CutPaste: Self-Supervised Learning for Anomaly Detection and Localization

By Chun-Liang Li, Kihyuk Sohn, Jinsung Yoon, Tomas Pfister. Computer Vision and Pattern Recognition, 2021

1,213 citations recorded by the source corpus.

We aim at constructing a high performance model for defect detection that detects unknown anomalous patterns of an image without anomalous data. To this end, we propose a two-stage framework for building anomaly detectors using normal training data only. We first learn self-supervised deep representations and then build a generative one-class classifier on learned representations. We learn representations by classifying normal data from the CutPaste, a simple data augmentation strategy that cuts an image patch and pastes at a random location of a large image. Our empirical study on MVTec anomaly detection dataset demonstrates the proposed algorithm is general to be able to detect various types of real-world defects. We bring the improvement upon previous arts by 3.1 AUCs when learning representations from scratch. By transfer learning on pretrained representations on ImageNet, we achieve a new state-of-the-art 96.6 AUC. Lastly, we extend the framework to learn and extract representations from patches to allow localizing defective areas without annotations during training.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/78d80c343d36baaf89f18e12d325cf6309fb6c8f

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

In Computer Vision and Pattern Recognition

Citations in the corpus

SimpleNet: A Simple Network for Image Anomaly Detection and Localization (2023, methodology): “The synthesizing-based methods [17, 30] estimate the decision boundary between the normal and anomalous by training on synthetic anomalies generated on anomaly-free images.”
CFLOW-AD: Real-Time Unsupervised Anomaly Detection with Localization via Conditional Normalizing Flows (2021, methodology): “Patch SVDD [42] and CutPaste [19] introduce a selfsupervised pretraining scheme for AD.”
Anomaly Detection via Reverse Distillation from One-Class Embedding (2022, methodology): “For example, CutPaste [23] generates pseudo outliers by adding small patches onto normal images and trains a model to detect these anomalous regions.”
Deep Industrial Image Anomaly Detection: A Survey (2023, neutral): “Both CutPaste [31] and CAVGA [42] treat anomaly detection as a classification problem, while GradCAM is used for pixel-level anomaly localization.”
EfficientAD: Accurate Visual Anomaly Detection at Millisecond-Level Latencies (2023, methodology): “Common methods include multivariate Gaussian distributions [16, 30, 48], Gaussian Mixture Models [38, 72], Normalizing Flows [22,47,51,52,68], and the k-Nearest Neighbor (kNN) algorithm [14, 39, 40, 50].”
SPot-the-Difference Self-Supervised Pre-training for Anomaly Detection and Segmentation (2022, methodology): “For both setups, existing state-of-the-art methods for surface anomaly detection commonly leverage supervised representations pre-trained on ImageNet [16], either as feature extractors [14,33] or as initialization for fine-tuning on the target dataset [26,44].”
WinCLIP: Zero-/Few-Shot Anomaly Classification and Segmentation (2023, methodology): “Due to the scarcity of anomalies, the major focus has been on oneclass methods with many normal images [7, 8, 20, 49, 51, 53].”
Self-Supervised Predictive Convolutional Attentive Block for Anomaly Detection (2021, methodology): “For the methods [18,34,37,39,49,79] chosen to serve as underlying models for SSPCAB, we use the official code from the repositories provided by the corresponding authors, inheriting the hyperparameters, e.”
Generalized Out-of-Distribution Detection: A Survey (2021, neutral): “These strategies include data augmentation [218], adversarial training [159], distillations [219], loss function enhancement [220], and usage of shallow [221], [222] and local features [223].”
CFA: Coupled-Hypersphere-Based Feature Adaptation for Target-Oriented Anomaly Localization (2022, background): “Self-supervised learningbased approaches such as [11, 21, 23] synthesized noise and used them as abnormal samples in learning.”
OpenOOD: Benchmarking Generalized Out-of-Distribution Detection (2022, background): “Although AD methods were originally designed to detect pixel-level appearance differences on MVTec-AD dataset, it proves to be potent when it comes to far-OOD detection such as DRAEM and CutPaste.”
AnomalyGPT: Detecting Industrial Anomalies using Large Vision-Language Models (2023, methodology): “Cut-paste [14] is a common technique in IAD domain for generating simulated anomaly images.”
FastFlow: Unsupervised Anomaly Detection and Localization via 2D Normalizing Flows (2021, methodology): “For example, they estimated the multidimensional Gaussian distribution (Li et al. 2021; Defard et al. 2020) by calculating the mean and variance for features, or used a clustering algorithm to estimate these normal features by normal clustering (Reiss et al. 2021; Roth et al. 2021).”
A survey on deep learning tools dealing with data scarcity: definitions, challenges, solutions, tips, and applications (2023, background): “An autoencoder model learns by giving input (input and target output) and generating input by encoding it to compact representation and later decoding it to its original [75].”
Student-Teacher Feature Pyramid Matching for Anomaly Detection (2021, methodology): “…two trends to achieve performance gains: (1) by pre-trained models, with a Wide-ResNet50 × 2 network [44], SPADE reports very competitive scores; (2) by self-training techniques, Cut-Paste [21] and Patch-SVDD [43] show this potential through designing proper pretext tasks for feature learning.”
A Unified Model for Multi-class Anomaly Detection (2022, background): “US [3] PSVDD [15] PaDiM [5] CutPaste [10] FCDD [11] MKD [12] DRAEM [16] Ours FLOPs(G) 60.”
DeSTSeg: Segmentation Guided Denoising Student-Teacher for Anomaly Detection (2022, background): “First, compared to painting a rectangle anomaly mask [17], the anomaly mask generated by random Perlin noise is more irregular and similar to actual anomalous shapes.”