that lime and purple sticks out like a sore thumb for some reason. Can you share a high qualtiy exr render, or alternatively, a blend file of this, I’d like to compare
FWIW a similar color checker exr I had found at some point looks like this
Top: tonemapped with my approach, bottom: sRGB clip (the teal in the third row on the right is actually outside sRGB)
it seems like I’m boosting greens at neutral EVs somewhat. Both the RGB-standin green in the third row and the like forest green actually end up a bit more saturated after my transform.
What gets boosted and what gets desaturated, relatively speaking, is entirely down to the luminance estimate. If I swap this out for something else, I can get quite significantly different results.
Above I used OKL³. Here is the exact same render using Y as luminance. It boosts green even more and reduces red and blue further
I also found one of my old spectral test renders. In this test the surprising thing is that every single object in the scene has the same exact spectrum attached to it, just as various different materials (emission, metallic, diffuse, glass)
That’s just a single spectrum but different colors - it’s a spectrum that features polychromatism. It simply looks different at different depths
If you have a hard time finding the differences here, the biggest ones are in the teal and pink parts of the quarter donut in the bottom right corner and in the pinks of the metallic Suzanne
If i am not wrong then these are the sRGB colors used in a mac beth checker
My render above was just a example what was happen with the Fresnel reflection on the checker hull,the checker image was just a photo from the checker which i have not compared the values of the numered versions.
Inaccurate. If you measure a typical macbeth spectrally, you will, in fact, find, that the teal patch does not fit into sRGB. It does have by far the lowest red value as shown here, but it ought to actually be a bit negative.
Yeah but the fact that the purple and lime stick out so much here
compared to this photo where they seem much more cohesive with the rest of the patches
made me curious
Also I think troy didn’t actually talk about the fresnel. He’s talking about the glowy outline of your patches here:
mine for comparison (this is the OKL³ version but the others are similar)
I think,he meantioned the min value of the colorswatch vs the surrounding black hull.
As said ,this is just black plastic hull,and the swatches have different diffuse or what ever produced materials which gives you the right amount of reflections.
at any rate, these two patches definitely stand out in ways the others do not
and while, in my takes, purple doesn’t seem to do so, the forest green (that looks more like lime in your example) certainly does a little bit, unless I just simply use the channel average as my luminance approximation…
You mean the Fresnel render or the Laptop bt2020 exr render?
Again,the Fresnel render was just a hull reflection example with a image from the x rite checker i posted above.
I know it’s just a slapdash quick test to show off the fresnel, but it still seems like those two swatches in particular are more powerful than they ought to be for some reason.
In your other test with the photographs, they do indeed look more reasonable.
@pixelgrip If you are looking for an accurate XRite digital color-checker reproduction you can find those here:
yep that was the intention for the dull look, trying to get something pretty neutral. However kind in mind that there is hundred of variable in the film development proces (physical and digital) that will affect the end-result, so my EXR is just one interpretation out of hundreds.
To get a better looking artistic reference you can add Saturation to the film EXR using the max algorithm (where luma is just max(r,g,b) instead of the BT709 weights)
this is an old test with the spectral branch, rendering the MacBeth chart with literally spectral values (top left of each swatch) and sRGB approximation (bottom right). For most patches the difference is nigh impossible to see. The cyan one is the most obvious discrepancy.
The point of the test was to see how reasonable the approximate RGB to spectrum process of the spectral branch was. Note, that the spectral branch for technical reasons was operating in Illuminant E white and attempted to correct for that afterward, so the whitepoint might be a bit off here.
presented here with OKL³ based luminance, average, and just clipped sRGB. Obviously this is a low dynamic range image so all that gamut compression isn’t actually necessary for the sake of luminance.
Yes i skiped the sigmoid for this render,it rolls of allot of highlights (from 95% to 80ish) even at default value.At one hand it is how its made for,on the other hand sometimes its to much.I am not happy with such tradeoff.
Yes,but for HDR images we have to rolloff the highlights as you know.It must be a more clever way to keep such images dynamic like this intact and roll of increase with higher dynamic range.A automatic calculation of the range.
