People are bad at reporting what they eat. That's a problem for dietary research • Hacker News Board

Re: People are bad at reporting what they eat. That's a problem for dietary research

#181

Post by wnorris510 » Tue, Jan 21, 2025, 3:29 PM UTC

mrgaro wrote: post #176

wnorris510 wrote: post #130

I've done research in this space for many years at Google AI and now at SnapCalorie. The thing I find interesting is how confident people are in their ability to estimate portion size visually, and in truth how wrong they all are. We published in CVPR (top peer reviewed academic conference for computer vision) and people are on average off by 53% and even trained professionals are still off by 40%. Basically if you w…

I downloaded SnapCalorie to try it out on Android. I went all the way through the sign-up phase, only to discovery that I would need to activate subscription in order to have the 7-day trial. Ended up uninstalling the app :(

We're an early stage startup and the models are expensive, we're trying to get the price as low as possible, but yes we need to charge to cover costs right now. Sorry about that!

You might get a yearly discount offer that is less than $2/month if you get lucky (A/B test split). But that's less than the cost of running the model for people, so hopefully others will consider paying full price.

Re: People are bad at reporting what they eat. That's a problem for dietary research

#182

Post by zahlman » Tue, Jan 21, 2025, 3:30 PM UTC

wnorris510 wrote: post #130

I've done research in this space for many years at Google AI and now at SnapCalorie. The thing I find interesting is how confident people are in their ability to estimate portion size visually, and in truth how wrong they all are. We published in CVPR (top peer reviewed academic conference for computer vision) and people are on average off by 53% and even trained professionals are still off by 40%. Basically if you w…

>people are on average off by 53% and even trained professionals are still off by 40%. Basically if you want to have a higher level of accuracy you need to use a food scale or something that measure the volume of food, people just can't estimate portion sizes visually.

I can typically estimate them accurately without direct measurement, and with feedback that will tend to make errors cancel out over time. My trick is to note package weights, and divide containers into N equal portions. That is: I decide a target portion size first, and then portion it out.

If the task is "measure out an ounce of butter" I realistically won't be 40% off - because I can very accurately divide a rectangular solid in half repeatedly, and the butter comes in a one-pound package. Similarly, I have a pretty good idea how much grilled chicken is on my plate, because I know how much raw chicken I cooked, because I made a whole piece from a pack of N roughly-equal pieces weighing X (values which I noted when I bought it).

Re: People are bad at reporting what they eat. That's a problem for dietary research

#183

Post by _dark_matter_ » Tue, Jan 21, 2025, 3:30 PM UTC

FredPret wrote: post #101

_dark_matter_ wrote: post #79

Earlier quoted context omitted.

There are ALL kinds of things we can't run experiments on. Climate, society, evolution, tech development, surgery, and more. We don't throw our hands in the air and say "science here is impossible!". Instead we roll up our sleeves and develop more and better causal inference models that improve over time.

OK, name such an approach in nutrition that doesn't already fall under regular biology. The reason we can do science on some things without doing experiments is that there's lots of hard, unambiguous data about relatively much simpler systems. Getting good data on the extremely complex thing that is homo sapiens is just not feasible, unless you're studying specific chemical reactions in the gut, in which case it'll t…

It sounds like your describing models. Yes we have small models. Yes we have big models. We have rat models, we have matching models, we have twin models. We have lots of models I don't know about.

I'm not sure what you're getting at. These models get better, our understanding is improved, and we will slowly uncover more truths. We already have so much more knowledge about nutrition than we did 100 years ago.

Re: People are bad at reporting what they eat. That's a problem for dietary research

#184

Post by mtntreks » Tue, Jan 21, 2025, 3:32 PM UTC

m_ke wrote: post #139

wnorris510 wrote: post #130

I've done research in this space for many years at Google AI and now at SnapCalorie. The thing I find interesting is how confident people are in their ability to estimate portion size visually, and in truth how wrong they all are. We published in CVPR (top peer reviewed academic conference for computer vision) and people are on average off by 53% and even trained professionals are still off by 40%. Basically if you w…

At Bitesnap we were surprised at how much interest there was from researchers to use our app for diet tracking. It turns out giving people a piece of paper to write “grilled cheese sandwich for lunch” is not a scalable and reliable way to collect research quality data. We even worked with USDA on putting together a food logging dataset: https://agdatacommons.nal.usda.gov/articles/dataset/SNAPMe_A...

This doesn't surprise me.

Just trying to keep track of calories for myself stupid things like supersized slices of bread becoming common in stores can really throw off my expected calorie counts.

It seems like this can completely throw off any attempt at figuring out nutrition from an app or research perspective.

Re: People are bad at reporting what they eat. That's a problem for dietary research

#185

Post by bluGill » Tue, Jan 21, 2025, 3:33 PM UTC

everdrive wrote: post #156

mmooss wrote: post #143

Earlier quoted context omitted.

> For whatever your metabolism, and whatever your category of calories, fewer calories will still produce weight loss. I thought that wasn't true, that the human body stores and burns calories at varying rates based on many signals, and that our bodies or some bodies effectively conserve weight or caloric stores at a certain level.

