ResTune

Dependencies

• Python
• PyTorch
• Numpy
• scikit-learn

Data preparation

Data should be downloaded in ./Datasets/. You may also use any path by setting the --dataset_root argument to /your/path/.

Supervised pre-training with labeled data

# 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 

Unsupervised clustering with unlabeled data)

# 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 

Data preparation

The models and results are saved in ./results/.