Reproduce “Dual-path Frequency Discriminators for Few-shot Anomaly Detection” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Yuhu Bai, Jiangning Zhang, Zhaofeng Chen, Yuhang Dong, Yunkang Cao, Guanzhong Tian

Leveraging frequency-domain analysis, a Dual-path Frequency Discriminators (DFD) network addresses few-shot anomaly detection by creating multi-level pseudo-anomalies. This approach delivers state-of-the-art pixel-level and image-level AUROC on MVTec AD and VisA benchmarks, alongside notable improvements in computational efficiency.

Key insights

  • Subtle anomalies, often hard to discern in the spatial domain, become more distinct and detectable when analyzed in the frequency domain.
  • Decomposing images into low- and high-frequency components provides complementary information, crucial for identifying various types of defects.
  • Generating pseudo-anomalies at both image and feature levels effectively creates diverse negative examples, allowing discriminative models to learn robust decision boundaries from minimal normal data.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.