Search papers, labs, and topics across Lattice.
This paper introduces a Foundation Model-based Anomaly Synthesis (FMAS) pipeline to generate realistic anomalous samples for training anomaly detection models, addressing the challenge of scarce anomalous data. They further propose a Wavelet Domain Attention Module (WDAM) that leverages frequency-domain characteristics to improve anomaly feature extraction. Experiments on MVTec AD and VisA datasets demonstrate that the WDAM module significantly improves anomaly detection performance when combined with FMAS.
Synthesizing training data with foundation models and attending to wavelet domains can dramatically boost anomaly detection, even without fine-tuning or class-specific training.
Industrial anomaly detection faces significant challenges due to the scarcity of anomalous samples and the complexity of real-world anomalies. In this paper, we propose a foundation model-based anomaly synthesis pipeline (FMAS) that generates highly realistic anomalous samples without fine-tuning or class-specific training. Motivated by the distinct frequency-domain characteristics of anomalies, we introduce aWavelet Domain Attention Module (WDAM), which exploits adaptive sub-band processing to enhance anomaly feature extraction. The combination of FMAS and WDAM significantly improves anomaly detection sensitivity while maintaining computational efficiency. Comprehensive experiments on MVTec AD and VisA datasets demonstrate that WDAM, as a plug-and-play module, achieves substantial performance gains against existing baselines.
Tsinghua AI