- Python
- PyTorch
- Numpy
- scikit-learn
Data should be downloaded in ./Datasets/
. You may also use any path by setting the --dataset_root
argument to /your/path/
.
# Train CIFAR-10 with 5 labeled classes
CUDA_VISIBLE_DEVICES=0 python cifar10_pretrain.py
# Train CIFAR-100 with 80 labeled classes
CUDA_VISIBLE_DEVICES=0 python cifar100_pretrain.py
# Train TinyImageNet with 80 labeled classes
CUDA_VISIBLE_DEVICES=0 python tinyimagenet_pretrain.py
# Train CIFAR-10 with 5 unlabeled classes
CUDA_VISIBLE_DEVICES=0 python cifar10_ResTune.py
# Train CIFAR-100 with 20 unlabeled classes
CUDA_VISIBLE_DEVICES=0 python cifar100_ResTune.py
# Train TinyImageNet with 20 unlabeled classes
CUDA_VISIBLE_DEVICES=0 python tinyimagenet_ResTune.py
The models and results are saved in ./results/
.