pseudo-anomaly synthesis

This work addresses the challenges of scarce anomaly samples and long-tailed distributions in industrial 3D anomaly detection. The authors propose Synthesis4AD, an end-to-end framework that pioneers the use of multimodal large language models to interpret design specifications and generate executable instructions for synthesizing anomalies. Coupled with MPAS—a primitive-guided, high-dimensional controllable synthesis engine—it enables large-scale generation of geometrically realistic anomalies with precise pixel-level masks. The framework further enhances training of Point Transformer detectors through spatial distribution normalization and geometry-preserving data augmentation strategies. Extensive experiments demonstrate state-of-the-art performance on Real3D-AD, MulSen-AD, and real-world industrial part datasets. Both the MPAS synthesis method and the 3D-DefectStudio platform will be publicly released.

This work addresses the challenge of 3D point cloud anomaly detection, where the scarcity and diversity of anomalous samples typically restrict training to normal data alone, thereby limiting model generalization. To overcome this, the authors propose a modular framework that enhances unsupervised training by synthesizing diverse pseudo-anomalies. Specifically, they construct a parametric deformation model based on local PCA coordinate systems, enabling anisotropic, direction-gated, and normal/tangential displacement fields to generate a rich variety of geometric defects. The approach is highly versatile, compatible with both reconstruction- and offset-prediction-based detection paradigms. Experiments on AnomalyShapeNet and Real3D-AD demonstrate significant improvements in both object-level and point-level anomaly detection and localization performance, while ablation studies confirm the effectiveness of individual components and robustness to noise.

This work addresses the neglect of mechanical causality in 3D anomaly detection by proposing the first defect-force-source-based detection paradigm. Methodologically, it formalizes 3D anomalies as outcomes of internal/external defect forces and introduces MC4AD—a mechanics-complementary framework comprising DA-Gen (anomaly generation) and CFP-Net (corrective force prediction), augmented by a symmetry loss, three-way decision theory, and the Anomaly-IntraVariance benchmark for intra-class variance awareness. Contributions include: (1) the first fine-grained anomaly modeling grounded in mechanical principles; (2) unified detection and causal attribution via interpretable corrective force prediction; and (3) a three-stage quality control strategy enhancing robustness. The method achieves nine state-of-the-art results across six benchmarks, with the fewest parameters and fastest inference speed. Code is publicly available.

Existing one-class anomaly detection methods suffer from poor generalization due to geometric bias in benchmark datasets, while prevailing pseudo-anomaly generation techniques fail to faithfully model the intrinsic structure of normal data and over-rely on image-domain operations. Method: This paper proposes a novel paradigm that synthesizes pseudo-anomalies exclusively within a frozen pre-trained feature space—bypassing image-level augmentation entirely. Contribution/Results: Key innovations include (1) an adaptive linear feature perturbation mechanism that dynamically tailors noise distribution per sample, and (2) a contrastive learning objective explicitly decoupling geometric bias from semantic anomaly modeling. Evaluated on both standard and geometric-bias-mitigated benchmarks, our method consistently outperforms state-of-the-art approaches, demonstrating superior generalization and robustness. The implementation is publicly available.

Existing anomaly detection models suffer from training bias and degraded generalization due to the lack of a unified standard for anomaly modeling. This paper addresses the limitation of reconstruction-based methods—specifically, their failure to account for inter-class discrepancies in anomaly characteristics during data augmentation—by proposing the first composable augmentation framework tailored for reconstruction models. We first systematically identify key factors by which synthetic anomalies influence reconstruction training. Then, we design a class-aware augmentation composition mechanism coupled with a decoupled, multi-stage training strategy comprising feature-space perturbation, class-conditional augmentation selection, and parameter freezing. On MVTec-AD, our method significantly outperforms state-of-the-art approaches, especially in object-level anomaly detection. Moreover, on a newly constructed multi-characteristic synthetic anomaly benchmark, it demonstrates superior cross-class generalization capability.

This study addresses the scarcity of real defects and the inefficiency and limited diversity of existing synthesis methods in industrial anomaly detection. We propose a "generate once, synthesize multiple times" paradigm that introduces a novel decoupling mechanism for defect extraction and synthesis. By integrating vision-language model guidance with image generation techniques, our approach achieves precise defect localization and seamless blending through object boundary suppression and multi-resolution spectral pyramid noise, enabling rapid construction of large-scale datasets via patch reuse. Evaluated on MVTec AD 2, the method attains an F1 score of 78.1%, approaching the upper bound of real-data performance, while accelerating synthesis by over 11.95×. Furthermore, it significantly improves calibration-transfer consistency, demonstrating its effectiveness for scalable and high-fidelity industrial anomaly detection.

This work addresses the challenge of detecting unknown-type defects in 3D point clouds from industrial manufacturing scenarios where only limited normal samples and a few known anomalies are available. To this end, we propose Open3D-AD, an open-set supervised 3D anomaly detection framework that jointly leverages normal data, synthetic anomalies, and a subset of real anomalies. Our approach models the probability density distributions of both positive (normal) and negative (anomalous) classes and introduces a distribution-aware subsampling strategy to enhance discriminability, complemented by a point-cloud-specific distribution disentanglement mechanism. We further contribute Open-Industry, the first high-quality dataset encompassing 15 categories of industrial products with five realistic defect types per category. Extensive experiments across multiple benchmarks demonstrate the effectiveness of our method, significantly advancing the state of the art in open-set 3D anomaly detection.

This work addresses the challenge of 3D point cloud anomaly detection, where the scarcity of real anomalous samples and high annotation costs hinder effective learning of discriminative features. To overcome this limitation, the authors propose the PA3AD framework, which generates diverse and plausible pseudo-anomalies from normal data through multi-physical modeling. The framework further incorporates a momentum-updated prototype mechanism and a discrepancy-aware fusion module to jointly guide the model in capturing distributional shifts between normal and anomalous patterns. By integrating these components with weight-shared 3D representation learning, PA3AD achieves state-of-the-art performance on both Anomaly-ShapeNet and Real3D-AD benchmarks, significantly outperforming existing methods.

This study addresses the challenge of industrial anomaly detection, where real anomalous samples are scarce and existing synthesis methods struggle to generate semantically realistic and diverse anomalies. To overcome this limitation, the work introduces an embodied agent paradigm for anomaly synthesis, proposing a novel anomaly synthesis agent endowed with self-reflection, knowledge retrieval, and iterative optimization capabilities. The agent leverages structured trajectories and a triple-reward mechanism to enable closed-loop refinement, augmented by tool-enhanced reinforcement learning within a two-stage training framework. Evaluated on MVTec-AD, the method achieves an Inception Score (IS) of 2.10 and Improved Consistency Loss (IC-L) of 0.33, attains a ResNet34 classification accuracy of 57.0%, and yields UNet image- and pixel-level average precisions of 99.3% and 74.2%, respectively—significantly outperforming current zero-shot state-of-the-art approaches.

This study addresses the limitation of existing industrial point cloud inspection methods that overlook object-level design or assembly rule violations by focusing solely on local geometric deviations. To this end, it formally defines logical anomalies and introduces the ILGAD benchmark dataset. Methodologically, a consistency reasoning framework is proposed, which integrates point-level annotations with multimodal feature analysis to detect anomalies by evaluating geometric morphology, structural coverage, and spatial relationships. Experimental results demonstrate that the proposed approach achieves superior performance in both object-level detection and point-level localization across multiple datasets. Furthermore, it effectively identifies logical anomalies while exhibiting strong generalization capability to conventional geometric defects.