SnapCalorie is an app where you can take a photo of any meal and get an accurate calorie count. Created by the co-founder of Google Lens, logging a meal with SnapCalorie takes seconds and on average is more accurate than a trained nutritionist.
If this works you guys will be overnight billionaires. I really struggle to see how, even with the best machine learning, you can discern from a photo the fat content of meat for example (eg minced beef is sold as 5%, 10%, 12% and 20% fat content where I live, they all look identical).
As a daily MyFitnessPal user, I'm looking forward to testing this. Could be a game changer.
Thanks @mickc79! I'm personally really motivated by how much potential there is to make people feel better and potentially live longer.
The main reason we're charging right now is because 1) we've found we get better feedback on what to build next if it's coming from people who invested some of their own money in the product and 2) the GPUs to run the algorithm are not cheap and so we can't scale infinitely with our current funding.
With regards to the % fat you are almost certainly right that the algorithm will struggle with this. It likely makes an educated guess with average values for beef. That being said, our research suggests that a majority of the error from people logging with existing solutions comes from estimating the portion size incorrectly (+/-50% on average) and not from the classification of the ingredient. Our hope is that by doing better on this we can actually be more accurate than current approaches with far less effort in logging.
We're also working on ways to add follow up questions where you can clarify non-visual aspects like this when the algorithm thinks it may be relevant.
I hope to have you as a user soon and would love to hear more on what you think when you try it.
@wade_norris Thanks for the reply. I appreciate it (and I have paid to sign up).
However still really stuck on things like sauces (which drastically affect calorie content).
For example if I'm making a chilli or a bolognese how can you ascertain via an image if i've used some sugar laden sauce from a jar, vs plain chopped tomatoes?
Anyway I will give it a bash for a month and see how it goes.
@mickc7 You're absolutely right that sometimes it will have to just guess average values for dishes if there aren't other visual cues. This is similar to how most people are logging right now in a lot of ways (e.g. I'm eating chilli so I'll just look up the average value in a database).
The big difference is that our algorithm is more accurate at the portion size estimation part.
In general we want to make logging easier and hope people leverage the average values over time as more of a signal than the exact values for each meal.
Can't wait to hear your feedback as you try it out! Feel free to email me any time with questions or feedback. wnorris@snapcalorie.com
Hi ProductHunt!
We’re building SnapCalorie for people across the world who want to live happier, healthier, and longer lives through improved diet and nutrition. There are few things in life that impact us as much as what we eat, but current tools for diet tracking take forever or leave you guessing when eating out.
With just a single photo, our AI algorithm will estimate the calories, carbs, fat, and protein of a meal. Our initial evaluations have shown that on average we are off by less than 60 calories per plate, while people logging without a scale are off by more than 125.
We’ve launched our app in a limited beta and the initial reaction has been great! We are excited to open this opportunity up to Product Hunt, and grow this community of users. In the long run we plan to add micronutrient tracking in addition to macros, suggest meal plans to fit your unique goals, and provide intelligent recommendations to improve your health.
If you want to help shape the direction of our product, we’re offering 30% off for motivated early adopters that want to join the beta now. Just use code PRODUCTHUNT on SnapCalorie.com
If you love the concept but want a more finalized product, sign up for our mailing list and we’ll notify you as soon as we’re out of beta.
@daniel_baum It can! This is where we focused most of our initial training data. We did this because we heard these types of complex mixed dishes when eating out are some of the hardest to log right now. Would love to hear your feedback if you give it a try.
After being a longtime DAU and paid subscriber of MyFitnessPal for a good decade, I've been using the SnapCalorie beta for the last week or so, and I'm super excited about where this is going! Congrats Wade and Scott!
This would be insane. I will use this side by side with My Fitness Pal to see how accurate it is. I guess for generic items it would be pretty accurate but when it comes to home cooked meals the difference from just the visual could be huge. You can recongnize mash potatoes or a salad but the amount of butter/dressing which is a significant percent of the calories would be really hard to recognize no?
@pablo_fatas The training data includes really challenging things like salads with dressing and mashed potatoes with oil and butter. The average caloric error numbers we quote (~60 calories of error on average) include those kinds of challenging images. The algorithm is likely guessing average values for non-visual things like oil and butter, but we've found this approach is actually far more accurate on average than users trying to eyeball portion size visually. We're also working on a way for users to assist the algorithm for non-visual elements if they know the answer.
Give it a try and let us know what you think!
Report
@wade_norris Thats incredible! Can’t wait to try it out
Report
Hi product hunt,
I am marketer
Congrats!, great idea, I would love to use it. Been waiting for this type of product.
Songbox
SnapCalorie
Songbox
SnapCalorie
SnapCalorie
Sleek Pay
SnapCalorie
Sleek Pay
Flow Club
SnapCalorie
SnapCalorie
SnapCalorie