DP Benchmarks

This repository contains the codebase for the paper Rethinking Benchmarks for Differentially Private Image Classification.

Prerequisites

This project runs on Python 3.10 and necessary installation packages are in requirements.txt

• Run pip install -r requirements.txt to install all necessary packages

Conda can be also be used to building environment

Run conda env create -f=environment.yml $ conda activate dpbenchmarks

For the ScatterNet experiments please use the following: conda env create -f=sn_env.yml $ conda activate sn_env

This repository also uses the functional module of Opacus and hence needs opacus to be installed from scratch. This can be done using:

git clone https://github.com/pytorch/opacus.git
cd opacus
pip install -e .

Datasets

Datasets can be obtained from the following links:

1. CheXpert
2. EyePACS: Data TestLabels

Please downlad the raw files in a folder data/raw

The weights for the WRN 28-10 pretrained on ImageNet-1k can be found here: WRN 28-10

How to run

The CONFIG.py can be used to tweak the experiments.

Run using python main.py