The body can compensate at the margins. Eat 5 fewer calories per day and you will see zero change. Eat 500 fewer calories per day, consistently every day, and you will absolutely see changes. (I'm not actually suggesting that it would be _healthy_ or advisable to drop your diet by 500 calories -- just pointing out that the body cannot compensate indefinitely.)

The compensation is often lower energy levels. Your body compensates by keeping you from doing as much.

Re: People are bad at reporting what they eat. That's a problem for dietary research

#186

Post by wnorris510 » Tue, Jan 21, 2025, 3:33 PM UTC

zahlman wrote: post #182

wnorris510 wrote: post #130

I've done research in this space for many years at Google AI and now at SnapCalorie. The thing I find interesting is how confident people are in their ability to estimate portion size visually, and in truth how wrong they all are. We published in CVPR (top peer reviewed academic conference for computer vision) and people are on average off by 53% and even trained professionals are still off by 40%. Basically if you w…

>people are on average off by 53% and even trained professionals are still off by 40%. Basically if you want to have a higher level of accuracy you need to use a food scale or something that measure the volume of food, people just can't estimate portion sizes visually. I can typically estimate them accurately without direct measurement, and with feedback that will tend to make errors cancel out over time. My trick is…

Yeah, dividing out a known portion size is a good hack that will probably help with accuracy. In our research most people's calories and error came from eating out where they didn't have these hints, but this is a good trick if you mostly cook for yourself!

Re: People are bad at reporting what they eat. That's a problem for dietary research

#187

Post by zahlman » Tue, Jan 21, 2025, 3:35 PM UTC

zeroonetwothree wrote: post #155

taeric wrote: post #153

Earlier quoted context omitted.

This goes a long way to further convince me that it is portion sizes in the US. Having traveled, it is quite absurd to see the difference in standard order sizes. Even for zero calorie things like water and unsweetened teas/coffees. You just get smaller cups. I'm not even sure you can get a US large in Japan for most drinks?

I don’t see why it would be bad to get more water to drink

There's actually something of a stereotype that Japanese places will give you unreasonably small portions of water with meals. (Dogen plays off this in some of his videos.)

But then, I think it's only been Americans I heard this from, so.

Re: People are bad at reporting what they eat. That's a problem for dietary research

#188

Post by ploynog » Tue, Jan 21, 2025, 3:36 PM UTC

wnorris510 wrote: post #181

mrgaro wrote: post #176

Earlier quoted context omitted.

I downloaded SnapCalorie to try it out on Android. I went all the way through the sign-up phase, only to discovery that I would need to activate subscription in order to have the 7-day trial. Ended up uninstalling the app :(

We're an early stage startup and the models are expensive, we're trying to get the price as low as possible, but yes we need to charge to cover costs right now. Sorry about that! You might get a yearly discount offer that is less than $2/month if you get lucky (A/B test split). But that's less than the cost of running the model for people, so hopefully others will consider paying full price.

I think the complaint is that you only get told that you require a paid plan AFTER signing up. At least on a brief look on the Play Store page and your website, it does not immediately mention it prominently.

That seems like a very dark pattern and is, honestly, pretty scummy.

Re: People are bad at reporting what they eat. That's a problem for dietary research

#189

Post by Spivak » Tue, Jan 21, 2025, 3:36 PM UTC

everdrive wrote: post #81

A lot of people seem to have a purely emotional relationship with resources which logic doesn't seem to be able to penetrate. Food and finances seem similar here. For years I tried to get my wife to stick to a grocery budget. That is, we have $n per week for all groceries. She'd blow badly over the limit every time. "But we needed [food]" or "These were toiletries, so they don't _count_ as groceries." Ultimately we n…

Oh hey, I'm the wife in this story. Having a fixed $/month budget for "things you buy at a grocery store" was doomed from the beginning. All the stuff in your house/pantry are on all kinds of weird replacement cycles that vary with usage and changes in habits. A monthly cadence also makes you sub-optimally plan around price movements.

An attainable goal is to reduce the average amount of monthly grocery spend and you do it by deciding, in advance, things you're no longer going to stock in the house, items you'll replace for cheaper options, or items you'll stock from wholesale clubs.

It's hard to bring the budget for gas down without people driving less. Your wife being the one tasked with filling up the tank is the messenger. It could be an emotional reaction as you describe but I would at least entertain the idea that her "bending the rules" is her way of trying to make an impossible ask doable. Whether she is consciously thinking about it or not, I bet the stuff that "doesn't count" aren't replaced every month and have spikey cost patterns.

Re: People are bad at reporting what they eat. That's a problem for dietary research

#190

Post by paul7986 » Tue, Jan 21, 2025, 3:36 PM UTC

Have used chatGPT for about a year to count my daily calorie in-take .

Since I eat out daily at fairly healthy places (Cava, Panera, chic fil a only grilled nuggets & fruit cups, Jersey Mikes number 7 mini, noodles & company, MOD pizza) GPT knows their menu & each items calories. Upon getting the food I just tell GPT what I'm about to eat each time and it counts & retains and calculates through the day.

In doing so as adult male (late 40s) 5'10 175 my body has gotten used to eating 1500 to 2k calories a day. Do weigh myself daily to ensure I'm not gaining as I do have a cheat day once a week.

I understand the sodium content is higher then if I cooked at home but I'm focused on maintaining a fit look & counting calories along with a few weekly gym visits I think keeps me as I seek.