Few-Normal-Shot Tile Anomaly Detection with Lightweight Multiscale Memories

Xueli Liu

doi:10.62677/IJETAA.2608152

Authors

Xueli Liu Ziwen Co., Limited, Beijing,China Author

DOI:

https://doi.org/10.62677/IJETAA.2608152

Keywords:

Ceramic tile inspection, Anomaly detection, Few-shot learning, Multiscale features, Memory bank

Abstract

Surface inspection of ceramic tiles is constrained by scarce representative defects and costly pixel annotation. This study evaluates lightweight memory banks fitted with only a few normal images. A frozen pretrained ResNet18 provides multiscale local descriptors, which are optionally rescaled using normal-image channel statistics and layer widths, projected into a compact space, and stored in a fixed-capacity coreset. The initial evaluation comprises 144 configurations on the Tile, Grid, and Wood categories of MVTec AD, with one, four, or sixteen normal images and three support-set seeds. For four-shot Tile, the normalized multiscale model achieves an image-level area under the receiver operating characteristic curve of 97.92 percent and a pixel-level value of 93.61 percent. Relative to a same-backbone adaptation of the authors' PatchCore code, these values improve by 8.25 and 0.94 percentage points, whereas pixel average precision decreases by 2.67 points. A larger WideResNet50 baseline remains more accurate. Supplementary controlled diagnostics do not establish that the combined normalization components outperform simpler variants. The contribution is a reproducible empirical assessment of accuracy and computational tradeoffs, rather than a claim of uniformly superior localization.

Downloads

Download data is not yet available.

References

P. Bergmann, K. Batzner, M. Fauser, et al., "The MVTec Anomaly Detection Dataset: A Comprehensive Real-World Dataset for Unsupervised Anomaly Detection," International Journal of Computer Vision, vol. 129, no. 4, 2021, pp. 1038–1059, doi: 10.1007/s11263-020-01400-4.

T. Defard, A. Setkov, A. Loesch, et al., "PaDiM: A Patch Distribution Modeling Framework for Anomaly Detection and Localization," in Pattern Recognition, ICPR International Workshops and Challenges, LNCS, 2021, pp. 475–489, doi: 10.1007/978-3-030-68799-1_35.

K. Roth, L. Pemula, J. Zepeda, et al., "Towards Total Recall in Industrial Anomaly Detection," in Proc. IEEE/CVF Computer Vision and Pattern Recognition (CVPR), 2022, pp. 14298–14308, doi: 10.1109/cvpr52688.2022.01392.

C.-L. Li, K. Sohn, J. Yoon, et al., "CutPaste: Self-Supervised Learning for Anomaly Detection and Localization," in Proc. IEEE/CVF Computer Vision and Pattern Recognition (CVPR), 2021, pp. 9659–9669, doi: 10.1109/cvpr46437.2021.00954.

V. Zavrtanik, M. Kristan, and D. Skocaj, "DRAEM - A Discriminatively Trained Reconstruction Embedding for Surface Anomaly Detection," in Proc. IEEE/CVF International Conference on Computer Vision (ICCV), 2021, pp. 8310–8319, doi: 10.1109/iccv48922.2021.00822.

G. Wang, S. Han, E. Ding, et al., "Student-Teacher Feature Pyramid Matching for Anomaly Detection," in Proc. British Machine Vision Conference (BMVC), 2021, paper 349, doi: 10.5244/c.35.349.

M. Rudolph, B. Wandt, and B. Rosenhahn, "Same Same but DifferNet: Semi-Supervised Defect Detection With Normalizing Flows," in Proc. IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2021, pp. 1906–1915, doi: 10.1109/wacv48630.2021.00195.

D. Gudovskiy, S. Ishizaka, and K. Kozuka, "CFLOW-AD: Real-Time Unsupervised Anomaly Detection With Localization via Conditional Normalizing Flows," in Proc. IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2022, pp. 1819–1828, doi: 10.1109/wacv51458.2022.00188.

