This is cool.
One of my final project ideas for my Artificial Intelligence class was to recreate expensive and cumbersome human eye tracking with Artificial Intelligence, specifically a Neural Network. Much to my chagrin, with a little Googling I find out it has already been done and done quite well:
The eye tracking video’s on YouTube claim that the application uses over 100,000 training samples. Also if you peruse through the developer’s video’s you can see examples of the computer generated results compared to real human eye tracking tests as well as being used in different mediums (ex. tv, web).
In my conversation with Dzmitry, the applications developer, he was understandably a little hesitant to reveal details on how the application works. But I sent him some more questions, namingly if he used a neural network or not. I will update the post if he responds.
Here’s a snippet of our conversation:
Hi Serena,
Thanks for your interest. … Please have a look at selected examples for www (early demos) here http://www.youtube.com/adstapler or retail and ad industry (more mature demos) http://www.youtube.com/shopperlabs. …
Best Regards,
Dzmitry
Below is my favourite example. Alright I might be a little web biaist. The results are great, considering all good PPC marketers know getting the third spot on the right hand for Adsense is ideal.
Note: I also did look at Feng-GUI’s similar application but found it nowhere near the accuracy as Dzmitry’s. For example, another blogger points out some accuracy issues here.