Hacker News Reader

Comment by psycr | Hacker News Reader

A web API that consumes phone camera captured photos of nutritional labels, and spits back a structured JSON response of the data.

Primary consumers would be developers who're building products that depend on the user inputting this type of data - or warehouse logistics companies who could catalog this sort of data and middleman it to everyone else.

View on HN

This would make my idea for a mobile app that renders pictures of the sugar/salt/fat content in foods by OCRing the label a trivial problem, and it could lead to many other applications like a shopping assistant for diabetics, athletes, etc.

https://assemblymade.com/golden-fleece

This was the startup I was CTO at. We organized peoples' whole diets and we used the barcodes rather than OCR to get the nutrition facts of labeled products.

The product recommended home-cooked meals (web-only), found the healthiest restaurant dishes near you (mobile and web), and evaluated the nutritional fit of labeled food products in stores (mobile-only). The recommendation algorithm was all backend.

Too bad we got acqui-hired.

I didn't follow the acqui-hire and went on to work for a genetics startup.

Email me if you want more info or to see about making it exist again. We had an alpha test group of some very happy users :)

Dibs. I'll start working on it. If anyone wants to help out, message me or email me karan[at]goel.im

(Apologies if you're an OCR guru and this is a stupid comment, but...) I wonder if it would be easier to get an MVP together using Amazon Mechanical Turk. A quick Google search reveals that there are lots of commercial products providing OCR on nutrition labels, but getting open source OCR to work reliably with phone-quality images (poor lighting, curvature of containers, etc) is tricky.

Either way, I think OCR-as-a-service would be really useful, and nutrition labels are probably a great place to start. I think there's lots of room for improvement in the nutrition/meal planning space.

I don't know how hard it might be, but we have a team already (wow HN), and have started the work.

I know FoodEssentials used to do this on mTurk. I did them myself on that platform. They haven't been around in a while, though. I'd be more than happy to enter nutritional label information on mTurk should that method be chosen.

You can scan barcode most of the time.

Just be aware that they're proposing a change to the current format used in the United States. This post covers the proposed changes:

http://www.fda.gov/NewsEvents/Newsroom/PressAnnouncements/uc...

The product should already have a UPC code that is trivial to capture. From there, it would be a simple matter of looking up that code to map it to its nutrition information. There may even be databases (eg: FDA) where this is available. MyFitnessPal seems to have the nutritional info on just about every product I've scanned, and I can't tell if they crowdsourced all that data input and/or if they seeded it with some database.

I believe it's crowdsourced. There are many items I eat that they don't have in their database. The app allows you to submit the nutrition info to their database and it gets recognized afterwards. MyFitnessPal is very popular. They should have one of the biggest databases out there.

Reply on news.ycombinator.com