Establish benchmark dataset

[bw4sz]

1. Across multiple taxa
2. Known performance on settings
3. Timed and profiled
4. Create a small scoring program that loads an image, ask whether there is a animal, hit y or n and associates it with some metadata. This will help with future feature selection.
5. Create new repo for new version, stop carrying around all this baggage.
6. Make an ec2 ami for testing