Computer vision can estimate calorie content of food at a glance | Hacker News Reader

Computer vision can estimate calorie content of food at a glance

newscientist.com

I'd rather see them coming up with a better matching metric of nutritional/energetic value to humans than burning food and measuring how long it burns.

I'm sympathetic, but I also have to confess that calory counting works better than most folks want to admit. Some of that likely comes from the necessity of avoiding overly caloric foods that don't fill me up, I'm sure. But the month I started basic calorie counting/restriction, I managed to get below what I thought was my realistic low weight. It was dramatic.

Yes, it works with counting calories. But there are a lot of edge cases where the caloric value doesn’t represent the amount of energy your body can take out of the food. Also basically all values printed on the boxes are very vague as nobody does own measurements but relies on previous measurements instead. Here’s a nice article summing up the issues with counting calories:

https://www.vice.com/en/article/vv7d78/weve-been-counting-ca...

Completely agreed. I'm just questioning if those edge cases are really what is holding folks back.

That is, yes, calorie counting is problematic. Is there a non problematic way? Maybe? But probably not. Just different problems.

There's a video from an account called "Thunderf00t", (who I understand has controversial views in some areas, but hopefully that can be ignored here) that claims to "debunk" the food scanning apps Tellspec and SCiO. I'm not claiming that the app in the article uses the same techniques.

https://www.youtube.com/watch?v=VHjmlQdzpW4

I admit that the source of this article (newscientist.com) has a better track record of reliability than the source of the other video (youtube.com) and of the original claims (indiegogo.com), so we're likely not even in the same ballpark, but some of the analysis done by thunderf00t on why EM radiation alone will be of limited use was interesting to me.

(There's also a responds video, but I've not seen it: https://www.youtube.com/watch?v=wdv_tWDqNEA)

I know a Swiss startup that's working on something similar!

I'd be curious to see how it distinguishes mashed potato with and without a metric buttload of butter though. On pure reflectance alone?

And if you make scrambled eggs like a French chef (use the eggs as a delivery vehicle for butter), the calorie range might also vary by something like 400%.

It will probably have large error but still be 1000 times better than the intuition of a hungry human.

But what about when the hungry human tries to trick the computer by cooking plain looking food that packs tons of calories but appears healthy on the food log

Sounds stupid but it's exactly the kind of fat logic my unconscious hungry brain will use.

Nice. Maybe version 2 can tell me which slice of pie is most tasty per calorie count

Reply on news.ycombinator.com