This is interesting and unexpected:<p>> hexagonal arrays are as computationally efficient [...] made possible by recent mathematical advances in the representation of hexagonal arrays, in particular Array Set Addressing (ASA) which makes processing of hexagonal array images almost as conceptually easy as processing square array images. Indeed, operations such as computing a fast Fourier transform (FFT) and edge detection are in fact faster per pixel over hexagonal arrays than square ones.
TIL that hexagonal image formats and hexagonal array neural networks are a thing:<p><a href="https://scholar.google.com/scholar?hl=en&as_sdt=0%2C6&as_vis=1&q=hexagonal+image+format&btnG=" rel="nofollow">https://scholar.google.com/scholar?hl=en&as_sdt=0%2C6&as_vis...</a><p><a href="https://scholar.google.com/scholar?q=hexagonal+neural+network&hl=en&as_sdt=0&as_vis=1&oi=scholart" rel="nofollow">https://scholar.google.com/scholar?q=hexagonal+neural+networ...</a><p>Google even has a recent related patent:<p><a href="https://patents.google.com/patent/US11200458B1/en" rel="nofollow">https://patents.google.com/patent/US11200458B1/en</a>
Main reasons for normal pixel arrays is that UI elements are built predominantly from vertical and horizontal lines. And that is enough to bury the idea of hex arrays until you have so many pixels that it doesn't matter anymore at which point there is no more advantages from hex arrays.