I’ve finally got to play with Deep dream. You know what they say, better late than never 🙂 Installing and compiling everything from source is a bit of a nightmare, especially on a Mac. There are several tutorials and “how to”s on the Internet – the best I found was this gist – but none of them got the job done, as they assume you’re starting from a freshly installed MacOS, which is almost never the case, so there will always be a different version of python or a different version of a library that something else depends on and so on. Just use patience and and keep a glass of warm milk nearby. Do not drink anything that might make you agitated or nervous, for the sake of your equipment :)))
Some blood, sweat and tears later, I did manage to install it, but there are easier alternative to get your pics deep-dreamed, the most easy to use being alpha.deepdreamr.com. Just play around with their tool and if you really like it, don’t forget to donate.
Here are some results. Click for a high-res image!
Pretty cool, especially if you start looking at the details.
And now the philosophical part 🙂
If you’ve read the original blog post from Google, they use multiple layers or artificial neurons and which level tries to make an assumption of what the image represents and then send that data to the next level, with the last level making the final call and these images are snapshots of what happens in between.
Since neuro-networks are supposed to mimic the human brain, in all its wonder, I can help asking my self: “is there something like that happening in my own head”? Do I also split in decision making into these small quanta of thought? Is every decision I take split into a set of several minor sub-decisions that I don’t have direct access to, as I always get the final result.
For example, every guy on the planet gets asked, sooner or later, a simple question: “what do you like at a girl?” and yet, none of us seems to be able to give a straight answer. Yes, the blonde photoshopped Barbie from underwear commercials seems to rank pretty high, but there are a lot of girls that don’t look like that and I like them – physically speaking – while others that are closer to that image I don’t.
Could the french expression “Je ne sais quoi” refer to a decision that you like something taken at an “intermediate” level of your thought pipeline that you can’t access?
I’m sure we’ll know in another 20-25 years 🙂 What a great time to be alive.
In the meantime, I wanted to see what happens if I feed an already “deep dream”ed picture back into the system, so I did so manually, with the pictures below. Between the first and the last, the figure of something that looks like a person started to take shape on my left hand:
I can’t help wonder what happens if you re-feed the image back a hundred times. Or a million? How much of the original image will still be there. Of course, a task like this cannot be completed manually, but since I have a box installed at home, I might try to script it and see what happens. That’s a nice project for a weekend afternoon 🙂