Can AI really estimate calories from food photos with 90% accuracy?

I found this claim about AI estimating calories from food images with 90% accuracy, but when I checked the website, there were no citations to back up the claim. It seems a bit hard to believe, but maybe I’m missing something. For people who care a lot about tracking their food intake, like those on keto, even small differences in carbs or calories matter. Does this really seem accurate to anyone else?

Welcome to this forum

Guidelines for Technical Discussions


Here are some helpful tips for posting:

  • Posts should have more than 100 characters. The more detail, the better.
  • Always include a direct link to the technical or research information.
  • Share your connection to the information - did you do the research or just find it useful?
  • Provide context and dialogue about the technical content.
  • If there are code repositories or models, be sure to include them.

Thanks for posting! Let the mods know if you have any questions or need help.

I really doubt that’s true, especially without any sources cited. Two servings that look almost the same could still have a big difference in calories depending on what’s actually in them. If this was proven true, I’m sure they would provide something more concrete to back it up.

@Whit
I bet they cherry-picked easy cases where the calorie count is pretty obvious. But it’s an interesting concept. I’d love to see something like this for making ethical purchasing decisions too. Like scanning a product and finding out if it’s made with something controversial—‘This item was probably made with Xinjiang cotton’ or something.

@Whit
It could work if you have a clear reference point for the size of the dish (to estimate the serving size) and a decent image of what the food is. After that, it’s just math, which can be calculated with a tool. You could even compare the image with recipe sites to get a more accurate estimate if you wanted to get fancy.