DeepDream is a type of software created by Google to perform the reverse of facial recognition using a convolutional neural network. Effectively, a neural network trained to recognize faces, animals, or object takes an image and adjusts it slightly to “enhance” what it perceives to be patterns resembling the data it is trained on. A free piece of software that performs this task is available at deepdreamgenerator.com, and this is what I used for my found system. I used an image of the Gates Hillman Center from the Carnegie Mellon Press Kit, and ran it through the Deep Dream Generator 25 times. The Deep Dream generator has multiple options for neural networks that you can use, and different images that you can “train” the network on. I opted to use different network settings for different loops, as I didn’t want the final image to consist of only one type of pattern. This resulted in a highly warped image consisting of what appear to be dogs, fruits, and colorful spirals. In the video below I showcase each iteration of the generator’s output, and list the network setting I used.
The collection of produced images can be found here: https://drive.google.com/drive/folders/1gBPbAUWQjrRO948-1sF7Zf4l6LjDVhHp?usp=sharing