RSS Feeds
organise
ALIPR Learns How To Auto-Tag Photos
Posted by Gina Trapani at 9:00 AM on October 12, 2008
New image recognition webapp ALIPR (Automatic Linguistic Indexing of Pictures in Real-Time) is a on a mission to assign relevant tags to digital images based on their content, and wants you to help it learn. While digital image recognition has come a long way in recent years, it's still got a long way to go, and ALIPR's got its share of hits and misses. Upload an image to ALIPR or hand it a URL of an image already online, and the engine will suggest tags, and ask you to add to its list. Some of ALIPR's suggestions are spot-on, but others are way off. You can confirm the hits and suggest other tags to help the engine learn. Check out how ALIPR did with a few images from my Flickr photostream.
Given a photos of boats in the harbor and a skyline, ALIPR does pretty well with its tag suggestions:
But a clear photo of a spider doesn't net any tags that apply:
Given a more confusing photo of a rock sculpture outside a building, ALIPR identifies the sky and buildings:
But a clear photo of a fire only returns the "orange" tag, not the word fire:
The concept of ALIPR is super exciting for anyone who's building a library of digital photos, because the most tedious parts of organising and searching your photos is adding relevant keywords to them. So automatic photo tagging, when done well and accurately, could be a huge time saver. Can't wait to see ALIPR-like technology built-in to a desktop photo organizer like Picasa. In the meantime, you can also use ALIPR's growing library of tagged photos to search for images by keyword as well.
Comments (AU Comments · US Comments)
There are currently no AU comments for this post.
mdevens
Posted 10:39 AM 12/10/08
This is really cool and kind of terrifying at the same time. The closest thing I can think of as far as giving computers the ability to see and then react to sight.
mdevens
goingthewongway
Posted 10:53 AM 12/10/08
Is that the Jacobs School of Engineering I spy? Random. Don't worry ALIPR, even real people can't figure out that the rocks are supposed to be a bear.
goingthewongway
Aprocalypse
Posted 11:28 AM 12/10/08
I dunno, the spider one picked out that it was indoors - technically that's an appropriate tag!
Aprocalypse
Mister Cow Pnoy
Posted 11:59 AM 12/10/08
That spider only has 7 legs...maybe it was looking for 8?
Mister Cow Pnoy
samgtaylor
Posted 12:18 PM 12/10/08
Hahaha... look at the "most voted" category... (potentially NSFW)
samgtaylor
Triborough
Posted 12:47 PM 12/10/08
After trying about half a dozen shots, it was basically worthless.
Triborough
zikman
Posted 5:20 PM 12/10/08
looks a little more entertaining than that Google photo-tagging "game." don't know if that's still around, but it was sorta fun for a minute.
zikman
mmillmor
Posted 7:37 PM 12/10/08
Prolific source of adult pictures too by the look of things
mmillmor
Torley
Posted 12:17 AM 13/10/08
Sounds like it learns as it goes along, so the more people use it, hopefully the more accurate it'll be? To that end, seems like it'll be quite familiar with the female anatomy in no time flat.
So far, ALIPR has been generally inaccurate for my purposes - I've been feeding in pictures from the virtual world Second Life - but a combo of this and human-intelligence tasks (like MTurk and TagCow) might make a practical experiment.
Torley
gzammit
Posted 2:49 AM 14/10/08
The whole concept is cool, but the suggested tags it gave me for some of the photos that I tried where absolutely wrong. If you have some spare time it's worth trying maybe you'll teach it more than I did!
Gil
[www.pstpanic.com]
gzammit