How Bad Is Your Colormap? (2014) | Hacker News Reader

How Bad Is Your Colormap? (2014)

jakevdp.github.io

133 pointskristianp22 comments

I highly recommend this accompanying video, which takes jake's idea and goes into the details of how viridis replaced the ever-naughty Jet.

Colormap selection is surprisingly fascinating.

"A Better Default Colormap for Matplotlib (SciPy 2015) Nathaniel Smith and Stéfan van der Walt" https://www.youtube.com/watch?v=xAoljeRJ3lU

Here is also the link to the change as introduced in Matplotlib 3, to put it in reference: https://matplotlib.org/users/dflt_style_changes.html

Really enjoyed that. Thank you! Bit rushed though, need to read more!

This video is excellent. Thanks for sharing it.

Are there any articles out there that give advice on choosing/adjusting color maps for a particular situation?

I understand why some maps are awful, but have had a hard time finding a color map which I like perceptually for interpreting LIDAR forest canopy maps.

I've tried using viridis and a few others, but exceptionally tall trees get lost in the noise (not enough contrast), especially when using transparency to overlay canopy height on the LIDAR intensity data.

Is this just a case of not clamping the ends of the range properly?

I guess if your underlying data is not uniform, a uniform color map isn't going to do great?

Say, if you have a bimodal underlying distribution and you're interested in coloring according to quantiles, you want to have a similarly bimodally distributed color map.

The easiest way to achieve this is to compute the quantiles for your data set, and color by that (rather than height).

Have you seen this?

http://colorbrewer2.org/#type=sequential&scheme=BuGn&n=3

Not enough people know about rainbow colormaps being problematic. This is a pretty cool tool that scans bioRxiv for preprints that use Jet and then automatically sends the authors an email asking them to improve the colormap, e.g. to viridis, parula.

https://jetfighter.ecrlife.org

The piece makes some good points about problems with converting certain color schemes to grayscale. This is a not insignificant thing. It is quite common to print charts and graphs in grayscale, either because the only available printer is black and white or because color printing is substantially more expensive.

That’s not the point.

Everyone’s eyes/brain interpret colors by taking signals from the 3 types of cone cell detectors and combining them into 2 color-difference signals (red–green and blue–yellow) and 1 brightness signal. The latter of these has the best spatial/temporal resolution and is used for interpreting textures/fine details, edges of shapes, motion, depth, etc.

In a sense, everyone’s vision is primarily grayscale, with lower-resolution color information layered over the top.

And then you have to take into account that our sensitivity to some colors then others. Our cones are tuned to what can be approximated as three different bands of wavelength, which are centered in and sufficiently overlap around the green wavelength to result in our being most able to discern the color. Human eyes can discern at least twice as many distinct shades of green than they can red or blue in a uniform distribution, and color science has to take that into account to prevent such an image from appearing overwhelmingly green.

Thank you.

One area where I would like to see more perception research is in 2D colormaps for values that have both an angle and a magnitude (complex numbers). The "state of the art" seems to be throwing it all into max-saturation HSV but I know we can do a lot better than that.

I wrote a Processing library [1] that allows you to select perceptually inform colors in any range (but remember there are holes where normal sRGB monitors cannot fill, see the API docs [2]). If you haven’t heard of Processing, check it out! [3]; it is a generative graphics tool with an Arduino-like interface and IDE.

[1] https://github.com/neilpanchal/Chroma

[2] https://github.com/neilpanchal/Chroma/blob/master/README.md

[3] https://processing.org/

Matlab (somewhat) recently ditched jet as the default colourmap in favour of Parula.

https://blogs.mathworks.com/steve/2014/10/13/a-new-colormap-...

Since most folks (typically students) simply use the default this should help end the tyranny of jet.

I do still like jet for on screen visualisation, the extra contrast can be useful.

Isn’t the whole point that the contrast is factually incorrect? It looks pretty but is a lie.

Indeed, but when you’re looking at small variations in a gradient it’s a reasonable tool to use before you delve deeper.

I do CFD and have really begun experimenting with various colormaps for each project and it has made a significant business impact.

Each color map has a specific type of story it is best at telling and using program defaults is a really bad practice. For the type of phenomena you are trying to highlight and explain there is a color map for that problem and I think this is something that is understudied and under discussed for its impact

You might like this colourmap set then: https://matplotlib.org/cmocean/

About that luminance-correct grayscale: https://entropymine.com/imageworsener/grayscale/

Can we tag this with [2014]?

Reply on news.ycombinator.com