FastFlow: Unsupervised Anomaly Detection and Localization via 2D Normalizing Flows. Yu, J., Zheng, Y., Wang, X., Li, W., Wu, Y., Zhao, R., & Wu, L. 11, 2021. ![FastFlow: Unsupervised Anomaly Detection and Localization via 2D Normalizing Flows [pdf]](https://bibbase.org/img/filetypes/pdf.svg "pdf")
Paper ![FastFlow: Unsupervised Anomaly Detection and Localization via 2D Normalizing Flows [link]](https://bibbase.org/img/filetypes/link.svg "link")
Website abstract bibtex Unsupervised anomaly detection and localization is crucial to the practical
application when collecting and labeling sufficient anomaly data is infeasible.
Most existing representation-based approaches extract normal image features
with a deep convolutional neural network and characterize the corresponding
distribution through non-parametric distribution estimation methods. The
anomaly score is calculated by measuring the distance between the feature of
the test image and the estimated distribution. However, current methods can not
effectively map image features to a tractable base distribution and ignore the
relationship between local and global features which are important to identify
anomalies. To this end, we propose FastFlow implemented with 2D normalizing
flows and use it as the probability distribution estimator. Our FastFlow can be
used as a plug-in module with arbitrary deep feature extractors such as ResNet
and vision transformer for unsupervised anomaly detection and localization. In
training phase, FastFlow learns to transform the input visual feature into a
tractable distribution and obtains the likelihood to recognize anomalies in
inference phase. Extensive experimental results on the MVTec AD dataset show
that FastFlow surpasses previous state-of-the-art methods in terms of accuracy
and inference efficiency with various backbone networks. Our approach achieves
99.4% AUC in anomaly detection with high inference efficiency.
@article{
title = {FastFlow: Unsupervised Anomaly Detection and Localization via 2D Normalizing Flows},
type = {article},
year = {2021},
websites = {https://arxiv.org/abs/2111.07677v2},
month = {11},
day = {15},
id = {d1f87dc6-22d8-330a-aba0-a8f0e8a4cb13},
created = {2023-06-12T12:12:14.961Z},
accessed = {2023-06-12},
file_attached = {true},
profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
last_modified = {2024-11-27T07:49:40.248Z},
read = {true},
starred = {false},
authored = {false},
confirmed = {false},
hidden = {false},
folder_uuids = {5a010301-acb6-4642-a6b2-8afaee1b741c,4b11b59b-846d-45b5-82b7-81d8bc3108b2,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},
private_publication = {false},
abstract = {Unsupervised anomaly detection and localization is crucial to the practical
application when collecting and labeling sufficient anomaly data is infeasible.
Most existing representation-based approaches extract normal image features
with a deep convolutional neural network and characterize the corresponding
distribution through non-parametric distribution estimation methods. The
anomaly score is calculated by measuring the distance between the feature of
the test image and the estimated distribution. However, current methods can not
effectively map image features to a tractable base distribution and ignore the
relationship between local and global features which are important to identify
anomalies. To this end, we propose FastFlow implemented with 2D normalizing
flows and use it as the probability distribution estimator. Our FastFlow can be
used as a plug-in module with arbitrary deep feature extractors such as ResNet
and vision transformer for unsupervised anomaly detection and localization. In
training phase, FastFlow learns to transform the input visual feature into a
tractable distribution and obtains the likelihood to recognize anomalies in
inference phase. Extensive experimental results on the MVTec AD dataset show
that FastFlow surpasses previous state-of-the-art methods in terms of accuracy
and inference efficiency with various backbone networks. Our approach achieves
99.4% AUC in anomaly detection with high inference efficiency.},
bibtype = {article},
author = {Yu, Jiawei and Zheng, Ye and Wang, Xiang and Li, Wei and Wu, Yushuang and Zhao, Rui and Wu, Liwei}
}
Downloads: 0
{"_id":"i5poH4HFWPaeEfR8k","bibbaseid":"yu-zheng-wang-li-wu-zhao-wu-fastflowunsupervisedanomalydetectionandlocalizationvia2dnormalizingflows-2021","author_short":["Yu, J.","Zheng, Y.","Wang, X.","Li, W.","Wu, Y.","Zhao, R.","Wu, L."],"bibdata":{"title":"FastFlow: Unsupervised Anomaly Detection and Localization via 2D Normalizing Flows","type":"article","year":"2021","websites":"https://arxiv.org/abs/2111.07677v2","month":"11","day":"15","id":"d1f87dc6-22d8-330a-aba0-a8f0e8a4cb13","created":"2023-06-12T12:12:14.961Z","accessed":"2023-06-12","file_attached":"true","profile_id":"f1f70cad-e32d-3de2-a3c0-be1736cb88be","group_id":"5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1","last_modified":"2024-11-27T07:49:40.248Z","read":"true","starred":false,"authored":false,"confirmed":false,"hidden":false,"folder_uuids":"5a010301-acb6-4642-a6b2-8afaee1b741c,4b11b59b-846d-45b5-82b7-81d8bc3108b2,bd3c6f2e-3514-47cf-bc42-12db8b9abe45","private_publication":false,"abstract":"Unsupervised anomaly detection and localization is crucial to the practical\napplication when collecting and labeling sufficient anomaly data is infeasible.