Reproduce “Fine-grained Abnormality Prompt Learning for Zero-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
Jiawen Zhu, Yew-Soon Ong, Chunhua Shen, Guansong Pang
FAPrompt introduces a Zero-shot Anomaly Detection framework that learns fine-grained abnormality semantics by leveraging a Compound Abnormality Prompt (CAP) module and a Data-dependent Abnormality Prior (DAP) module. The framework achieves superior image-level and pixel-level anomaly detection across 19 industrial and medical datasets, with up to 3.9% AUROC and 4.2% AP improvements over existing methods.
Key insights
- Learning multiple, complementary abnormality prompts via a compound structure and orthogonal constraint is crucial for capturing diverse and fine-grained anomaly semantics.
- Dynamically adapting prompts with a data-dependent prior derived from the most anomalous image patches significantly enhances cross-dataset generalization.
- A dedicated abnormality prior learning loss is essential to ensure the prior is derived solely from genuinely abnormal features, preventing noise from normal samples.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.