Multimodal Task Representation Memory Bank vs. Catastrophic Forgetting in Anomaly Detection
This work addresses a significant problem in unsupervised anomaly detection for applications where data may be multimodal and continuously streaming, such as in industrial inspection or medical monitoring.
The authors tackled the problem of catastrophic forgetting in unsupervised continuous anomaly detection, achieving an average detection accuracy of 0.921 with their proposed Multimodal Task Representation Memory Bank method. This outperforms state-of-the-art methods and achieves the lowest forgetting rate.
Unsupervised Continuous Anomaly Detection (UCAD) faces significant challenges in multi-task representation learning, with existing methods suffering from incomplete representation and catastrophic forgetting. Unlike supervised models, unsupervised scenarios lack prior information, making it difficult to effectively distinguish redundant and complementary multimodal features. To address this, we propose the Multimodal Task Representation Memory Bank (MTRMB) method through two key technical innovations: A Key-Prompt-Multimodal Knowledge (KPMK) mechanism that uses concise key prompts to guide cross-modal feature interaction between BERT and ViT. Refined Structure-based Contrastive Learning (RSCL) leveraging Grounding DINO and SAM to generate precise segmentation masks, pulling features of the same structural region closer while pushing different structural regions apart. Experiments on MVtec AD and VisA datasets demonstrate MTRMB's superiority, achieving an average detection accuracy of 0.921 at the lowest forgetting rate, significantly outperforming state-of-the-art methods. We plan to open source on GitHub.