This paper introduces a novel unsupervised anomaly detection framework for medical images, particularly effective for ultrasound imaging, using a diffusion model trained with synthetic anomaly noise and a multi-stage diffusion process for enhanced accuracy and detail preservation.
Dinomaly, a novel framework for multi-class unsupervised anomaly detection, achieves state-of-the-art performance by leveraging a simplified Transformer architecture with four key elements: foundation Transformers, a noisy bottleneck, linear attention, and loose reconstruction.
This survey paper provides a comprehensive overview of unsupervised anomaly detection methods in industrial settings, focusing on RGB, 3D, and multimodal approaches. It categorizes existing methods, discusses their strengths and weaknesses, and highlights future research directions.
Training unsupervised anomaly detection models on a carefully selected subset of prototypical in-distribution samples can outperform training on the entire dataset, challenging the assumption that more data always leads to better performance.
This paper introduces DMDD, a novel knowledge distillation-based method for unsupervised anomaly detection in images, which leverages a decoupled student-teacher network architecture and dual-modeling distillation to achieve state-of-the-art localization performance by effectively capturing both the edges and centers of anomalies.
Adapted-MoE addresses the challenges of feature distribution variations within the same category and distribution bias between training and test data by employing a Mixture of Experts model with a routing network and test-time adaptation.
The proposed Masked Diffusion Posterior Sampling (MDPS) method models the problem of normal image reconstruction as multiple diffusion posterior samplings based on a devised masked noisy observation model and a diffusion-based normal image prior under Bayesian framework, enabling robust normal image reconstruction and accurate anomaly localization.
A multi-feature reconstruction network using crossed-mask restoration is proposed to effectively detect anomalies in images without any labeled data.
Leveraging language models to provide detailed, understandable explanations for anomaly maps generated by unsupervised anomaly detection methods.
The core message of this paper is to propose a new anomaly detection model that fuses dictionary learning and one-class support vector machines (OC-SVM) to improve unsupervised anomaly detection performance.