Hacker News Reader

Comment by actuator | Hacker News Reader

The fat measurement bit seems interesting. It seems entirely dependent on CV.[1] It does acknowledge that people gain fat differently but I am very apprehensive of this feature. Definition of muscles unless you go sub 10% BF looks very different among people. Also, clicking full body pictures and uploading it in a third party service even though this is Amazon feels very violating for some reason.

I think traditional home/gym scales use electrical impedance to measure this like this Inbody one. [2] They also claim to have been validated by studies. I would rather want these devices to become more portable and affordable.

[1] https://blog.aboutamazon.com/devices/a-better-measure-of-hea...

[2] https://inbodyusa.com/products/inbody770/

View on HN

I'm very intrigued by this because I have a bioelectrical impedance scale, one of the top of the market. I've generally found it to be extremely accurate. Amazon is saying this beats even those, which if it can do from just photos (with our without depth map?), that's impressive.

Yeah, the top ones generally are fairly accurate even from my experience, though the full body ones fare better than just the foot ones. Even if they are off by ~5%, it still can be good to track changes.

As an aside, for anyone who wants to track changes, I made an iOS shortcut which takes weight, waist circumference and neck measure and estimates body fat and muscle mass using the navy formula. It then logs weight, waist, body fat, muscle fat and bmi to apple health.

It is surely inaccurate, but directionally accurate. I only check weight and waist each day. Neck circumference changes slowly.

This uses imperial measures and is for men. You need to set the apple health data to imperial for weight and waist.

For women, you also need hip measures, and the calculations would differ a bit. If you’re interested in building this reach out and I can help modify this one.

Shortcut runs fast, and the two measures are quite quick to do daily.

https://www.icloud.com/shortcuts/fb69f71fc1c54cb09191b3610cf...

Hey, thanks for telling about this. This sounds interesting, I hope someone who uses iOS gives it a try and can give feedback.

I use it every day, so it definitely works! Dunno how to replicate on android, but maybe possible? Not familiar with the scripting/automation tools there. Or the health data integrations. But you could always log to a spreadsheet.

I have this scale[1]. It's extremely flawed in their body fat calculation. It simply displays 10% of your weight as your body fat percentage :-)

[1] - https://www.amazon.com/gp/product/B07MBPF2HW

Anything that goes from just your feet is gonna be pretty inaccurate just due to the laws of physics. You really want one that also has hand paddles.

The state of the art [1] for body fat estimation using depth sensing is actually quite good (R2 value of 0.78).

The hard part will be accurately extracting the anthropometric features from the front/side/back images that Halo captures. There's a chance they would initially launch an iPhone only app and leverage the face scanning depth sensor.

[1] https://www.nature.com/articles/s41430-020-0603-x

Electrical impedance has been shown to be absolutely trash in accurately predicting bf%. I've used both Garmin's and Fitbit's smart scales, which are supposed to be top-of-the-line, and compared to DXA... the numbers aren't even close. Most people report similar experiences.

That's not something you can have at your home though, no? The DEXA/DXA scans machine I know of are something you can't keep at the comfort of your home.

In my personal experience I have found impedance ones to be fairly accurate(~10%, BF range 10-16) and for most people the accuracy of full body ones is good to have a general idea about their body. Probably when you reach sub 10 levels, accuracy matters more.

Btw, here is the link to studies they give: https://inbodyusa.com/studies/

I've also found about a 10-point discrepancy. There's a huge difference between 15% and 25% bf.. one is approaching athletic while the other is obese.

DXA is definitely less convenient, but there's no real reason to test your bf% daily (just like weighing yourself daily isn't productive, you need to look at average changes over time). Considering the amount of time it takes to significantly change your bf%, twice a year testing would be plenty.. and would probably cost you the same amount as this subscription service.

Maybe weighing daily IS productive, but looking at the result daily is not :)

Hydration and other factors affect accuracy of impedance body fat estimation. So consistency is key, like measuring at the same time in the same conditions (for myself, after waking and before any consumptions of food or water) daily. Still, absolute measurement will not be as accurate as DXA, hydrostatic, skin calipers, etc.

However, for relative changes over time when consistently measured, impedance works just fine (trending upward or downward).

I'm very skeptical of their claim that the photo measurements are more accurate than other means, but perhaps they've done massive validation of their methods and have achieved adequate reliability and validity.

>for relative changes over time when consistently measured, impedance works just fine

Agreed, but for change over time it's just as accurate as skin calipers... which don't require an investment or subscription service.

While skin calipers are most affordable regarding equipment, the reliability is poor unless the person measuring is very consistent (well-trained and/or experienced for consistent pinch/folds). Calipers also will be inconvenient and possibly unaffordable for many because access to the person measuring is necessary whenever you want an updated measurement.

In my view, impedance evens out with calipers because of the convenience of impedance measurements over time. I just stand on the scale for a few seconds every morning, and I have stats synced that give me a good enough idea of my status (bonus for the added understanding of my hydration which calipers won't do). There's no cost or subscription for my scale aside from initial purchase (Withings aka Nokia Body Cardio), and it is regularly available for as low as $99 on sale. So after a couple years and hundreds of measurements the cost is negligible given the benefit.

If I had someone regularly accessible to do caliper measurements, I would have it done to compare regularly but still continue with impedance because it's so convenient to do and track.

It's still useless if the data is garbage, which it really is. I had DXA scans done every half year for the last few years, and I've owned a few smart scales in that time (first Fitbit, then Garmin). DXA has shown me consistently dropping bf% (17.8%, 15.0%, 13.1%, 11.2%) while the scales (taking monthly averages) are all over the place (10.1%, 11.1%, 11.6%, 10.0% for the same months the scans were taken in). Impedance can show you a ballpark (and we're talking a wide, 10%ish ballpark) of where you are, but honestly you're probably better off just looking in the mirror and comparing to images online.

I only mentioned calipers because they'd at least show you trends, which is where impedance seems to fail. Impedance is simply a poor way to measure bf%. Hydration might be a different story though.

I'd recommend you do yourself a favor and just spend the $100 annually to see where you're at.

I've had better experience with the Body Cardio scale. Though I haven't validated against other measures, I believe the body fat percentage is accurate enough based on my subjective opinion (it has not been generous to me like another poster's experience). Research has determined that impedance can be +/- 4% from actual in optimal conditions and my scale seems to fall close enough within that range.

Regardless, I use it for all the variables tracked. I look at body fat percentage but body weight is ultimately the value of concern. I don't agree $99 a year is a good value just for bf%, because I do agree that visual impression means more than a specific number.

The subscription cost for Amazon's service may be a great value (privacy concerns aside) if their photo algorithm is within a couple points of DXA or hydrostatic.

>The subscription cost for Amazon's service may be a great value if their photo algorithm is within a couple points of DXA

If it's accurate enough, I definitely agree. I'm looking forward to testing it out once it's available; since I'm one of those data nerds with actual ongoing DXA results, I'll be able to compare the two accurately. Although, pessimistically, I think if they had proven results within a few points they'd be a lot more quantitative in their marketing rather than the ambiguous language in the posts referenced above.

I can confirm that the Withings Body+ scale body fat measurement is also total garbage. For me it reads several percent lower than a DXA scan, basically just a vanity number for people who want to believe they aren't fat. The readings also fluctuate a lot based on how well hydrated I am.

Reply on news.ycombinator.com