I knew “fluorescents are weird” well enough to check again after we went outside in the sun, but I was not prepared for how much they would change.

In college I (the lighting designer) went fabric shopping with the costume designer. I brought my swatchbook of lighting color gels.

We went to this big warehouse in an old factory building, acres of cloth under fluorescent lights. I looked at how the samples would look in the colors of my lights.

Same here…

when I say, “my field,” I mean perceptual neuroscience, biophysics, engineering…

but tech is where I first learned that color metamerism existed and just how bananas it can be.

My field broadly if not narrowly:

cloth, paper, paint, and the like can look radically different colors under different lights even if the lights are the same color.

Whoops, Carthago is feminine, that should be Carthago deleta est.

I think this could have worked equally well as a post about sea-level rise.

Seriously, what map tool lets you just delete Tunisia and literally flood-fill it with blue?

https://www.nytimes.com/2023/07/29/world/africa/africa-coups-niger.html?smid=nytcore-ios-share&referringSource=articleShare

It does! It comes from the French word because it means exiting your fortress to make a quick strike at the folks besieging it. Or so Wikipedia would have you believe: https://en.m.wikipedia.org/wiki/Sortie

We used a partial version of the balancing scheme from the PANN and PSLA/AST papers. Full balancing (PANN, PSLA/AST) hurts mAP on a held-out set, but partial balancing gives a small boost over baseline:

The training speedup is probably the most surprising finding. Other papers have suggested this, but the magnitude of the speedup is quite large: 6x with full balancing (but poor generalization) 2x with partial balancing at optimal generalization performance.