README

Contents: Description of how to install and run facebookresearch's Detectron on AWS. Website Resources:

1. Website for installing nvidia-docker on AWS: https://github.com/NVIDIA/nvidia-docker/wiki/Deploy-on-Amazon-EC2
2. Website for

A) Need to install docker and docker-machine locally. B) Need to have an AWS *

Cost record: *3/30/18 1 Hr charge for EC2 Micro Instance ($0.006) *3/30/18 1 Hr charge for EC2 p2.xlarge GPU instance ($1)

Command to create AWS instance: docker-machine create --driver amazonec2
--amazonec2-region us-east-2
--amazonec2-zone a
--amazonec2-ami ami-916f59f4
--amazonec2-instance-type p2.xlarge
--amazonec2-vpc-id vpc-***
--amazonec2-access-key AKI**********
--amazonec2-secret-key *************Fbf
aws01

##################################################################### #RUNNING THE CODE WITH DOCKER. ##################################################################### 0) Ssh into the gpu machine, and run "screen" to open a new screen session (alternatively you can connect to a previous screen session). This is important as screen allows you to reconnect to a previous ssh session and check the output of a running job so that you don't have to stay ssh'd into the computer running your desired job.

1. Run "./run_docker.sh". This command will build the desired docker container, and will then run the docker container with the git repo (current directory) mounted to the root folder at "/ML_OD_Benchmarking". It will also open up a bash terminal at the root directory.
2. Run "cd /ML_OD_Benckmarking/detectron_scripts &", (Note this is run in the background so you can track each log process)
3. Run "./benchmark.sh", which will run each of the deep nets using images from "/ML_OD_Benchmarking/data/inputs/", and will output their results to "/ML_OD_Benchmarking/data/outputs/<model_name>.json"