Again, I noticed, as have others, that images of dogs keep appearing, as in, almost all the time. “Why is this?”, I wondered. Well, it turns out there’s a very clear and simple reason for this:
A neural network’s ability to recognize what’s in an image comes from being trained on an initial data set. In Deep Dream’s case, that data set is from ImageNet, a database created by researchers at Stanford and Princeton who built a database of 14 million human-labeled images. But Google didn’t use the whole database. Instead, they used a smaller subset of the ImageNet database released in 2012 for use in a contest… a subset which contained “fine-grained classification of 120 dog sub-classes.”In
If I understand this correctly, it means that Deep Dreams sees what we’ve told it to see. We’re getting better at training computers to recognize images, but we haven’t got to the point of being able to teach them how to teach themselves. Yet.