PyTorch Out-of-Distribution Detection

Out-of-Distribution (OOD) Detection with Deep Neural Networks based on PyTorch.

The library provides:

Out-of-Distribution Detection Methods
Loss Functions
Datasets
Neural Network Architectures as well as pretrained weights
Useful Utilities

and is designed such that it should be compatible with frameworks like pytorch-lightning and pytorch-segmentation-models. The library also covers some methods from closely related fields such as Open-Set Recognition, Novelty Detection, Confidence Estimation and Anomaly Detection.

📚 Documentation

The documentation is available here.

NOTE: An important convention adopted in pytorch-ood is that OOD detectors predict outlier scores that should be larger for outliers than for inliers. If you notice that the scores predicted by a detector do not match the formulas in the corresponding publication, it may be possible that we multiplied the scores by negative one to comply with this convention.

⏳ Quick Start

Load model pre-trained on CIFAR-10 with the Energy-Bounded Learning Loss [6], and predict on some dataset data_loader using Energy-based Out-of-Distribution Detection [6], calculating the common OOD detection metrics:

from pytorch_ood.model import WideResNet
from pytorch_ood.detector import EnergyBased
from pytorch_ood.utils import OODMetrics

# Create Neural Network
model = WideResNet(num_classes=10, pretrained="er-cifar10-tune").eval().cuda()

# Create detector
detector = EnergyBased(model)

# Evaluate
metrics = OODMetrics()

for x, y in data_loader:
    metrics.update(detector(x.cuda()), y)

print(metrics.compute())

You can find more examples in the documentation.

🛠 ️️Installation

The package can be installed via PyPI:

pip install pytorch-ood

Dependencies

torch
torchvision
scipy
torchmetrics

Optional Dependencies

scikit-learn for ViM
gdown to download some datasets and model weights
pandas for the examples.
segmentation-models-pytorch to run the examples for anomaly segmentation

📦 Implemented

Detectors:

Detector	Description	Year	Ref
OpenMax	Implementation of the OpenMax Layer as proposed in the paper Towards Open Set Deep Networks.	2016	[1]
Monte Carlo Dropout	Implements Monte Carlo Dropout.	2016	[4]
Maximum Softmax Probability	Implements the Softmax Baseline for OOD and Error detection.	2017	[5]
ODIN	ODIN is a preprocessing method for inputs that aims to increase the discriminability of the softmax outputs for In- and Out-of-Distribution data.	2018	[2]
Mahalanobis	Implements the Mahalanobis Method.	2018	[3]
Energy-Based OOD Detection	Implements the Energy Score of Energy-based Out-of-distribution Detection.	2020	[6]
Entropy	Uses entropy to detect OOD inputs.	2021	[26]
Maximum Logit	Implements the MaxLogit method.	2022	[17]
KL-Matching	Implements the KL-Matching method for Multi-Class classification.	2022	[17]
ViM	Implements Virtual Logit Matching.	2022	[24]

Objective Functions:

Objective Function	Description	Year	Ref
Objectosphere	Implementation of the paper Reducing Network Agnostophobia.	2016	[7]
Center Loss	Generalized version of the Center Loss from the Paper A Discriminative Feature Learning Approach for Deep Face Recognition.	2016	[12]
Outlier Exposure	Implementation of the paper Deep Anomaly Detection With Outlier Exposure.	2018	[8]
Deep SVDD	Implementation of the Deep Support Vector Data Description from the paper Deep One-Class Classification.	2018	[9]
Energy Regularization	Adds a regularization term to the cross-entropy that aims to increase the energy gap between IN and OOD samples.	2020	[6]
CAC Loss	Class Anchor Clustering Loss from Class Anchor Clustering: a Distance-based Loss for Training Open Set Classifiers	2021	[11]
Entropy Maximization	Entropy maximization and meta classification for OOD in semantic segmentation	2021	[26]
II Loss	Implementation of II Loss function from Learning a neural network-based representation for open set recognition.	2022	[10]
MCHAD Loss	Implementation of the MCHAD Loss friom the paper Multi Class Hypersphere Anomaly Detection.	2022	[23]

Image Datasets:

Dataset	Description	Year	Ref
TinyImages	The TinyImages dataset is often used as auxiliary OOD training data. However, use is discouraged.	2012	[19]
Textures	Textures dataset, also known as DTD, often used as OOD Examples.	2013	[18]
FoolingImages	OOD Images Generated to fool certain Deep Neural Networks.	2014	[14]
TinyImages300k	A cleaned version of the TinyImages Dataset with 300.000 images, often used as auxiliary OOD training data.	2018	[8]
MNIST-C	Corrupted version of the MNIST.	2019	[16]
CIFAR10-C	Corrupted version of the CIFAR 10.	2019	[13]
CIFAR100-C	Corrupted version of the CIFAR 100.	2019	[13]
ImageNet-C	Corrupted version of the ImageNet.	2019	[13]
ImageNet - A, O, R	Different Outlier Variants for the ImageNet.	2019	[15]
MVTech-AD	MVTech Anomaly Segmentation Dataset	2021	[22]
StreetHazards	Anomaly Segmentation Dataset	2022	[17]
PixMix	PixMix image augmentation method	2022	[25]

Text Datasets:

Dataset	Description	Year	Ref
Multi30k	Multi-30k dataset, as used by Hendrycks et al. in the OOD baseline paper.	2016	[20]
WikiText2	Texts from the wikipedia often used as auxiliary OOD training data.	2016	[21]
WikiText103	Texts from the wikipedia often used as auxiliary OOD training data.	2016	[21]
NewsGroup20	Textx from different newsgroups, as used by Hendrycks et al. in the OOD baseline paper.

