Show HN: An algorithm to automatically turn photos of food into faces | Hacker News Reader

Show HN: An algorithm to automatically turn photos of food into faces

aaronrandall.com

211 pointsaaronrandall28 comments

Not bad, but not quite Arcimbaldo [0].

https://en.m.wikipedia.org/wiki/Giuseppe_Arcimboldo#/media/F...

Whoa. That's pretty rad! Why don't more people know about this guy? Considering he predates the surrealists by literally hundreds of years, you'd think he would get more press. This stuff would be right at home in an issue of Juxtapoz magazine.

Striking, isn't it? I ran across his work when taking cog psych/cog neuro courses. Arcimbaldo's paintings provide a great demonstration of face perception and — one could argue — face-seeking that our brains perform with no conscious effort.

Just to say, he's quite well known amongst artists and art historians, though I don't recall learning about him in my Western art history course. Along similar lines, but more well known is Hieronymus Bosch http://lmgtfy.com/?q=hieronymus+bosch.

Fun project, excellent article! Your figures made the algorithms incredibly easy to grasp.

Thanks, glad you enjoyed it :)

I don't understand how the part of the plate that was covered with food is reconstructed.

Does this assume that the plate has an uniform color and round symmetry? Does this work with plates that have circular color stripes?

It discards original plate and puts everything on image of an empty plate [0][1]. The effect is surely a bit more impressive when the plates match :)

[0] https://github.com/AaronRandall/Megabite/blob/b982d587677b93...

[1] https://github.com/AaronRandall/Megabite/blob/master/Megabit...

That's a great question! My original plan that once I had extracted all the items from the plate, the app would then detect the average colour of the remaining regions and "fill" the gaps - that wouldn't work well with plates that have different patterns, but would probably be good enough for plates of a single colour.

Due to time constraints (and ultimately solving the majority of the problems I had originally aimed to with this project), the app replaces the cropped circular region with a plain white plate (a separate image contained in the app) before placing the extracted items.

Maybe I'll go for the more advanced implementation if/when I spend more time on this project!

Great article, well explained and illustrated. The name is funnier than the app though. It literally translates to "giant dick" in French.

So does megabyte and megabit. Grow up! ;)

Those are homophones, not homonyms. I'll admit I had to do a double take on that name.

That's why they call it a mégaoctet.

I'm just happy that I got to see a rotational transform of bacon today.

This is a fantastic example of an interesting academic task, and very well explained. Thanks for sharing!

Awesome, glad you enjoyed it :)

This seems like a great example of a technically/algorithmically challenging problem that the layman would not appreciate, more so if this could be refined and applied to a wider set of images.

An elevator pitch for this, if you will, would likely not impress most non-technical people because, on it's face, what's so hard about rearranging foods into faces?

There are tons of tasks that are easy for a human to perform, but challenging for a computer to perform. Deepdream certainly seems to impress the layman.

Think of how revolutionary this is for people who incessantly take photos of their food and post to Instagram.

We need a "food->face" filter stat. If not for the poster, then one I can apply to any of my friends' photos they post.

Or maybe I need better friends.

Or eat more photogenic food myself.

Are there other example images? I don't have access to an iOS device right now.

Great! This article is a perfect example of excellent documentation for source code. It would be great if any project had such a good documentation.

Great! It's also a nice reminder that it's possible to do automatic image manipulation with tools other than neural networks.

We truly live in the future.

Nitpick: The area should be in units of pixels^2. Other than that, this is awesome.

Bacon to Kevin Bacon?

Am I the only person who read it as "feces"!!!

Nope.

Ugh, millennials.

http://buttersafe.com/2016/02/02/a-new-generation-of-art/

Reply on news.ycombinator.com