Anyone using an app like Lose It to count calories knows how difficult the process actually is: unless you’re eating all pre-packaged meals, estimating the caloric count of what we’re consuming is cumbersome — and, according to studies, generally incorrect.
Speaking at the Rework Deep Learning Summit this week, Google researcher Kevin Murphy discussed a project Google is developing a photos-based algorithm to take the heavy lifting out of keeping a food journal. Called Im2Calories, through a sophisticated process of image recognition and algorithmic computer learning, Google is hoping to create a database of photogenic nutrition information — similar to the text-based ones found in apps like Lose It, for example.
According to Popular Science, Im2Calories will allow users to adjust nutritional information if the technology produces incorrect results. Murphy stated that if Im2Calories was even 30% accurate upon roll-out, “it’s enough that people will start using it, we’ll collect data, and it’ll get better over time”.
Obviously, no image-based calorie counter will ever be perfect — there’s even some who feel that our standard nutritional labels, using 19th-century calorie counting practices, aren’t entirely accurate themselves. However, an photographic database would certainly take some of the effort out of keeping a food diary; and, the likelihood of human-introduced bias as well.
Neither Google nor Kevin Murphy has set a timetable for Im2Calories to debut.