[?BUG?] less dynamic resolution on UTM EXR than TM EXR

[zuegs]

A histogram comparision of a tonemapped and a untonemapped EXR from same render:


There seams something wrong with UTM output ???
I think is a float rounding problem... i think UTM should get a addition gain of 2^5 to not be clipped at the lower resolution border of 16bit floats.

When importing in Photoshop i have a set exposure to 21 to see something, that means values get multiplied by 2^21 and as Half-floats only have a range of 6.0×10^-8 - 6.5×10^4 or 2^-24 - 2^16 it's very probable that some colors gets rounded at lower border of half-floats.
Any one understood what i tried to say in pseudo "english"

[zuegs]

I found out that it's dependent of the total light energie in the scene. Above example was rendered with a backround:

Code: Select all

	<background>
		<spectrum>
			<rgb>
				<rgb>1 1 1</rgb>
			</rgb>
		</spectrum>
	</background>

To get a good enough result for untonemapped EXRs it needs a much stronger light as:

Code: Select all

	<background>
		<spectrum>
			<rgb>
				<rgb>1000 1000 1000</rgb>
			</rgb>
		</spectrum>
	</background>

That's not realy nice I will check some other light source types but,

Nick can you please increase the untonemapped EXR gain by about 1000000 to be in a better range?

[OnoSendai]

Yes.. i ran into this problem in some of my own tests.
I'm considering whether to 'normalise' the image (scale until the luminances lie in some decent range) before saving to exr, however this may be confusing, as it would basically be doing tonemapping, for the untonemapped option.

One workaround for you Zuegs is to increase your aperture radius to something more reasonable, i'm guessing it's like 0.00001. Smaller radius => less light

[zuegs]

Hi Ono,
yes i've noticed "aperture" dependence... I normaly use 0.0001 (1/10mm).

I also asked me why the utm-EXR luminance doesn't increase over time? Current implementation seams already to normalize utm-EXR over total-mutations . But in some cases it could be useful to get a "copy" of the render-buffer (where luminance increases over time) because if utm-EXR is fully dependent of mutations (time) it would be possible to add multiple utm-EXRs with different render-times. I mean if I have a 30min utm-exr and a 60min utm-exr with around the double luminance than the 30mins I could just be added together to get a 90min utm-exr without playing with mixing gain .

But I think currently a "normalise" feature for utm-exr would be very nice