So I may have become a little bit obsessed with Google’s new Deep Dream hallucination algorithm. As I understand it, they use their image-recognition algorithm and make it enhance the features that it thinks it sees. Then, they run the new image back through the algorithm several more times, so that those features become amplified even more. However, the features that it thinks it sees tend not to correspond to actual features of the image, leading to hilarious/terrifying results.
Google has set up a repository for it on Github, but I couldn’t figure out how to set it up. So I thought that was the end of it. Until! I found out that there is a website that will process images for you! This was a very exciting development, so of course I had to go process some of my own pictures.
Here is a picture of one of our cats, Clyde:
Here’s a nice relaxing scene of a cliff by the ocean…
Here’s a Voodoo Donut (already kind of a weird thing):
Here’s Clyde again, but using a different filter:
Finally, I wondered what it would do to non-photo images. I fed it one of my screenshots from Monument Valley that I posted in the last post.
Anyway, back to business as usual, I guess. If you want to make your own Deep Dream images, you can use the nice website at dreamscopeapp.com. Have fun!