Reproduce “A Hierarchical Transformation-Discriminating Generative Model 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
Shelly Sheynin, Sagie Benaim, Lior Wolf
This paper introduces a method for few-shot anomaly detection that leverages a hierarchical generative model integrated with self-supervised transformation discrimination. The approach effectively defines normality and detects anomalies using as few as one training image, consistently outperforming existing baselines across various datasets and demonstrating robust defect localization capabilities.
Key insights
- Even a single natural image contains rich internal statistical regularities across patches and scales, which can be modeled for anomaly detection.
- Self-supervised transformation tasks provide a strong supervisory signal for discriminators, enabling them to learn robust features of normal data from limited examples.
- The synergistic combination of hierarchical multi-scale generation, transformation discrimination, and a generative component is essential for effective few-shot anomaly detection.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.