Trying to run Caffe on EC2
Complete instructions to configure an Ubuntu EC2 g-instance up to have Caffe and ffmpeg which is enough to create deep dream inception videos. These are my notes, take them for what they are.
Good for applying:
- https://github.com/google/deepdream (haven't run this yet)
- https://github.com/graphific/DeepDreamVideo (this one's scripts need a bit of care)
To use:
- Start an EC2 instance
- A g instance such as g2.8xlarge instance (cheaper if you go for spot instances)
- OS: Ubuntu Server 14.04 LTS (HVM), SSD Volume Type - ami-5189a661 (Amazon's current offering)
- Must go to instance details to make sure you have more storage (I used 64GB)
- Follow the many steps in steps.txt
Credit to the original deep dream inception authors:
- Alexander Mordvintsev
- Christopher Olah
- Mike Tyka
- http://googleresearch.blogspot.ch/2015/06/inceptionism-going-deeper-into-neural.html
To the Caffe project:
- http://caffe.berkeleyvision.org
- BVLC AMI: https://github.com/BVLC/caffe/wiki/Caffe-on-EC2-Ubuntu-14.04-Cuda-7
An alternative is a ready to go Docker image (I have not tested this):
And to the authors of the referenced guides (credited where they are used in steps.txt).
See also:
- https://photos.google.com/share/AF1QipPX0SCl7OzWilt9LnuQliattX4OUCj_8EP65_cTVnBmS1jnYgsGQAieQUc1VQWdgQ?key=aVBxWjhwSzg2RjJWLWRuVFBBZEN1d205bUdEMnhB
- https://github.com/BVLC/caffe/wiki/Install-Caffe-on-EC2-from-scratch-(Ubuntu,-CUDA-7,-cuDNN)
- https://github.com/graphific/DeepDreamVideo (this one's scripts need a bit of care)
- https://github.com/samim23/DeepDreamAnim (this one was slow for me, and did not re-assemble frames in correct order)
- http://rocknrollnerd.github.io/ml/2015/05/27/leopard-sofa.html
- https://news.ycombinator.com/item?id=9749660
- http://www.win-vector.com/blog/2015/06/neural-net-image-salad-again-with-code/
- https://www.reddit.com/r/deepdream/
- https://news.ycombinator.com/item?id=9865398