This compilation of computer vision is really interesting to watch if you imagine seeing it without the source footage behind the computer graphics. I really like the description Kyle McDonald gives in the comments of the aesthetic qualities of computer vision graphics and will add to it myself:
“color choices tend toward high contrast saturated primaries (easy colors to code), almost no text (it’s never descriptive, only enumerative), trails are used to show history, ellipses and rectangle/bounding boxes are used as placeholders for complex shapes…”
Robot readable world from Timo on Vimeo.
I did a series of tests of iPhoto’s facial recognition capabilities to determine what it can and cannot recognize as a face. It only tracks human faces, so I wanted to find the limit of what the software considers to be human. You can read details in the PDF, but essentially I found that it can see anything that looks like real human face, and not anything else. So chimpanzees that look a little less human than others do not not get recognized, and neither do cartoon characters. My favorite moment in the test was when it identified a friend of mine as possibly being in a photo that in fact contained no faces; it thought his face was found in some branches in front of the sky (image above).