\nMost existing representation-based approaches extract normal image features\nwith a deep convolutional neural network and characterize the corresponding\ndistribution through non-parametric distribution estimation methods. The\nanomaly score is calculated by measuring the distance between the feature of\nthe test image and the estimated distribution. However, current methods can not\neffectively map image features to a tractable base distribution and ignore the\nrelationship between local and global features which are important to identify\nanomalies. To this end, we propose FastFlow implemented with 2D normalizing\nflows and use it as the probability distribution estimator. Our FastFlow can be\nused as a plug-in module with arbitrary deep feature extractors such as ResNet\nand vision transformer for unsupervised anomaly detection and localization. In\ntraining phase, FastFlow learns to transform the input visual feature into a\ntractable distribution and obtains the likelihood to recognize anomalies in\ninference phase. Extensive experimental results on the MVTec AD dataset show\nthat FastFlow surpasses previous state-of-the-art methods in terms of accuracy\nand inference efficiency with various backbone networks. Our approach achieves\n99.4% AUC in anomaly detection with high inference efficiency.","bibtype":"article","author":"Yu, Jiawei and Zheng, Ye and Wang, Xiang and Li, Wei and Wu, Yushuang and Zhao, Rui and Wu, Liwei","bibtex":"@article{\n title = {FastFlow: Unsupervised Anomaly Detection and Localization via 2D Normalizing Flows},\n type = {article},\n year = {2021},\n websites = {https://arxiv.org/abs/2111.07677v2},\n month = {11},\n day = {15},\n id = {d1f87dc6-22d8-330a-aba0-a8f0e8a4cb13},\n created = {2023-06-12T12:12:14.961Z},\n accessed = {2023-06-12},\n file_attached = {true},\n profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},\n group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},\n last_modified = {2024-11-27T07:49:40.248Z},\n read = {true},\n starred = {false},\n authored = {false},\n confirmed = {false},\n hidden = {false},\n folder_uuids = {5a010301-acb6-4642-a6b2-8afaee1b741c,4b11b59b-846d-45b5-82b7-81d8bc3108b2,bd3c6f2e-3514-47cf-bc42-12db8b9abe45},\n private_publication = {false},\n abstract = {Unsupervised anomaly detection and localization is crucial to the practical\napplication when collecting and labeling sufficient anomaly data is infeasible.\nMost existing representation-based approaches extract normal image features\nwith a deep convolutional neural network and characterize the corresponding\ndistribution through non-parametric distribution estimation methods. The\nanomaly score is calculated by measuring the distance between the feature of\nthe test image and the estimated distribution. However, current methods can not\neffectively map image features to a tractable base distribution and ignore the\nrelationship between local and global features which are important to identify\nanomalies. To this end, we propose FastFlow implemented with 2D normalizing\nflows and use it as the probability distribution estimator. Our FastFlow can be\nused as a plug-in module with arbitrary deep feature extractors such as ResNet\nand vision transformer for unsupervised anomaly detection and localization. In\ntraining phase, FastFlow learns to transform the input visual feature into a\ntractable distribution and obtains the likelihood to recognize anomalies in\ninference phase. Extensive experimental results on the MVTec AD dataset show\nthat FastFlow surpasses previous state-of-the-art methods in terms of accuracy\nand inference efficiency with various backbone networks. Our approach achieves\n99.4% AUC in anomaly detection with high inference efficiency.},\n bibtype = {article},\n author = {Yu, Jiawei and Zheng, Ye and Wang, Xiang and Li, Wei and Wu, Yushuang and Zhao, Rui and Wu, Liwei}\n}","author_short":["Yu, J.","Zheng, Y.","Wang, X.","Li, W.","Wu, Y.","Zhao, R.","Wu, L."],"urls":{"Paper":"https://bibbase.org/service/mendeley/bfbbf840-4c42-3914-a463-19024f50b30c/file/402245b0-4fb3-12e3-7025-6e70a829b52a/full_text.pdf.pdf","Website":"https://arxiv.org/abs/2111.07677v2"},"biburl":"https://bibbase.org/service/mendeley/bfbbf840-4c42-3914-a463-19024f50b30c","bibbaseid":"yu-zheng-wang-li-wu-zhao-wu-fastflowunsupervisedanomalydetectionandlocalizationvia2dnormalizingflows-2021","role":"author","metadata":{"authorlinks":{}},"downloads":0},"bibtype":"article","biburl":"https://bibbase.org/service/mendeley/bfbbf840-4c42-3914-a463-19024f50b30c","dataSources":["2252seNhipfTmjEBQ"],"keywords":[],"search_terms":["fastflow","unsupervised","anomaly","detection","localization","via","normalizing","flows","yu","zheng","wang","li","wu","zhao","wu"],"title":"FastFlow: Unsupervised Anomaly Detection and Localization via 2D Normalizing Flows","year":2021}