OpenMatch: Open-set Consistency Regularization for Semi-supervised Learning with Outliers (NeurIPS 2021)

OpenMatch: Open-set Consistency Regularization for Semi-supervised Learning with Outliers (NeurIPS 2021)

This is an PyTorch implementation of OpenMatch. This implementation is based on Pytorch-FixMatch.

Requirements

• python 3.6+
• torch 1.4
• torchvision 0.5
• tensorboard
• numpy
• tqdm
• sklearn
• apex (optional)

See Pytorch-FixMatch for the details.

Usage

Dataset Preparation

This repository needs CIFAR10, CIFAR100, or ImageNet-30 to train a model.

To fully reproduce the results in evaluation, we also need SVHN, LSUN, ImageNet for CIFAR10, 100, and LSUN, DTD, CUB, Flowers, Caltech_256, Stanford Dogs for ImageNet-30. To prepare the datasets above, follow CSI.

mkdir data
ln -s path_to_each_dataset ./data/.

## unzip filelist for imagenet_30 experiments.
unzip files.zip

All datasets are supposed to be under ./data.

Train

Train the model by 50 labeled data per class of CIFAR-10 dataset:

sh run_cifar10.sh 50 save_directory

Train the model by 50 labeled data per class of CIFAR-100 dataset, 55 known classes:

sh run_cifar100.sh 50 10 save_directory

Train the model by 50 labeled data per class of CIFAR-100 dataset, 80 known classes:

sh run_cifar100.sh 50 15 save_directory

Run experiments on ImageNet-30:

sh run_imagenet.sh save_directory

Evaluation

Evaluate a model trained on cifar10

sh run_eval_cifar10.sh trained_model.pth

Trained models

Coming soon.

• CIFAR10-50-labeled
• CIFAR100-50-labeled-55class
• ImageNet-30

Acknowledgement

This repository depends a lot on Pytorch-FixMatch for FixMatch implementation, and CSI for anomaly detection evaluation. Thanks for sharing the great code bases!

Reference

This repository is contributed by Kuniaki Saito. If you consider using this code or its derivatives, please consider citing:

@article{saito2021openmatch,
  title={OpenMatch: Open-set Consistency Regularization for Semi-supervised Learning with Outliers},
  author={Saito, Kuniaki and Kim, Donghyun and Saenko, Kate},
  journal={arXiv preprint arXiv:2105.14148},
  year={2021}
}