PyTorch implementation of AutoAugment

This repository contains code for AutoAugment (only using paper's best policies) based on AutoAugment: Learning Augmentation Policies from Data implemented in PyTorch.

Requirements

WideResNet28-10 baseline on CIFAR-10:

python train.py

WideResNet28-10 +Cutout, AutoAugment on CIFAR-10:

python train.py --cutout True --auto-augment True

Model	Error rate	Loss	Error rate (paper)
WideResNet28-10 baseline	3.82	0.1576	3.87
WideResNet28-10 +Cutout	3.40	0.1280	3.08
WideResNet28-10 +Cutout, AutoAugment	2.91	0.0994	2.68

Learning curves of loss and accuracy.

Name		Name	Last commit message	Last commit date
Latest commit History 5 Commits
LICENSE		LICENSE
README.md		README.md
acc.png		acc.png
auto_augment.py		auto_augment.py
example.png		example.png
loss.png		loss.png
test.py		test.py
train.py		train.py
utils.py		utils.py
wide_resnet.py		wide_resnet.py