Reconstruction by inpainting for visual anomaly detection

By Vitjan Zavrtanik, Matej Kristan, Danijel Skocaj. Pattern Recognition, 2020

732 citations recorded by the source corpus.

Abstract Visual anomaly detection addresses the problem of classification or localization of regions in an image that deviate from their normal appearance. A popular approach trains an auto-encoder on anomaly-free images and performs anomaly detection by calculating the difference between the input and the reconstructed image. This approach assumes that the auto-encoder will be unable to accurately reconstruct anomalous regions. But in practice neural networks generalize well even to anomalies and reconstruct them sufficiently well, thus reducing the detection capabilities. Accurate reconstruction is far less likely if the anomaly pixels were not visible to the auto-encoder. We thus cast anomaly detection as a self-supervised reconstruction-by-inpainting problem. Our approach (RIAD) randomly removes partial image regions and reconstructs the image from partial inpaintings, thus addressing the drawbacks of auto-enocoding methods. RIAD is extensively evaluated on several benchmarks and sets a new state-of-the art on a recent highly challenging anomaly detection benchmark.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/2c89b183df320c3ef698989bdc5d1d4731c4d65d

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Hanqiu Deng
2. Xingyu Li
3. Yaochu Jin
4. Guoyang Xie
5. Feng Zheng
6. Shangwen Li
7. Jingbao Wang
8. Chengjie Wang

In Computer Vision and Pattern Recognition

1. Hanqiu Deng
2. Xingyu Li
3. Yaochu Jin
4. Guoyang Xie
5. Feng Zheng
6. Shangwen Li
7. Jingbao Wang
8. Chengjie Wang

Citations in the corpus

• Revisiting Reverse Distillation for Anomaly Detection (2023, methodology): “Later methods use deep models to improve the quality of reconstructed images [41, 42].”
• AnomalyGPT: Detecting Industrial Anomalies using Large Vision-Language Models (2023, methodology): “RIAD [33], SCADN [30], InTra [21] and AnoDDPM [28] employ different reconstruction network architectures, ranging from”
• DRÆM – A discriminatively trained reconstruction embedding for surface anomaly detection (2021, methodology): “Class US[4] RIAD[31]”
• Anomaly Detection via Reverse Distillation from One-Class Embedding (2022, methodology): “To address this issue, memory mechanism [11, 16, 26] , image masking strategy [42, 46] and pseudo-anomaly [28, 45] are incorporated in reconstruction-based methods.”
• Mixed supervision for surface-defect detection: from weakly to fully supervised learning (2021, unknown): “Various unsupervised [34, 4, 43] and weakly supervised [17, 44, 47] deep-learning approaches have been developed to reduce the need for a large amount of annotated data.”
• Deep Industrial Image Anomaly Detection: A Survey (2023, methodology): “RIAD [87] randomly masks a portion of the training set image at the patch level and reconstructs it using a U-Net encoder-decoder network [92].”