Project motivation#
Making scientific data and visualizations accessible to as many people as possible is clearly an excellent goal. However, everyone has different aesthetic preferences, and sacrificing visual appeal for accessibility is not generally necessary. If you have found that you are unsatisfied or bored with existing perceptually uniform colormaps, this project may be what you need.
Color perception differs between individuals#
Red-green colorblindness is a common form of color vision deficiency (CVD), but it is not the only one. Problems with one or more optical cones in the human eye can lead to very different perceptions of the same image. The table below gives a rough idea of how common — or uncommon — different kinds of CVD are today.
It is not sufficient, then, to mash a few colors together and call the result a “good” colormap. Instead, we want to ensure that the eye perceives a linearly varying intensity along the colormap, independent of hue. The colors themselves become an aesthetic element that is not necessary to understand the data being conveyed.
As a side note, perceptually uniform colormaps should also produce images that can be understood in grayscale, which is helpful if an image needs to used in print media without color.
Caution
This table uses unverified data from Wikipedia. It is, at best, a rough guideline for illustrative purposes only. This data is not mine, and you should not cite it from this page (or, indeed, from Wikipedia, as citations were not listed at time of writing). If you require quantitative data on color vision deficiency, please consult a medical journal.
Well-known perceptually uniform colormaps#
If you use matplotlib at all, you are undoubtedly familiar with its
perceptually uniform colormaps: matplotlib.cm.viridis,
matplotlib.cm.plasma, matplotlib.cm.inferno,
matplotlib.cm.magma, and matplotlib.cm.cividis.
And, to be clear, there’s nothing wrong with these colormaps! Personally,
I like magma, but the lightest color is sometimes too light for a particular
application; similarly, I like plasma, but the bright yellow looks bad on
a white background you might use in a journal article figure. plasma,
inferno, and magma also all use similar base colors, making them less
distinct than, say, viridis compared to plasma.
Another good source of pre-designed colormaps is the
cmasher library. It includes some maps that I
really like, such as cmasher.ember, cmasher.bubblegum,
and, of course cmasher.pride. Still, for a particular project,
perhaps I need or want something a bit different.
What sandman can do#
Simply put, sandman will take any colors you dream up and try to make them into a “nice” sequential, diverging, or cyclic colormap. Your imagination (and, perhaps, number of available CPU cores) is the limit! In the end, you will have a JSON file with lots of new colormaps, a score that indicates how close to or far from perceptually uniform that colormap is, and, just for fun, a pseudo-random, pronounceable name for that colormap based on the input parameters that generated it. sandman can also plot the “best” colormaps from one of these JSON blobs so you can choose the one(s) you like best.
What sandman cannot do#
While sandman assigns a numeric score to the colormaps it generates to give the user a rough idea of how well the optimization step worked, this is relative and somewhat subjective: You could easily choose a different optimization objective function and get a different result. So these numbers are useful but should not be treated as absolute quality measures.
Furthermore, sandman attempts to simulate human color perception
using the CAM02-UCS model from
colorspacious, but, since this is just a model,
it is probably not perfect, and it could be revised in the future. Treat it as
you would any other model.