Deep Dream

[T. Brown 1st MRB]

Deep dream pics. I put my kitty through it, the results remind me of The Thing movie.

Info

Code here

Original

Deep Dream

[Melnyk 1st MRB]

Lovely

[Ford 1st MRB]

We were just talking about this yesterday in class! Pretty amazing images have already been coming out from the use of this code.

It's like someone figured out how to write the code to recreate an acid trip.

Edited July 15, 2015 by Ford 1st MRB

[Guest]

[J. Hill]

wouldn't know about that. NEVER did the drug scene, although I did experience one night of hallucinations after getting severely dehydrated. The chemical imbalance really fucked me up bad. Saw the faces of demons in the dark when I went to bed. Did NOT get any sleep that night. It was truly terrifying. I could not move or talk. I ended up getting dehydrated the next day and took my second trip to the hospital. That's when I spent the weekend in the hospital.

[Woz 1st MRB]

We were just talking about this yesterday in class! Pretty amazing images have already been coming out from the use of this code.

It's like someone figured out how to write the code to recreate an acid trip.

I'm starting to specialize in machine learning, so I've been following this topic for about a year now.

It's interesting that you say that, because what you are essentially seeing is what your brain actually does to images, but without any higher level context filters. They generate these images by essentially creating a force feedback loop through certain lower layers until the low level filters are exaggerated so much that they are visible. Keep in mind that the low level filters look like simple edges, lines, and curves, and anything above that is a recombination of previous filters to achieve more complex shapes and therefore more complex filters.

The human brain doesn't really have a problem with filtering out a lot of the unneeded information, because it uses context clues to optimize the understanding of an object. These context clues can come from any of the 5 senses.

Now, when it comes to acid, a lot of these optimizations are thrown out the window due to increased activity in (some part of the brain I can't recall at the moment), causing our image recognition to deteriorate significantly.

Here is an example of a first level filter set for AlexNet (Keep in mind that a computer LEARNED these filters, and that the filters were not created by humans!)

Here is an example showing the evolution of feature detection in higher level filters.

Here is a network map for a simple AlexNet (8 layers):

Network map of GoogleNet (Look at all of those layers!)