Filter or Compensate: Towards Invariant Representation from Distribution Shift for Anomaly Detection

Recent Anomaly Detection (AD) methods have achieved great success with\nIn-Distribution (ID) data. However, real-world data often exhibits distribution\nshift, causing huge performance decay on traditional AD methods. From this\nperspective, few previous work has explored AD with distribution shift, and the\ndistribution-invariant normality learning has been proposed based on the\nReverse Distillation (RD) framework. However, we observe the misalignment issue\nbetween the teacher and the student network that causes detection failure,\nthereby propose FiCo, Filter or Compensate, to address the distribution shift\nissue in AD. FiCo firstly compensates the distribution-specific information to\nreduce the misalignment between the teacher and student network via the\nDistribution-Specific Compensation (DiSCo) module, and secondly filters all\nabnormal information to capture distribution-invariant normality with the\nDistribution-Invariant Filter (DiIFi) module. Extensive experiments on three\ndifferent AD benchmarks demonstrate the effectiveness of FiCo, which\noutperforms all existing state-of-the-art (SOTA) methods, and even achieves\nbetter results on the ID scenario compared with RD-based methods. Our code is\navailable at https://github.com/znchen666/FiCo.\n