Towards Total Recall in Industrial Anomaly Detection

Towards Total Recall in Industrial Anomaly Detection. Roth, K., Pemula, L., Zepeda, J., Scholkopf, B., Brox, T., & Gehler, P. Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2022-June:14298-14308, IEEE Computer Society, 6, 2021.

Website doi abstract bibtex

Being able to spot defective parts is a critical component in large-scale industrial manufacturing. A particular challenge that we address in this work is the cold-start problem: fit a model using nominal (non-defective) example images only. While handcrafted solutions per class are possible, the goal is to build systems that work well simultaneously on many different tasks automatically. The best performing approaches combine embeddings from ImageNet models with an outlier detection model. In this paper, we extend on this line of work and propose \textbfPatchCore, which uses a maximally representative memory bank of nominal patch-features. PatchCore offers competitive inference times while achieving state-of-the-art performance for both detection and localization. On the challenging, widely used MVTec AD benchmark PatchCore achieves an image-level anomaly detection AUROC score of up to $99.6\%$, more than halving the error compared to the next best competitor. We further report competitive results on two additional datasets and also find competitive results in the few samples regime.\freefootnote$^*$ Work done during a research internship at Amazon AWS. Code: github.com/amazon-research/patchcore-inspection.

@article{
 title = {Towards Total Recall in Industrial Anomaly Detection},
 type = {article},
 year = {2021},
 keywords = {Recognition: detection,Self-& semi-& meta- Vision applications and systems,categorization,retrieval},
 pages = {14298-14308},
 volume = {2022-June},
 websites = {https://arxiv.org/abs/2106.08265v2},
 month = {6},
 publisher = {IEEE Computer Society},
 day = {15},
 id = {81d41879-9748-31cb-b0f0-53c69350d9bf},
 created = {2023-06-12T12:10:59.236Z},
 accessed = {2023-06-12},
 file_attached = {false},
 profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},
 group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},
 last_modified = {2023-06-12T12:10:59.236Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {false},
 hidden = {false},
 private_publication = {false},
 abstract = {Being able to spot defective parts is a critical component in large-scale
industrial manufacturing. A particular challenge that we address in this work
is the cold-start problem: fit a model using nominal (non-defective) example
images only. While handcrafted solutions per class are possible, the goal is to
build systems that work well simultaneously on many different tasks
automatically. The best performing approaches combine embeddings from ImageNet
models with an outlier detection model. In this paper, we extend on this line
of work and propose \textbfPatchCore, which uses a maximally representative
memory bank of nominal patch-features. PatchCore offers competitive inference
times while achieving state-of-the-art performance for both detection and
localization. On the challenging, widely used MVTec AD benchmark PatchCore
achieves an image-level anomaly detection AUROC score of up to $99.6\%$, more
than halving the error compared to the next best competitor. We further report
competitive results on two additional datasets and also find competitive
results in the few samples regime.\freefootnote$^*$ Work done during a
research internship at Amazon AWS. Code:
github.com/amazon-research/patchcore-inspection.},
 bibtype = {article},
 author = {Roth, Karsten and Pemula, Latha and Zepeda, Joaquin and Scholkopf, Bernhard and Brox, Thomas and Gehler, Peter},
 doi = {10.1109/CVPR52688.2022.01392},
 journal = {Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition}
}

