ARMORY is a test bed for running scalable evaluations of adversarial defenses. Configuration files are used to launch local or cloud instances of the ARMORY docker container. Models, datasets, and evaluation scripts can be pulled from external repositories or from the baselines within this project.
pip install armory-testbed
Upon installing armory, a directory will be created at ~/.armory. This user
specific folder is the default directory for downloaded datasets and evaluation
outputs. Defaults can be changed by editing ~/.armory/config.json
ARMORY works by running an evaluation configuration file within the armory docker
ecosystem. To do this, simply run armory run <path_to_evaluation.json>.
Please see example configuration files for runnable configs.
The current working directory and armory installation directory will be mounted
inside the container and the armory.eval.Evaluator class will proceed to run the
evaluation script that is written in the evaluation['eval_file'] field of the
config.
For more detailed information on the evaluation config file please see the documentation.
Note: Since ARMORY launches Docker containers, the python package must be ran on system host.
As an example:
pip install armory-testbed
git clone https://github.com/twosixlabs/armory-example.git
cd armory-example
armory run example_config.json
Debugging evaluations can be performed interactively by passing --interactive and
following the instructions to attach to the container in order to use pdb or other
interactive tools. There is also support for --jupyter which will open a port on
the container and allow notebooks to be ran inside the armory environment.
At the moment our evaluations are created so that attacks and defenses may be interchanged. To do this we standardize all attacks and defenses as subclasses of their respective implementations in adversarial-robustness-toolbox
Armory is intended to be a lightweight python package which standardizes all evaluations inside a docker container. Users are encouraged to use the available images on dockerhub:
docker pull twosixarmory/tf1:0.3.3
docker pull twosixarmory/tf2:0.3.3
docker pull twosixarmory/pytorch:0.3.3
However if there are issues downloading the images (e.g. proxy) they can be built within this repo:
docker build --target armory-tf1 -t twosixarmory/tf1:0.3.3 .
docker build --target armory-tf2 -t twosixarmory/tf2:0.3.3 .
docker build --target armory-pytorch -t twosixarmory/pytorch:0.3.3 .
By default when launching an ARMORY instance the current working directory will be mounted as your default directory.This enables users to run modules from ARMORY baselines, as well as modules from the user project.
Depending on the task, docker memory for an ARMORY container must be at least 8 GB to run properly (preferably 16+ GB). On Mac and Windows Desktop versions, this defaults to 2 GB. See the docs to change this: