• Great breakdown. I think the "ahead-of-time" stuff﻿ is a big part of what makes pointerpointer amazing...there seem to be a lot of photos involved - I wondered if they used some kind of machine learning algorithm to figure out where the fingers in the photos are pointing.

• Voronoi is an incredibly simple concept. Out of a field of points, at any given coordinate, find whatever point is closest to you, and you're part of that point's region. So if you have a field of colored points, and then iterated over each pixel in the screen and made that pixel's color the color﻿ of its closest point, then you'd see a voronoi diagram!

• im quite curious what are the chrome extension PI uses. who can tell by icons?﻿

• Fuck, I'd use a standard grid, and just select with a small equation....

boxNumberCursorIsIn=int(x/10)+­int(y/10*10);

Alternatively, find the centre of the boxes, and then find the closest point to the mouse.

Using what they have, seems a bit﻿ overkill!

• They might have, but i think it's faster to do manually. Search flickr for photos tagged with finger pointing.﻿ Once you have 400 or so, for each determine one or several points where the finger is pointing to (like make a slide show and record clicks). Once you know what point refers to what image, voronoify! It's a great idea

• I checked out the voronoi diagram behind the scenes and then tried to trick it by moving my pointer around some of the big "cells" in the diagram where the author(s) hadn't found many pictures.

The script surprised me by still being super accurate!﻿ I guess it also recenters the image on the point in that cell as well. Nice extra touch.

• Read the first comment, it explains why you would use a voronoi grid to get better accuracy﻿

• I don't understand why the Voronoi grid is necessary if all the source images have﻿ already been mapped to a grid point?

• Nice.﻿ I guess you can do this with just css using '~' selector.

• Well, the diagram ensures that the whole space is covered by at least one image,﻿ that perhaps would be more difficult to accomplish using just collision testing and a set of xy coords.

• more﻿ than than doing mouse x,y collision testing against a rectangular cell? Can't imagine how... but sure I'm missing something, thanks! :)