afmhot is Not A Perceptually Uniform Colormap

[Miserlou]

Cool picture, guys! But it seems like you've chosen a non-perceptually uniform colormap, which means that humans are likely to slightly misinterpret the image. It'd be great to see this historic image in something more human-comprehensible, like viridis or plasma rather than afmhot!

[achael]

Hi Miserlou! You are absolutely right that afmhot is not perceptually uniform, however the images plotted in our papers (https://iopscience.iop.org/journal/2041-8205/page/Focus_on_EHT) do not use amfhot. Instead, they use "amfhot_10us", a new colormap that is perceptually uniform and symmetric-in-chroma while approximating the appearance of afmhot. It was designed for the eht and implemented in the ehtplot library by Chi-kwan Chan and Lia Medeiros (https://github.com/liamedeiros/ehtplot).

Unfortunately, vanilla afmhot does remain the current default in the eht-imaging library for historical reasons, but this could change soon. Rest assured we thought of this issue and the paper images are indeed perceptually uniform!

[Miserlou]

Ah, wonderful, thanks for your reply!

[jbednar]

Ah, that explains a lot! The only perceptually uniform hot colormap that I've seen previously is the fire colormap included with our colorcet package, and your results looked perceptually uniform but not precisely those same colors. On investigating, it was very confusing to see afmhot throughout the eht-imaging source code, and to see very nonuniform plots when re-running your examples, but from this issue it's clear. I would highly recommend that you change the default in eht-imaging to avoid encouraging the spread of afmhot.

For anyone who needs to know why not to use afmhot, you can see that it completely muddles the top end of the color values, making most of those colors indistinguishable:

The fire and presumably afmhot_10us maps allow all the numerical values to be distinguished across the full range, making the "comb" modulations here equally visible for all values:

These differences are very dramatic in the actual data as well, with afmhot looking overexposed because all the bright values blur together:

compared to fire showing all the detail for every numerical value range:

Glad to see you caught that before publication, and please help encourage others to use perceptually uniform colormaps as well!