This always happens when you correct for a cast in an image. What I'm wondering now is how DC Raw even gets white highlights when correcting for the WB? Is it scaling the image? If so, doesn't that effect the exposure??? How else would it get the primaries to line up?

@GiantCowFilms No. That colour simply is interpreted as a white, and calculated accordingly. Remember the chart has explicit XYZ values baked into the colours. The RGB are simply diagnosed against that.

@GiantCowFilms And please... Any RGB value where you have an unknown channel due to overexposure, the result is non data. It is not a colour.

So stop abusing terms if you don't mean clamped.

So, try again. Redefine what you mean by "clipped"?

Did you overexposed the chart?

Remember that RGB is nothing more than Euclidean points, and transforming via 3x3 is simply a scale, translate, and rotation baked into one transform.

@troy_s No

@troy_s Not sure what you mean by that? How else am I to calculate the white point?

@troy_s When I use the primaries and white point to convert an image from say ND Camera Space to sRGB, over exposed pixels (blown out highlights), don't show up as white. That is the problem.

Let me explain what I'm getting at differently, maybe it will make sense. Normally the white point is set to the color that correspond with the R=G=B point. In this case, since there is a color cast on the input images, it isn't set to the R=G=B point. Therefore the R=G=B point isn't white. What I want to know is what on earth am a supposed to do in such a scenario.

@GiantCowFilms What do you mean they don't show up as white? Did you adapt?

@troy_s Yes

But the quite point isn't the R=G=B point

because there is a cast

@GiantCowFilms If you adapted, then R=G=B.

@troy_s Well not like that

I use a custom white point

so that white on the chart would map to white

@GiantCowFilms You are muddling I think. White is white with ND. It is whatever it ends up being in "ND space".

Basically, in my images, achromatics are not achromatic

You would take the ND white space white XYZ, do the scale and Bradford as per Bradford, and go to say, D65 white.

@GiantCowFilms We covered this. They never are until you adapt.

@troy_s No, I'm not talking about the output

I'm talking about the inputs

The achromatics in the input aren't achromatic

Right. Inputs won't be R=G=B.

As I said.

Generate the matrix on the ND shot.

@troy_s did that

That matrix should hold up.

@troy_s But it needs a white point

What white point does it reveal?

In RGB?

Your achromatic swatches should give you an idea (in the TI3)

@troy_s I have a value

and I adapted with that value

You can try the xicclu with Lab to get the white triplet.

@troy_s But xicclu needs a white point for that conversion

And its guess won't agree with what I want, because it assumes the chart was already WB'd which it wasn't

@troy_s Yes. That works great. And the white patches on my chart a white. Yay! However, clipped highlights aren't white, which also makes sense, but something needs to be done about it.

@GiantCowFilms You can see the whites of the achromatics. Are they consistent in the TI3s?

@troy_s Roughly yes

Those achromatics are fine

@GiantCowFilms I have no clue what clipped means. Overexposed? Non data.

@troy_s Overxposed, Non data. yes

If you overexposed, you simply have to cut the channel entirely.

@troy_s As in?

There is never anything you can do about that. You are missing information. Most DSLRs wisely simply put to max.

@troy_s Correct

I understand that

so I want to know what to do in light of that fact

What strategy should I use to make that non data white?

You can try all sorts of bullshit magic of "highlight recovery", but in the end, you have non data. It is essentially a black pixel of unknown. To make it seem appropriate, make it max 1.0 for each channel.

Non data is non data.

@troy_s Yes

but I can't just make it max

To do that I need to scale or something, but that affects the rest of the image.

So how do I max it to one for each channel?

...without messing up the rest of my picture

You could max to the known channel if you wish.

Doesn't matter too much. It just needs to end up achromatic.

@troy_s Won't that change the exposure?

@GiantCowFilms If it is non data, what is the exposure?

It is a guess on undefined data.

That is the kind of thing a desaturation LUT can address.

@troy_s Well, I mean in order to make it max at one, all the pixels will have to be adjusted to fit, including the non overexposed ones.

Otherwise there will be a band around the highlight

@GiantCowFilms Typically there is a gradation up to them. Needless to say, no.

@GiantCowFilms This is what 99.95% of DSLRs do in their processing software. If (R|G|B=max) RGB=Max

@troy_s AH HA!

That is where those stupid bands came from

I new something was messing with my data!

Look at your raws.

It //is undefined//

The actual value is unknown. It isn't "messing with your data"

@troy_s Look at my raws is very very vauge statement

There is no data there.

@troy_s No, the graduation up to it. That graduation. that is de-linearlizing my data. It was showing up in my merges

You will see what I just described. Any posterization is more than likely bad meeting.

You will see what I just described. Any posterization is more than likely bad merging.

@troy_s still vauge. With what, how?

@troy_s No. Its not posterization

There is no "de linearizing"

@troy_s You say it has a graudation up to it?

In all areas of non data, all three channels would be pinned to max.

God.

Think about a hot light.

It blows out at a point, and there is a gradation away from it.

Typically there is a gradation up to them.

That is just the nature of physics.

@troy_s Yes

No magic.

Oh I see.

I see now

got it.

Okay, that was rather thick of me. Still wonder where those band come from though :/

The only additional processing after the massaged raw data is as I described above. Any fully saturated photosite ends up setting all three channels to max.

@GiantCowFilms Are they in your raw photos?

The stills?

If not, bad merging.

@troy_s a tentative yes.

Tentative as in I haven't figured out a way to objectively prove they exist, but I can see them even with a proper transform. And they really stand out in the merge.

Then it is a camera thing sadly.

@troy_s Okay

Might be CRF influenced

Its fine anyway, its trival to get rid of that data

just annoying

cya tomorrow

Ciao!

Hi .... If you don't mind asking ... What are the most relevant CG industries//business ? I mean work you could do with blender ? I can think of gaming, architecture visualisation, product visualisation, .... But what else ?

@OldMan Visual Effects

Engineering applications (like models to be 3D Printed).

@GiantCowFilms Visual Effects, is that not Nuke and that kind of compositie

@GiantCowFilms sorry, is VFx not more like Nuke and less blender ?

@OldMan Well visual effects isn't just compositing. Nearly all modern VFX shots require a huge amount of CGI, and some are now 100% CGI.

@GiantCowFilms interesting. You mean cgi work is integrated in motion ?

@OldMan integrated in motion? What do you mean by that?