H. Deng, and X. Li, "Anomaly Detection via Reverse Distillation From One-Class Embedding," in Proc. IEEE/CVF Computer Vision and Pattern Recognition (CVPR), 2022, pp. 9727–9736, doi: 10.1109/cvpr52688.2022.00951.

Z. Liu, Y. Zhou, Y. Xu, et al., "SimpleNet: A Simple Network for Image Anomaly Detection and Localization," in Proc. IEEE/CVF Computer Vision and Pattern Recognition (CVPR), 2023, pp. 20402–20411, doi: 10.1109/cvpr52729.2023.01954.

X. Zhang, S. Li, X. Li, et al., "DeSTSeg: Segmentation Guided Denoising Student-Teacher for Anomaly Detection," in Proc. IEEE/CVF Computer Vision and Pattern Recognition (CVPR), 2023, pp. 3914–3923, doi: 10.1109/cvpr52729.2023.00381.

T. D. Tien, A. T. Nguyen, N. H. Tran, et al., "Revisiting Reverse Distillation for Anomaly Detection," in Proc. IEEE/CVF Computer Vision and Pattern Recognition (CVPR), 2023, pp. 24511–24520, doi: 10.1109/cvpr52729.2023.02348.

J. Jeong, Y. Zou, T. Kim, et al., "WinCLIP: Zero-/Few-Shot Anomaly Classification and Segmentation," in Proc. IEEE/CVF Computer Vision and Pattern Recognition (CVPR), 2023, pp. 19606–19616, doi: 10.1109/cvpr52729.2023.01878.

X. Li, Z. Zhang, X. Tan, et al., "PromptAD: Learning Prompts with only Normal Samples for Few-Shot Anomaly Detection," in Proc. IEEE/CVF Computer Vision and Pattern Recognition (CVPR), 2024, pp. 16848–16858, doi: 10.1109/cvpr52733.2024.01594.

Q. Zhou, G. Pang, Y. Tian, et al., "AnomalyCLIP: Object-agnostic Prompt Learning for Zero-shot Anomaly Detection," in Proc. International Conference on Learning Representations (ICLR), 2024. [Online]. Available: https://openreview.net/forum?id=buC4E91xZE, preprint DOI: 10.48550/arXiv.2310.18961.

S. Damm, M. Laszkiewicz, J. Lederer, et al., "AnomalyDINO: Boosting Patch-Based Few-Shot Anomaly Detection with DINOv2," in Proc. IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2025, pp. 1319–1329, doi: 10.1109/wacv61041.2025.00136.

Z. Gu, B. Zhu, G. Zhu, et al., "UniVAD: A Training-free Unified Model for Few-shot Visual Anomaly Detection," in Proc. IEEE/CVF Computer Vision and Pattern Recognition (CVPR), 2025, pp. 15194–15203, doi: 10.1109/cvpr52734.2025.01415.

S. Wu, Y. Wang, X. Liu, et al., "DFM: Differentiable Feature Matching for Anomaly Detection," in Proc. IEEE/CVF Computer Vision and Pattern Recognition (CVPR), 2025, pp. 15224–15233, doi: 10.1109/cvpr52734.2025.01418.

K. Batzner, L. Heckler, and R. König, "EfficientAD: Accurate Visual Anomaly Detection at Millisecond-Level Latencies," in Proc. IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2024, pp. 127–137, doi: 10.1109/wacv57701.2024.00020.

X. Zhang, M. Xu, and X. Zhou, "RealNet: A Feature Selection Network with Realistic Synthetic Anomaly for Anomaly Detection," in Proc. IEEE/CVF Computer Vision and Pattern Recognition (CVPR), 2024, pp. 16699–16708, doi: 10.1109/cvpr52733.2024.01580.

J. Santos, T. Tran, and O. Rippel, "Optimizing PatchCore for Few/many-shot Anomaly Detection," arXiv preprint arXiv:2307.10792, 2023, preprint DOI: 10.48550/arXiv.2307.10792.

M. Fučka, V. Zavrtanik, and D. Skočaj, "ObjectCore - Efficient Few-shot Logical Anomaly Detection using Object Representations," in Proc. IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), 2026, pp. 3857–3867, doi: 10.1109/wacv61042.2026.00376.