Newer posts are loading.
You are at the newest post.
Click here to check if anything new just came in.

Deep Dream


Very informative article first.

Artificial Neural Networks used by Google are taught, by feeding them thousands of photos, to recognize various things (dogs, for instance). The way of testing if a Network learned the right thing, is feeding it a random picture (or just pure pixel noise) and asking to intensify any shapes that it can see - and it can see only what it was tought to see. That way, we can check if Network has learned the thing correctly.


…Or if we made a mistake and it now thinks that hands are part of dumbbells.

Google has made the code (called Deep Dream) public. There is also a website that let’s one create his own pictures. This particular Network seems to be taught to recognize dogs, but there are also lot of other types, for example - buildings and landscapes:


There is also an online live generator that recognizes about thousand different things (more about and more cool pictures here).

My favorites so far:


Not only humans see shapes in clouds; Alice in Wonderland, anyone?


I will never eat donut again in my life.

Don't be the product, buy the product!