On: Ask HN: What are you working on (hacking)?

[user24]

computer vision JSON webservice - you supply image, it returns tags/keywords super fast. Spare time project.

Hoping to build a freemium model out of it for image libraries to use. Happy to speak to anyone with any kind of CV / object detection knowhow.

[paulgeo → user24]

A company in Berkeley is doing this: www.iqengines.com (see Developer API), and demonstration app www.omoby.com. HTML Post image and JSON/XML return label (also face, barcode, ocr, etc).

[user24 → paulgeo]

thanks! Hopefully I can kick their asses ;)

edit: which doesn't seem like it'll be too difficult. On their developer test I just get {"data": {"error": 0}} back for any image I try to upload.

[tocomment → user24]

How far have you gotten on this? I didn't know that level of CV technology existed to tag arbitrary images. It sounds like a really cool project.

[user24 → tocomment]

thanks, it is a really fun project to work on :)

At the moment I can return colour names and faces.

Example:

input: http://cm1.theinsider.com/media/0/91/63/mini-kim-kardashian-...

output: face, yellow, blue, black.

So just colours and faces at the moment. You're right that arbitrary images are ridiculously complicated, I'm hoping to start off on a smaller domain and build up :)

(edit: obviously I can return the coords of the face too, as well as coords of empty parts of the image etc, but tagging is really what I'm focusing on at the moment)