Downloads: 0

{"_id":"TkA8zXn27Kjqqg9AJ","bibbaseid":"roth-pemula-zepeda-scholkopf-brox-gehler-towardstotalrecallinindustrialanomalydetection-2021","author_short":["Roth, K.","Pemula, L.","Zepeda, J.","Scholkopf, B.","Brox, T.","Gehler, P."],"bibdata":{"title":"Towards Total Recall in Industrial Anomaly Detection","type":"article","year":"2021","keywords":"Recognition: detection,Self-& semi-& meta- Vision applications and systems,categorization,retrieval","pages":"14298-14308","volume":"2022-June","websites":"https://arxiv.org/abs/2106.08265v2","month":"6","publisher":"IEEE Computer Society","day":"15","id":"81d41879-9748-31cb-b0f0-53c69350d9bf","created":"2023-06-12T12:10:59.236Z","accessed":"2023-06-12","file_attached":false,"profile_id":"f1f70cad-e32d-3de2-a3c0-be1736cb88be","group_id":"5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1","last_modified":"2023-06-12T12:10:59.236Z","read":false,"starred":false,"authored":false,"confirmed":false,"hidden":false,"private_publication":false,"abstract":"Being able to spot defective parts is a critical component in large-scale\nindustrial manufacturing. A particular challenge that we address in this work\nis the cold-start problem: fit a model using nominal (non-defective) example\nimages only. While handcrafted solutions per class are possible, the goal is to\nbuild systems that work well simultaneously on many different tasks\nautomatically. The best performing approaches combine embeddings from ImageNet\nmodels with an outlier detection model. In this paper, we extend on this line\nof work and propose \\textbfPatchCore, which uses a maximally representative\nmemory bank of nominal patch-features. PatchCore offers competitive inference\ntimes while achieving state-of-the-art performance for both detection and\nlocalization. On the challenging, widely used MVTec AD benchmark PatchCore\nachieves an image-level anomaly detection AUROC score of up to $99.6\\%$, more\nthan halving the error compared to the next best competitor. We further report\ncompetitive results on two additional datasets and also find competitive\nresults in the few samples regime.\\freefootnote$^*$ Work done during a\nresearch internship at Amazon AWS. Code:\ngithub.com/amazon-research/patchcore-inspection.","bibtype":"article","author":"Roth, Karsten and Pemula, Latha and Zepeda, Joaquin and Scholkopf, Bernhard and Brox, Thomas and Gehler, Peter","doi":"10.1109/CVPR52688.2022.01392","journal":"Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition","bibtex":"@article{\n title = {Towards Total Recall in Industrial Anomaly Detection},\n type = {article},\n year = {2021},\n keywords = {Recognition: detection,Self-& semi-& meta- Vision applications and systems,categorization,retrieval},\n pages = {14298-14308},\n volume = {2022-June},\n websites = {https://arxiv.org/abs/2106.08265v2},\n month = {6},\n publisher = {IEEE Computer Society},\n day = {15},\n id = {81d41879-9748-31cb-b0f0-53c69350d9bf},\n created = {2023-06-12T12:10:59.236Z},\n accessed = {2023-06-12},\n file_attached = {false},\n profile_id = {f1f70cad-e32d-3de2-a3c0-be1736cb88be},\n group_id = {5ec9cc91-a5d6-3de5-82f3-3ef3d98a89c1},\n last_modified = {2023-06-12T12:10:59.236Z},\n read = {false},\n starred = {false},\n authored = {false},\n confirmed = {false},\n hidden = {false},\n private_publication = {false},\n abstract = {Being able to spot defective parts is a critical component in large-scale\nindustrial manufacturing. A particular challenge that we address in this work\nis the cold-start problem: fit a model using nominal (non-defective) example\nimages only. While handcrafted solutions per class are possible, the goal is to\nbuild systems that work well simultaneously on many different tasks\nautomatically. The best performing approaches combine embeddings from ImageNet\nmodels with an outlier detection model. In this paper, we extend on this line\nof work and propose \\textbfPatchCore, which uses a maximally representative\nmemory bank of nominal patch-features. PatchCore offers competitive inference\ntimes while achieving state-of-the-art performance for both detection and\nlocalization. On the challenging, widely used MVTec AD benchmark PatchCore\nachieves an image-level anomaly detection AUROC score of up to $99.6\\%$, more\nthan halving the error compared to the next best competitor. We further report\ncompetitive results on two additional datasets and also find competitive\nresults in the few samples regime.\\freefootnote$^*$ Work done during a\nresearch internship at Amazon AWS. Code:\ngithub.com/amazon-research/patchcore-inspection.},\n bibtype = {article},\n author = {Roth, Karsten and Pemula, Latha and Zepeda, Joaquin and Scholkopf, Bernhard and Brox, Thomas and Gehler, Peter},\n doi = {10.1109/CVPR52688.2022.01392},\n journal = {Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition}\n}","author_short":["Roth, K.","Pemula, L.","Zepeda, J.","Scholkopf, B.","Brox, T.","Gehler, P."],"urls":{"Website":"https://arxiv.org/abs/2106.08265v2"},"biburl":"https://bibbase.org/service/mendeley/bfbbf840-4c42-3914-a463-19024f50b30c","bibbaseid":"roth-pemula-zepeda-scholkopf-brox-gehler-towardstotalrecallinindustrialanomalydetection-2021","role":"author","keyword":["Recognition: detection","Self-& semi-& meta- Vision applications and systems","categorization","retrieval"],"metadata":{"authorlinks":{}},"downloads":0},"bibtype":"article","biburl":"https://bibbase.org/service/mendeley/bfbbf840-4c42-3914-a463-19024f50b30c","dataSources":["2252seNhipfTmjEBQ"],"keywords":["recognition: detection","self-& semi-& meta- vision applications and systems","categorization","retrieval"],"search_terms":["towards","total","recall","industrial","anomaly","detection","roth","pemula","zepeda","scholkopf","brox","gehler"],"title":"Towards Total Recall in Industrial Anomaly Detection","year":2021}