Component-aware anomaly detection framework for adjustable and logical industrial visual inspection

Liu, Tongkun; Li, Bing; Du, Xiao; Jiang, Bingke; Jin, Xiao; Jin, Liuyi; Zhao, Zhuo

doi:10.1016/j.aei.2023.102161

Cited by 6 publications

(6 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We selected several SOTA IAD models as our baselines, including E cientAD-M [31], AST [44], S-T [43], and GCAD[6] based on S-T networks, FastFlow [32] based on normalizing ows, PatchCore [30] based on features, ComAD [27] based on pre-segmentation, DRAEM [39] and CutPaste [40] based on reconstruction, SLSG [24] based on graph self-attention mechanism, and DiffusionAD [42] based on a diffusion model. For FastFlow and PatchCore, we utilized WideResNet-50 as the backbone.…”

Section: Comparison With Other Methodsmentioning

confidence: 99%

“…ComAD [27] partitioned the image into k different feature components through clustering and established a memory bank of multi-class components based on the training set [30]. During testing, anomaly detection is achieved by calculating distances from the contents in the memory bank at the component level.…”

Section: Introductionmentioning

confidence: 99%

“…PatchCore [30] directly establishes a query library, known as a memory bank, for the characteristics of normal patch blocks, using the maximum feature distance for anomaly scoring. Subsequently, some related works have been proposed based on the memory bank [27,33]. However, even with the memory bank size minimized, a complex feature matching process is still required during the inference stage, signi cantly constraining the model's inference speed.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

GLAE: Global-local Feature Autoencoder for Image Logical Anomaly Detection

Peng,

Ye,

Zhuang

et al. 2024

Preprint

View full text Add to dashboard Cite

Self-supervised anomaly detection in industrial images holds significant practical value. While existing anomaly detection models excel in numerous public benchmarks, their substantial computational complexity and inability to detect logical anomalies hinder their widespread practical application. To address this issue, we proposes a global-local feature autoencoder (GLAE) for anomaly detection, which can be seamlessly integrated into the lightweight student-teacher (S-T) framework in parallel. GLAE uses a novel global feature extractor as the encoder (GFencoder), comprising solely of simple convolutions. This significantly reduces the computational cost while effectively capturing the global semantic information of the image, enabling the completion of global-local information reconstruction for the image. During the training stage, the student and GLAE were exclusively trained on normal samples, and they were unable to accurately capture the local or global features of abnormal samples during testing. By computing the feature distance between the teacher, student, and autoencoder, the local and global anomalies of the image were determined. We evaluated our method using three industrial anomaly detection dataset collections, and GLAE demonstrated state-of-the-art (SOTA) performance in image-level logical anomaly detection, efficiently handling anomalies with less than 7 ms latency on an NVIDIA RTX 3090 GPU. This establishes it as a cost-effective solution applicable in industrial scenarios, and it introduces a new approach for utilizing convolution to extract global features from images.

show abstract

Section: Comparison With Other Methodsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

GLAE: Global-local Feature Autoencoder for Image Logical Anomaly Detection

Peng,

Ye,

Zhuang

et al. 2024

Preprint

View full text Add to dashboard Cite

show abstract

“…SIN-BAD (Tzachor, Hoshen et al 2023) extracts a set of orderless elements and randomly projects element features to compute a histogram, with anomaly scores obtained via density estimation. ComAD (Liu et al 2023a) applied K-Means clustering on pre-trained features to segment multiple components within an image. However, performance was limited because a precise discrimination of different components is challenging.…”

Section: Related Workmentioning

confidence: 99%

“…When using multiple anomaly scores and aggregating them, previous works simply add the scores (Tsai et al 2022) or manually set hyper-parameters to scale them (Liu et al 2023a). However, these approaches may degrade performance when the multiple scores follow different distributions or the hyper-parameters are incorrectly set (Table 3).…”

Section: Related Workmentioning

confidence: 99%

Few Shot Part Segmentation Reveals Compositional Logic for Industrial Anomaly Detection

Kim,

An,

Chikontwe

et al. 2024

AAAI

View full text Add to dashboard Cite

Logical anomalies (LA) refer to data violating underlying logical constraints e.g., the quantity, arrangement, or composition of components within an image. Detecting accurately such anomalies requires models to reason about various component types through segmentation. However, curation of pixel-level annotations for semantic segmentation is both time-consuming and expensive. Although there are some prior few-shot or unsupervised co-part segmentation algorithms, they often fail on images with industrial object. These images have components with similar textures and shapes, and a precise differentiation proves challenging. In this study, we introduce a novel component segmentation model for LA detection that leverages a few labeled samples and unlabeled images sharing logical constraints. To ensure consistent segmentation across unlabeled images, we employ a histogram matching loss in conjunction with an entropy loss. As segmentation predictions play a crucial role, we propose to enhance both local and global sample validity detection by capturing key aspects from visual semantics via three memory banks: class histograms, component composition embeddings and patch-level representations. For effective LA detection, we propose an adaptive scaling strategy to standardize anomaly scores from different memory banks in inference. Extensive experiments on the public benchmark MVTec LOCO AD reveal our method achieves 98.1% AUROC in LA detection vs. 89.6% from competing methods.

show abstract

MCAD: Multi-classification anomaly detection with relational knowledge distillation

Li,

Ge,

Yue

et al. 2024

Neural Comput & Applic

View full text Add to dashboard Cite

With the wide application of deep learning in anomaly detection (AD), industrial vision AD has achieved remarkable success. However, current AD usually focuses on anomaly localization and rarely investigates anomaly classification. Furthermore, anomaly classification is currently requested for quality management and anomaly reason analysis. Therefore, it is essential to classify anomalies while improving the accuracy of AD. This paper designs a novel multi-classification AD (MCAD) framework to achieve high-accuracy AD with an anomaly classification function. In detail, the proposal model based on relational knowledge distillation consists of two components. The first one employs a teacher–student AD model, utilizing a relational knowledge distillation approach to transfer the interrelationships of images. The teacher–student critical layer feature activation values are used in the knowledge transfer process to achieve anomaly detection. The second component realizes anomaly multi-classification using the lightweight convolutional neural network. Our proposal has achieved 98.95, 96.04, and 92.94% AUROC AD results on MNIST, FashionMNIST, and CIFAR10 datasets. Meanwhile, we earn 97.58 and 98.10% AUROC for AD and localization in the MVTecAD dataset. The average classification accuracy of anomaly classification has reached 76.37% in fifteen categories of the MVTec-AD dataset. In particular, the classification accuracy of the leather category has gained 95.24%. The results on the MVTec-AD dataset show that MCAD achieves excellent detection, localization, and classification results.

show abstract

Component-aware anomaly detection framework for adjustable and logical industrial visual inspection

Cited by 6 publications

References 23 publications

GLAE: Global-local Feature Autoencoder for Image Logical Anomaly Detection

GLAE: Global-local Feature Autoencoder for Image Logical Anomaly Detection

Few Shot Part Segmentation Reveals Compositional Logic for Industrial Anomaly Detection

MCAD: Multi-classification anomaly detection with relational knowledge distillation

Contact Info

Product

Resources

About