image_classification

Single class image classification

Setup environment

1. Same as the default setup.

Download and preprocess CIFAR10 dataset

Downloads CIFAR10 dataset from the original source and preprocess into pickles.

python -m dataset.cifar10.download_preprocess --output_dir data/cifar10

Run benchmarks

Benchmarks are implemented for both PyTorch and TensorFlow, available in image_classification/pytorch and image_classification/tf respectively. The commands are the same.

# pytorch or tf
framework=pytorch

CIFAR10 IID no DP:

python image_classification/${framework}/train.py --args_config image_classification/configs/baseline.yaml

CIFAR10 non-IID no DP:

python image_classification/${framework}/train.py --args_config image_classification/configs/baseline.yaml --dataset cifar10 --central_num_iterations 3000

CIFAR10 IID central DP:

python image_classification/${framework}/train.py --args_config image_classification/configs/baseline.yaml --central_privacy_mechanism gaussian_moments_accountant --central_num_iterations 3000

CIFAR10 non-IID central DP:

python image_classification/${framework}/train.py --args_config image_classification/configs/baseline.yaml --dataset cifar10 --central_privacy_mechanism gaussian_moments_accountant --central_num_iterations 3000