FedVision

FedVision is a Visual Object Detection Platform Powered by Federated Learning

Quick start

Prerequisites

Too run FedVision, following dependency or tools required:

• machine to install fedvision-deploy-toolkit:

  • python virtualenv with Python>=3

  • setup SSH password less login to machine(s) for deploy fedvision framework.

• machine(s) to deploy fedvision framework:

  • Python>=3.7(with pip)

  • an isolated directory (each directory will be deployed with a copy of code)

Deploy

1. install fedvision deploy toolkit

   # ceate a python virtual envirement (recommanded) or use an exist one.
   source <virtualenv>/bin/activate
   python -m pip install -U pip && python -m pip install fedvision_deploy_toolkit

2. generate deploy template

   fedvision-deploy template standalone

3. read comments in generated template standalone_template.yaml and modify as you want.

4. run deploy cmd

   fedvision-deploy deploy deploy standalone_template.yaml

Services start and stop

Services could be start/stop with scripts in Fedvision/sbin or, use fedvision deploy toolkits:

fedvision-deploy services all start standalone_template.yaml

Run examples

Jobs could be submitted at each deployed machine with master service started.

cd /data/projects/fedvision
source venv/bin/activate
export PYTHONPATH=$PYTHONPATH:/data/projects/fedvision/FedVision
sh FedVision/examples/paddle_mnist/run.sh 127.0.0.1:10002

Architecture and Runtime Framework

We utilize PaddleFL to makes PaddlePaddle programs federated and utilize PaddleDetection to generate object detection program. This project may be extended to utilize pytorch's Ecology in future versions as well.

At runtime, each Party connects with coordinator and proposal jobs to or subscribe jobs from coordinator to participate in federated learning training cycle.

Documentation

Documentation can be generated using Mkdocs. Users can build the documentation using

source venv/bin/activate # venv to build docs
pip install -r requirement.txt
pip install -r docs/requirements.txt
mkdocs serve

Open up http://127.0.0.1:8000/ in your browser.

The latest version of online documentation can be found at

License

FedVision is released under Apache License 2.0. Please note that third-party libraries may not have the same license as FedVision.

Contributing

Any contributions you make are greatly appreciated!

• Please report bugs by submitting a GitHub issue.

• Please submit contributions using pull requests.