Impossible Colors?!?!

[blue-j]

In converting a wavelength of 480.5 to various RGBs via XYZ, I noted that when I went to encode linear RGB [-0.16635786 0.21035504 0.78802606] but the negative R value was not acceptable to:

/models/rgb/transfer_functions/gamma.py:96: RuntimeWarning: invalid value encountered in power return as_float(a**exponent)

The command I used was:

encoded_rgb = np.array(colour.RGB_COLOURSPACES['Adobe RGB (1998)'].cctf_encoding(linear_rgb))

This inspired me to ask the question about impossible colors. Should we allow processing and report the negative values? Lindbloom opts for that. I am not a professional color scientist and am unsure about which direction to go for this. Right now I am thinking that if there is a negative value, it cannot be represented in the target space, and we should omit it.

• J

[blue-j]

I've decided to show the negative values as useful information, but I am not sure how to get around the encoding function. Maybe just report linear RGB for those? clamping to zero is misinformation in my view.

[KelSolaar]

I would recommend preserving negative colour values as how long you can, a negative sRGB colour might be a positive BT.2020 or DCI-P3 one.