🤝 Contributing

We encourage everyone to contribute to this project by adding implementations of OOD Detection methods, datasets etc, or check the existing implementations for bugs.

📝 Citing

pytorch-ood was presented at a CVPR Workshop in 2022. If you use it in a scientific publication, please consider citing:

@InProceedings{kirchheim2022pytorch,
    author    = {Kirchheim, Konstantin and Filax, Marco and Ortmeier, Frank},
    title     = {PyTorch-OOD: A Library for Out-of-Distribution Detection Based on PyTorch},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
    month     = {June},
    year      = {2022},
    pages     = {4351-4360}
}

🛡️ ️License

The code is licensed under Apache 2.0. We have taken care to make sure any third party code included or adapted has compatible (permissive) licenses such as MIT, BSD, etc. The legal implications of using pre-trained models in commercial services are, to our knowledge, not fully understood.

🔗 References

[1]	Bendale, A., & Boult, T. E. (2016). Towards open set deep networks. CVPR.

[2]	Liang, S., Li, Y., & Srikant, R. (2017). Enhancing the reliability of out-of-distribution image detection in neural networks. ICLR.

[3]	Lee, K., Lee, K., Lee, H., & Shin, J. (2018). A simple unified framework for detecting out-of-distribution samples and adversarial attacks. NeurIPS.

[4]	Gal, Y., & Ghahramani, Z. (2016). Dropout as a bayesian approximation: Representing model uncertainty in deep learning. ICML.

[5]	Hendrycks, D., & Gimpel, K. (2016). A baseline for detecting misclassified and out-of-distribution examples in neural networks. ICLR.

[6]	(1, 2, 3, 4) Liu, W., Wang, X., Owens, J., & Li, Y. (2020). Energy-based out-of-distribution detection. NeurIPS.

[7]	Dhamija, A. R., Günther, M., & Boult, T. (2018). Reducing network agnostophobia. NeurIPS.

[8]	(1, 2) Hendrycks, D., Mazeika, M., & Dietterich, T. (2018). Deep anomaly detection with outlier exposure. ICLR.

[9]	Ruff, L., et al. (2018). Deep one-class classification. ICML.

[10]	Hassen, M., & Chan, P. K. (2020). Learning a neural-network-based representation for open set recognition. SDM.

[11]	Miller, D., Sunderhauf, N., Milford, M., & Dayoub, F. (2021). Class anchor clustering: A loss for distance-based open set recognition. WACV.

[12]	Wen, Y., Zhang, K., Li, Z., & Qiao, Y. (2016). A discriminative feature learning approach for deep face recognition. ECCV.

[13]	(1, 2, 3) Hendrycks, D., & Dietterich, T. (2019). Benchmarking neural network robustness to common corruptions and perturbations. ICLR.

[14]	Nguyen, A., Yosinski, J., & Clune, J. (2015). Deep neural networks are easily fooled: High confidence predictions for unrecognizable images. CVPR.

[15]	Hendrycks, D., Zhao, K., Basart, S., Steinhardt, J., & Song, D. (2021). Natural adversarial examples. CVPR.

[16]	Mu, N., & Gilmer, J. (2019). MNIST-C: A robustness benchmark for computer vision. ICLR Workshop.

[17]	(1, 2, 3) Hendrycks, D., Basart, S., Mazeika, M., Mostajabi, M., Steinhardt, J., & Song, D. (2022). Scaling out-of-distribution detection for real-world settings. ICML.

[18]	Cimpoi, M., Maji, S., Kokkinos, I., Mohamed, S., & Vedaldi, A. (2014). Describing textures in the wild. CVPR.

[19]	Torralba, A., Fergus, R., & Freeman, W. T. (2007). 80 million tiny images: a large dataset for non-parametric object and scene recognition. IEEE Transactions on Pattern Analysis and Machine Learning.

[20]	Elliott, D., Frank, S., Sima'an, K., & Specia, L. (2016). Multi30k: Multilingual english-german image descriptions. Proceedings of the 5th Workshop on Vision and Language.

[21]	(1, 2) Merity, S., Xiong, C., Bradbury, J., & Socher, R. (2016). Pointer sentinel mixture models. ArXiv

[22]	Bergmann, P., Batzner, K., et al. (2021) The MVTec Anomaly Detection Dataset: A Comprehensive Real-World Dataset for Unsupervised Anomaly Detection. IJCV.

[23]	Kirchheim, K., Filax, M., Ortmeier, F. (2022) Multi Class Hypersphere Anomaly Detection. ICPR

[24]	Wang, H., Li, Z., Feng, L., Zhang, W. (2022) ViM: Out-Of-Distribution with Virtual-logit Matching. CVPR

[25]	Hendrycks, D, Zou, A, et al. (2022) PixMix: Dreamlike Pictures Comprehensively Improve Safety Measures. CVPR

[26]	(1, 2) Chan R, et al. (2021) Entropy maximization and meta classification for out-of-distribution detection in semantic segmentation. CVPR

kkirchheim/pytorch-ood