Federated Learning for Computer Vision

Federated Learning for vision classification using Flower and Pytorch

Description

This repository implements several models and custom strategies for federated learning in computer vision using flower for multilabel classification.

Getting Started

Dependencies

• Docker

File structure

flower_federated_learning/
└── ofb-flower/
    ├── server/
    │   ├── flower/
    │   │   ├── run.sh
    │   │   ├── requirements.txt
    │   │   ├── ...
    │   │   └── src/
    │   ├── build_run.sh
    │   ├── ...
    │   └── mlflow/
    └── client/
        ├── src/
        ├── requirements.txt
        ├── ...
        └── run_python.sh   

Setting environment

Environment variables used in docker-compose are in client/.env and server/.env You have to set at least the correct IP address and port for the clients to target mlflow and flower server.

Start server and clients

server and client builder located respectively in ofb-flower/server and ofb-flower/client

bash build_run.sh

Help

If you want to run server without docker make sure you have the necessary requirements for Python packages and at least Python3.7

python3 -m pip install requirements.txt

Authors

Contributors names and contact info

Hugo Math @mathugo