Photoree is a web-based recommendation engine for images. It can learn what kind of images that the user likes by serving images and asking for a simple thumbs-up, thumbs-down or no rating, and can use this information to point the user to images/photographs that it thinks he/she will like. It is based on a database of more than 1 million indexed images.
Do you enjoy good photography? Maybe you frequently find yourself downloading images or wallpaper backgrounds because of their artistic merit? If so then you probably visit Flickr and you probably are aware of the fantastic photography that can be found there.
The only problem is that it can be quite time consuming to surf Flickr or sites like it in order to find the best stuff, and this is where Photoree, a recommendation engine for images and photographs, can be very useful. To quote the site: “Photoree is to Photos what Last.FM is to music and StumbleUpon to websites”.
Here are more notes on this web service:
- The image database: the site states that Photoree has a database of more than 1 million images. All the images seem to share 2 characteristics (a) they all originate from Flickr, and (b) they are all under the Creative Commons license. Oh, and (c) they are mostly very good. All images are linked to their page on Flickr, where you can find out more information if you need to.
- Rating: Photoree will serve images to you based on what it learned from your previous ratings. Images which you “thumbs up” will be saved in your favorites. Curiously, once you rate an image up or down Photoree will let you skip rating a couple of images but will force you to make a judgment call after that, which feels weird at times but certainly makes the process more efficient.
- Your favorites: you can view these as pages of thumbnails in a grid, and can either share them with everyone, with people who have registered with Photoree only, with your contacts only, or just yourself. Note that you can add people as contacts very easily and at will.
- Stats: once you make 100 ratings Photoree will present you some ’stats’. The interesting thing here is a list of other users whose ratings profile were similar to yours (e.g. 80% similarity, 50% similarity etc). In theory the higher the similarity score the more taste should be similar to these useres.
- Adding your own images: it is possible to recommend images from Flickr for addition to the database. They ask that you supply a URL and that these are likely to interest people rather than of a personal nature (and that they have a Creative Commons license).
But does it work?: I’m really not sure. After rating more than 200 images presumably the algorithm has more of a grasp of what I like, and I was expecting that the stream of served images should, in theory, contain more frequency of images that I like, but that did not seem to be the case. Moreover I would have thought that I would like many of the favorite images that belong to some of the other users whose ratings profile was similar to mine (at 83% and 77% similarity), but that wasn’t quite the case either.
On the other hand, a concept like ’taste’ is not easy to pin down directly even by a human let alone an algorithm, so I am willing to give it the benefit of the down. One thing is certain, however, the images served by Photoree are of extremely high quality and are mostly extremely interesting and well done, so that alone makes it worth using. Check it out!
Version Tested: the site as of Feb 23, 2008.
Go to the Photoree site.