Don’t try this when you’re a dinner guest. On the other hand, if you are in a restaurant with friends, go ahead. Have fun with PlateMate. Whip out your smartphone. Snap a photo. Then ask for crowdsourcing advice on how many calories, fat, carbohydrates and protein you are about to consume. The answer will come back in about an hour, while you’re still suffering heartburn from the Double Down Sandwich Combo you just ate.
Four Harvard students — Jon Noronha, Eric Hysen, Haoqi Zhang and Krzysztof Gajos — created PlateMate after being inspired by the Remote Food Photography Method. RFPM is a means of gaining more accurate information about food intake than asking people to keep a food log. Instead, they use smartphones to photograph what they are about to eat.
When they have had enough, they photograph whatever is left on the plate. Registered dietitians look at the photographs and analyze the amount of food eaten and wasted. The data is fed into a computer application that automatically calculates grams, calories, nutrients, waste and intake based on a USDA database.
The students wanted to make a similar system available to anyone via a smartphone app, so they tinkered with the technology, created a method, and tested it on two groups of students. PlateMate yielded results only slightly less accurate than trained dietitians. Using Amazon Mechanical Turk, the crowd (or Turkers) looked over the photos, identified foods and quantities, and fed them into software that did the estimating.
They did run into some challenges. PlateMate kept identifying a low-fat latte as coffee with cream and low-fat salad dressing as the full-fat version. However, trained dietitians overestimated caloric intake by 5.5%, PlateMate by 7.4%. Either way, the fudge factor is in the right direction since presumably weight loss would be the main reason for using the application.
Meal Snap already offers crowd-sourced calorie counting, though has some accuracy challenges to work out. The engineering students, now graduates working for Microsoft and Google, figure PlateMate can be optimized to give faster and more accurate results. Expect the app to be commercially available before long.
Related Care2 Stories
Photo from beglib via morgueFile.com