If you find this code useful in your research, please consider citing:
@article{zhong2017random,
title={Random Erasing Data Augmentation},
author={Zhong, Zhun and Zheng, Liang and Kang, Guoliang and Li, Shaozi and Yang, Yi},
journal={arXiv preprint arXiv:1708.04896},
year={2017}
}
Thanks for Marcus D. Bloice, Marcus D. Bloice reproduces our method in Augmentor. Augmentor is an image augmentation library in Python for machine learning.
ResNet-20 + Random Erasing on Fashion-MNIST: python fashionmnist.py --dataset fashionmnist --arch resnet --depth 20 --p 0.5
Other architectures
For ResNet: --arch resnet --depth (20, 32, 44, 56, 110)
For WRN: --arch wrn --depth 28 --widen-factor 10
Our results
You can reproduce the results in our paper:
CIFAR10
CIFAR10
CIFAR100
CIFAR100
Fashion-MNIST
Fashion-MNIST
Models
Base.
+RE
Base.
+RE
Base.
+RE
ResNet-20
7.21
6.73
30.84
29.97
4.39
4.02
ResNet-32
6.41
5.66
28.50
27.18
4.16
3.80
ResNet-44
5.53
5.13
25.27
24.29
4.41
4.01
ResNet-56
5.31
4.89
24.82
23.69
4.39
4.13
ResNet-110
5.10
4.61
23.73
22.10
4.40
4.01
WRN-28-10
3.80
3.08
18.49
17.73
4.01
3.65
NOTE THAT, if you use the latest released Fashion-MNIST, the performance will slightly lower than the results reported in our paper. Please refer to the issue.
If you have any questions about this code, please do not hesitate to contact us.