"colour.XYZ_to_hdr_CIELab" and "colour.XYZ_to_hdr_IPT" definitions are generating NaNs.

[priikone]

I'm getting nans with these when going back to XYZ. Example:

import colour
from colour.models.rgb import (RGB_COLOURSPACE_ACESCG)

print("hdr-CIELab")
RGB = [0.1, 0.0, 0.0]
print(RGB)
XYZ = colour.RGB_to_XYZ(RGB, RGB_COLOURSPACE_ACESCG.whitepoint, RGB_COLOURSPACE_ACESCG.whitepoint, RGB_COLOURSPACE_ACESCG.matrix_RGB_to_XYZ)
print(XYZ)
LAB = colour.XYZ_to_hdr_CIELab(XYZ)
print(LAB)
XYZ = colour.hdr_CIELab_to_XYZ(LAB)
print(XYZ)
RGB = colour.XYZ_to_RGB(XYZ, RGB_COLOURSPACE_ACESCG.whitepoint, RGB_COLOURSPACE_ACESCG.whitepoint, RGB_COLOURSPACE_ACESCG.matrix_XYZ_to_RGB)
print(RGB)

print("hdr-IPT")
RGB = [0.1, 0.0, 0.0]
print(RGB)
XYZ = colour.RGB_to_XYZ(RGB, RGB_COLOURSPACE_ACESCG.whitepoint, RGB_COLOURSPACE_ACESCG.whitepoint, RGB_COLOURSPACE_ACESCG.matrix_RGB_to_XYZ)
print(XYZ)
LAB = colour.XYZ_to_hdr_IPT(XYZ)
print(LAB)
XYZ = colour.hdr_IPT_to_XYZ(LAB)
print(XYZ)
RGB = colour.XYZ_to_RGB(XYZ, RGB_COLOURSPACE_ACESCG.whitepoint, RGB_COLOURSPACE_ACESCG.whitepoint, RGB_COLOURSPACE_ACESCG.matrix_XYZ_to_RGB)
print(RGB)

And that gives output:

hdr-CIELab
[0.1, 0.0, 0.0]
[ 0.06624542  0.02722287 -0.00055746]
[ 28.5406125   66.4841375   67.05114917]
.../colour/colour/colorimetry/luminance.py:409: RuntimeWarning: invalid value encountered in log
  np.log(
[ 0.06624542  0.02722287         nan]
[ nan  nan  nan]
hdr-IPT
[0.1, 0.0, 0.0]
[ 0.06624542  0.02722287 -0.00055746]
[ 21.68092623  44.55999847  41.05035691]
[ nan  nan  nan]
[ nan  nan  nan]

It's probably the negative Z, but should this work?

[KelSolaar]

Hi @priikone,

The failure here is expected, hdr-IPT and hdr-Lab do simply not support anything that is either negative or equal to zero, i.e. log(R <= 0) is undefined.

We tend to rarely handle NaN/Inf and leave that to the end user because there is no obvious way to handle those. Here, for example, we could set them to zero and swallow the warnings but then you might get unpredictable behaviour down the line. You might assume that the conversion works fine while instead there are some issues that might prevent inversion, etc...

Cheers,

Thomas