TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

Trained GPT to analyze body fat% within 5% error of DEXA scan

3 pointsby aakashboss9 months ago

1 comment

aakashboss9 months ago
Hey everyone, been working on this for a while and finally got it to a place where I&#x27;m confident in the accuracy. Started off by seeing how good GPT vision is at identifying real world foods (given the amount of variance in how the same piece of food can look) and then how well GPT could understand depth&#x2F;volume if given an image from an iPhone camera. These two things together make for pretty interesting calorie counting use cases. There&#x27;s a few calorie counting apps out there that try to do this but I don&#x27;t think any of them are specializing their models on this use case (at worst they&#x27;re scanning barcodes, at best they&#x27;re using out of the box GPT or Claude etc).<p>I decided to take it one step further to see if I could apply what success I saw with food identification and calorie estimation to see if I could replicate the results of a DEXA scan (for the uninitiated this is a 200 dollar body fat scan). Proceeded to try and fine tune GPT 4o on a dataset of dexa scans, photos, and demographic data (age, height, weight, etc) and this is the result! Would lvoe to hear what you think if you try it out!<p>PS: I&#x27;ve done little amrekting but it has gotten to dozens of paid users, if you have any thoughts on features to add to make it more valuable for the paid users please let me know. Didn&#x27;t think I would get this far (ramen profitable).