 |
 |
As ACQUINE is an A.I. project and as its developers state on the web site, the algorithm will undergo algorithmic changes time to time. The paper (linked below) highlights the algorithm's main data source as MIT's photo.net which is online since 1997. The algorithm is constructed upon human ratings and obviously this is the key point. The researchers are essentially trying to "teach" Acquine about those human emotional reactions to visual stimulus. Ritendra Datta, Dhiraj Joshi, Jia Li and James Z. Wang, ``Studying Aesthetics in Photographic Images Using a Computational Approach,'' *Lecture Notes in Computer Science, vol. 3953, Proceedings of the European Conference on Computer Vision*, Part III, pp. 288-301, Graz, Austria, May 2006. http://infolab.stanford.edu/~wangz/project/imsearch/Aesthetics/ECCV06/datta.pdf<http://infolab.stanford.edu/%7Ewangz/project/imsearch/Aesthetics/ECCV06/datta.pdf> Obviously the question is how the machine can learn about it. Well, we - the human - give the feedback it requires in terms of ratings and the machine couples the ratings with machine-readable data, clusters the exceptions, thus learns. The details are explained in the paper. One of the most remarkable steps beyond machine readability of photos is the Exif *Exchangeable image file format *specification. http://en.wikipedia.org/wiki/Exif Recently I was involved as a consultant in a similar yet much smaller-in-scale project. The project involved learning from metadata within exif and coupling/clustering the data with the human feedback. The relevance is meaningful when you have a large dataset both in terms of pics and ratings. Hopefully ACQUINE will drive huge attention and offer relevant output (and input for a better algorithm). In order to achieve this, low rated pics should be motivated for rating besides the new and featured pics. Quick rating tools like the one in hotornot.com or a redeem mechanism like in The Amazon Mechanical Turk may be listed here as some good methods to motivate random rating. On the other hand ACQUINE only does the machine work and lets Photo.net deal with ratings. However it doesn't warn the user if the uploaded photo will be available on another website. Also, I think it'd be better if these two sites had been put together in one space, preferably in the most popular one. On the other hand a Facebook application would yield to awful results I guess:) Anyway thanks a lot for the post and no, I'm not mad with the numbers I received to my pics:) Bengi Turgan Phd Student Department of Industrial Design Istanbul Technical University