Wade Norris wanted to create a life-changing project at Google. He co-founded Google Lens, a computer vision app that displays object information. It didn’t satisfy.
Norris and aerospace systems engineer Scott Baron founded SnapCalorie, a health-focused startup, several years ago. AI-powered SnapCalorie estimates a meal’s calories and macronutrients from a smartphone photo.
Accel, Index Ventures, Eric Roza, and Y Combinator invested $2 million in SnapCalorie this month. Pre-seed investors gave the company $125,000.
Norris said humans are bad at visualizing food portions. SnapCalorie uses new technologies and algorithms to improve the status quo.
SnapCalorie isn’t the first computer-vision-based calorie counter. Foodadvisor, Bite, Calorie Mama, and Lose It AI has tried and failed. Norris says SnapCalorie’s depth sensors for measuring portion size and human reviewers for “an added layer of quality” set it apart.
Norris says the team reduces caloric errors by 20% on average. “Other apps can use AI to do photo-based meal tracking, but none of them help with portion size estimation—the most important part to reduce error.”
Photo-driven calorie estimation tools are skepticized in the health industry for good reason. Calorie Mama, one of the most popular AI-based calorie counters, was only 63% accurate in a 2020 study.
How has SnapCalorie improved? Norris points to a company algorithm that can estimate food calories better than humans. SnapCalorie’s algorithm recognizes food types and measures portion sizes to estimate caloric content.
SnapCalorie’s food journal or Apple Health can store results.
Norris says SnapCalorie’s unique training data set of 5,000 meals—soups, burritos, oils, “mystery sauces,” and more—explains the algorithm’s strong performance.
“We weighed out every single ingredient on a scale and made sure these had all of the diverse and challenging conditions you’d see in the real world,” Norris said. “The traditional pipeline for training an AI model is to download public web images, have people label them, and train the model to predict those labels. Food can’t be labeled after the fact because people are bad at visualizing portion sizes.
Since SnapCalorie collected most of its initial training data in the U.S., Norris admits the algorithm may favor American food. He says SnapCalorie is adding other cultural cuisines to its training data using user photos and internal data.
No app can accurately calculate the calories in a meal, regardless of the algorithm. After all, apps don’t account for cooking methods or food breakdown time.
Norris doesn’t claim SnapCalorie’s calorie-estimating tools are 100% accurate, suggesting they’re just one part of the nutrition puzzle. He also highlighted SnapCalorie’s ChatGPT-powered chatbot, which makes meal suggestions based on a user’s preferences, past preferences, and nutritional values.
“People are becoming more aware of what they eat. Norris said that processed foods’ health risks are becoming more apparent every day. “We’ve heard that our users really like SnapCalorie especially when eating out, as many restaurants don’t post nutrition values, and they would otherwise have no way of logging the meal.”
As for popularity, SnapCalorie is on track to reach 1,000 new users this month. Norris called the burn rate “very conservative” even though the company is expanding rather than monetizing.
“Our incredible organic growth rate seems to indicate that our value proposition resonating well with consumers—people try it, love it, and recommend it to their friends and family,” he said.