New App Counts Calories Through Photos of Food

[xxchef]

"Want to know how many calories you are going to consume when you reach for that cheese burger or slice of pie? Take a picture of it using your iPhone. A new app called MealSnap allows users to photograph food and get the calorie count."

So claims an article today in Discovery News: http://news.discovery.com/tech/new-app-counts-calories-through-photos-food-110413.html

Anybody seen/tried this out?

[highchef]

Thank you for the link, there was also a further link on a color changing film that darkens as food spoils. This would be lovely used on that 25 bucks a pound lump crab meat.

[Mjx]

I'm sceptical, although it sounds like an idea with potential.

I'm not clear on how this app could determine the actual amount of fat, sugar, or various fillers present (e.g. steamed kale versus kale sauteed in bacon fat, a desser made with sugar, or the identical one made with an artificial sweetener, an actual breaded cutlet versus one that is a composite of animal byproduct and TVP), factors which can make a tremendous difference in calorie count.

[kathryn]

They appear to be using Amazon's Mechanical Turk service to identify foods, and a lot of user reviews in the App Store are complaining about inaccuracies and lag time to get calorie counts back.

[xxchef]

I'm sceptical, although it sounds like an idea with potential.

I'm not clear on how this app could determine the actual amount of fat, sugar, or various fillers present (e.g. steamed kale versus kale sauteed in bacon fat, a desser made with sugar, or the identical one made with an artificial sweetener, an actual breaded cutlet versus one that is a composite of animal byproduct and TVP), factors which can make a tremendous difference in calorie count.

And even if it can correctly identify a food correctly, how can it tell how much of the food is really there without something to give it scale? Is that a cup of soup or a bowl? Is that brownie 2x2" and 1" thick or 4x4" and 2" thick?

The app does give a range of calories (sometimes a large range) but it would be interesting to know what parameters define the ranges - is it a question of portion size or variation in possible ingredients involved, etc?

Edited April 14, 2011 by xxchef (log)

[HungryC]

I had the same thoughts about inaccuracies: a burger with a no-sugar-added sourdough bun vs. a brioche bun, a bowl of whole-milk yogurt & fruit vs. fat free yogurt (ditto for ice cream), deep-fat french fries vs baked, and so on.

[abadoozy]

I downloaded it and it's clever, but not THAT clever. Calorie-wise, it's a crapshoot, even when it correctly identifies an item. For example, I took a picture of 9 Jelly Belly jelly beans, an easily calculable calorie count of about 35 cals. It correctly identified the food as Jelly Belly jelly beans, but reported the calorie count as something like 120-140 cals. Totally off in an easily identified picture.

So if it can't do things like that correctly, how the heck can it figure out more complex food? As others have pointed out, there's only so much info you can get from looking at an item. Whole-milk versus fat-free, etc, can't be identified by sight. And even if they could, I don't see how it can do portion sizes accurately.

I just don't think it's possible to do what the app says it can do. It's hard enough for humans to figure out accurate nutrition by just looking at a plate (trust me - I do carb counting as I'm a Type 1 diabetic and my health depends on it), there's no way a computer can do it better with a simple picture, at least not with today's technology.

[xxchef]

I just don't think it's possible to do what the app says it can do. It's hard enough for humans to figure out accurate nutrition by just looking at a plate (trust me - I do carb counting as I'm a Type 1 diabetic and my health depends on it), there's no way a computer can do it better with a simple picture, at least not with today's technology.

I can't see it working either. You can probably take several picture of the same mixed plate of food and have it come up with a different result every time.

Another cool idea waiting for technology to catch up.

So, it's a $2.99 rip off (or maybe it's a cheap toy for doing parlor tricks at the dinner table)? That it actually recognized the BRAND of your jelly beans is pretty amazing!

[kathryn]

It's not computers doing the food identification, it's humans, being paid very low amounts for the task, through Amazon's Mechanical Turk service. A few pennies per photo.

Looks like they’re using Mechanical Turk to identify the foods at anywhere from $0.02-$0.05 per picture and then using the data returned from Mechanical Turk to search for calorie information in their already well established database of food.