danielvarga / deepdream-aws

Setting up and running the Deep Dream algorithm on AWS EC2

deepdream-aws

Setting up and running the Deep Dream algorithm on AWS EC2.

Go through the steps of readme.sh to create and set up an EC2 instance. We use Tito Costa's Torch+Caffe EC2 AMI, documented at http://blog.titocosta.com/post/110345699197/public-ec2-ami-with-torch-and-caffe-deep-learning . Big thanks, Tito!

setup.sh is what sets up the algorithm and its inputs on the instance, at the /home/ubuntu/deepdream-aws directory. Beware, its last line starts the layers.sh script as well, you'll probably want to comment that line out.

python dream.py IMAGE.jpg LAYER runs the algorithm for a given GoogLeNet layer. The output is IMAGE.LAYER.jpg.

bash layers.sh IMAGE.jpg runs the Deep Dream algorithm for a given image, for each layer of the neural network. Images with alpha channel are not allowed.

Forking

If you fork this, you'll have to change it to point to your fork at TWO places:

1. In readme.sh: https://raw.githubusercontent.com/danielvarga/deepdream-aws
2. In setup.sh: [email protected]:danielvarga/deepdream-aws.git

deepdream-aws's Issues

how to read output?

setting up and running the scripts generates a plethora of output files, presumably corresponding to individual network layers -- but how are they read out? opening them as jpgs doesn't seem to work.

files look like:
cout.daniel.jpg.pool4-3x3_s2

thanks!