ABSTRACT
The objects are identified by their particular features, for example the four legs, black nose, fur and so on of dogs, and even though objects may have vastly different individual features, viewed in different poses and aspects, they must still be classified as dogs; for example, the genus dog includes the tiny short-hair Chihuahuas to gigantic shaggy Sheepdogs. The migration from top-down to bottom-up classification for computer vision began when Princeton's Li Fei-Fei realized that since infants cannot innately recognize objects and scenes, the algorithm should learn through experience just as humans do. Professor Li immediately saw the Mechanical Turk as a data-gathering tool that could scale, and accessing the crowdsourcing website, she and her team supervised some 50,000 participants in 167 different countries to collect, classify, label, and manually annotate nearly one hundred million images for her project, including 62,000 images of cats alone.
