UDA : Unsupervised Data Augmentation
October 4, 2019 ยท View on GitHub
Unofficial PyTorch Implementation of Unsupervised Data Augmentation.
- Experiments on Text Dataset need to be done. Any Pull-Requests would be appreciated.
- Augmentation policies for SVHN, Imagenet using AutoAugment are not available publicly. We use policies from Fast AutoAugment.
Most of codes are from Fast AutoAugment.
Introduction
todo.
Run
$ python train.py -c confs/wresnet28x2.yaml --unsupervised
Experiments
Cifar10 (Reduced, 4k dataset)
Reproduce Paper's Result
| WResNet 28x2 | Paper | Our Converged(Top1 Err) | Our Best(Top1 Err) |
|---|---|---|---|
| Supervised | 20.26 | 21.30 | |
| AutoAugment | 14.1* | 15.4 | 13.4 |
| UDA | 5.27 | 6.58 | 6.27 |
SVHN
todo.
ImageNet
todo.
References
- Unsupervised Data Augmentation : https://arxiv.org/abs/1904.12848v1
- Official Tensorflow Implementation : https://github.com/google-research/uda
- Fast AutoAugment : https://github.com/kakaobrain/fast-autoaugment