arXiv 2505.066032025-05-10cs.CV

ReplayCAD：用于持续异常检测的生成式扩散回放

ReplayCAD: Generative Diffusion Replay for Continual Anomaly Detection

South China University of Technology（华南理工大学）；China United Network Communications Corporation Limited Guangdong Branch（中国联合网络通信集团有限公司广东分公司）；Pazhou Laboratory（琶洲实验室）；Guangdong University of Technology（广东工业大学）

Lei Hu, Zhiyong Gan, Ling Deng, Jinglin Liang, Lingyu Liang, Shuangping Huang, Tianshui Chen

备注：: Accepted by IJCAI 2025

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

针对持续异常检测中的灾难性遗忘与小异常区域分割难题，提出ReplayCAD扩散生成回放框架，利用语义与空间特征压缩并回放多样化像素级历史数据，在VisA和MVTec上取得显著分割提升。

摘要：

持续异常检测（Continual Anomaly Detection，CAD）使异常检测模型能够在学习新类别的同时保留对历史类别的知识。CAD面临两个关键挑战：灾难性遗忘和小型异常区域的分割。现有CAD方法通过存储图像分布或图像块特征来缓解灾难性遗忘，但它们无法保留像素级细节特征以实现准确分割。为克服这一局限，我们提出ReplayCAD，这是一种新颖的扩散驱动生成回放框架，可回放高质量历史数据，从而有效保留像素级细节特征。具体而言，我们通过在预训练扩散模型的条件空间中搜索类别语义嵌入来压缩历史数据，该嵌入能够引导模型回放具有细粒度像素细节的数据，进而提升分割性能。然而，仅依赖语义特征会导致有限的空间多样性。因此，我们进一步使用空间特征引导数据压缩，实现对样本空间的精确控制，从而生成更加多样化的数据。我们的方法在分类和分割上均达到最先进性能，其中分割提升尤为显著：在VisA上提升11.5%，在MVTec上提升8.1%。源代码可在https://github.com/HULEI7/ReplayCAD获取。

英文摘要：

Continual Anomaly Detection (CAD) enables anomaly detection models in learning new classes while preserving knowledge of historical classes. CAD faces two key challenges: catastrophic forgetting and segmentation of small anomalous regions. Existing CAD methods store image distributions or patch features to mitigate catastrophic forgetting, but they fail to preserve pixel-level detailed features for accurate segmentation. To overcome this limitation, we propose ReplayCAD, a novel diffusion-driven generative replay framework that replay high-quality historical data, thus effectively preserving pixel-level detailed features. Specifically, we compress historical data by searching for a class semantic embedding in the conditional space of the pre-trained diffusion model, which can guide the model to replay data with fine-grained pixel details, thus improving the segmentation performance. However, relying solely on semantic features results in limited spatial diversity. Hence, we further use spatial features to guide data compression, achieving precise control of sample space, thereby generating more diverse data. Our method achieves state-of-the-art performance in both classification and segmentation, with notable improvements in segmentation: 11.5% on VisA and 8.1% on MVTec. Our source code is available at https://github.com/HULEI7/ReplayCAD.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。