11th on the list. To be fair, the cake was a dark color and from a quick glance I might’ve guessed it was chocolate.
One night for dinner I ate tri-colored tortellini pasta with cherry tomatoes and parsley pesto, plus a kale salad and a glass of milk. The app’s top suggestions were guacamole (maybe what it thought the pesto was), salad, and ravioli—not entirely accurate, but fairly close. Notably, the app didn’t register that there was a glass of milk sitting next to the plate. It was easy to quickly search the app’s database and fill in the missing and misidentified foods, though.
On the other hand, the app struggled to recognize a homemade Ethiopian-style stew with red lentils, yams, cabbage, and tomatoes, so I had to type that dish in manually and hazard a wild guess about the calorie count. But I’ll cut the software some slack because there are a lot of similarly colored ingredients in that meal. It’s also a relatively obscure recipe that I’m guessing not many U.S. users have logged in the Lose It food database.
Overall, a mixed bag is what one might expect from an early version of a software product, especially one using image recognition—a technology that has come a long way in recent years, but is far from perfect.
Teague compares the beta version of Snap It with semi-autonomous car technologies currently on the market, like Tesla’s Autopilot. “It’s this assistive technology,” Teague says of Snap It. “We can’t make it quite as easy as taking a picture and then everything’s done. But we’ve done a lot of the heavy lifting for the user.”
He also notes that in the first version of the product, “we intentionally focused the recognition on the most popular categories of foods to improve its efficacy.”
“Ethiopian stew may not be recognized by the deep learning framework today,” Teague says, “but as our millions of users access and use Snap It, we expect it to become increasingly accurate and specific, evolving to a place where it can define the precise foods, portion sizes, and nutritional information, all from a single photo.”
That will take some work. But the bigger question is whether Snap It can consistently make it faster for users to log their meals—or, at least, make people think tracking meals with it is easier. Users may grumble about how long the process normally takes, but Teague’s company has actually timed it with a stopwatch and found that it typically requires less than five minutes to manually log everything a person eats in a day.
“What’s getting to people is the cognitive burden—I’ve got to remember to do it, it’s annoying,” Teague says. Snap It is “likely to improve both” the speed of logging meals and the perceived hassle, “but I expect it to have even more meaningful impact on the mental burden.”
