This is an unofficial implementation of the NeurIPS 2020 paper Unsupervised Data Augmentation (UDA).
| Augmentation | Paper | Reproduced |
|---|---|---|
| Crop and flip | 10.94 | 10.93 |
| CutOut | 5.43 | 6.00 |
- Setting: CIFAR-10 4000
- Training with 4,000 labeled samples and 46,000 unlabeled samples
- Python >= 3.6
- PyTorch >= 1.5
- torchvision >= 0.6
- numpy
- Pillow
- ruamel.yaml
- sklearn
- tqdm
See train.py and config files in the config folder for more information
GPLv3