'computational photography' is interesting because it requires a different way of thinking about what an image is than what we were used to. As if going throug

'computational photography' is interesting because it requires a different way of thinking about what an image is than what we were used to. As if going through a certain shaped lens is the only way to make "real" photos. My own thinking on the subject got severely reshaped when I took it upon myself to write some alternative lens models for Chunky[1] (the Minecraft scene path-tracer) and later my own path-tracer[2]. Once I realized that every pixel on the screen could be mapped to any virtual light-sensor position and direction, suddenly projections were arbitrary. I could define a lens that wrapped around the subject, pointing inwards if I wanted, and that was no less 'real' than a pinhole or fisheye lens. Of course modern phones take things a bit further than just straightforward transformations. Guessing what's between the pixels or selectively blending multiple images together maybe crosses my subjective threshold between what I think of as a 'real image' and a 'made-up' one. Especially as the AI gets smarter. Because I can no longer understand exactly what happened. [1] https://chunky.llbit.se/ https://chunky.llbit.se/ [2] https://github.com/TOGoS/SolidTree/ https://github.com/TOGoS/SolidTree/