Behold a database of everything
Thingd's computer servers constantly crawl tens of thousands of product-rich websites, such as J. Crew (JCG) and Best Buy (BBY). When the program comes across an unfamiliar object—whether it be a shoe or a stapler or a truck—it collects an image, extracts whatever related facts it can find, and then adds the item to the database. So far, the crawler has identified more than 100 million items, and it aims for as much specificity as possible—a white 1999 Ford Explorer, for instance, rather than just an SUV. The Fancy is a crowdsourced supplement. When people upload photos to The Fancy, says Einhorn, "they're just training our database." The database becomes smarter over time—once it's seen several pictures of, say, a chandelier, it starts to recognize the same chandelier in other pictures, even when no identifying information is available.
No comments:
Post a Comment