Curating the chaos of industrial anomaly detection research

What it does This repository is a curated bibliography of academic papers, public datasets, and open-source code focused on industrial image anomaly and defect detection. It organizes research into a detailed taxonomy — from classic Teacher-Student and Memory-bank methods to newer diffusion models and multimodal large language models — and tracks recent publications by conference venue. The maintainers, M-3LAB, also link to their own related survey papers and the IM-IAD benchmark.

The interesting bit Instead of dumping links, the repo builds a “Paper Tree” that classifies representative methods by their underlying mechanics, making it easier to compare approaches like normalizing flows against reconstruction-based networks at a glance. It also stretches beyond pure 2D RGB into 3D, logical, and continual anomaly detection — reflecting where the field is actually heading.

Key highlights

• SOTA methods table with direct links to papers and GitHub repos, tagged by topic (e.g., Teacher-Student, Memory-bank, Diffusion Model).
• Granular taxonomy covering unsupervised, supervised, few-shot, zero-shot, noisy, and MLLM-based anomaly detection.
• Links to companion resources: the maintainers’ own survey paper, the IM-IAD benchmark, and spin-off lists for 3D anomaly detection and anomaly synthesis.
• Dataset section with BibTeX citations for reproducibility.

Verdict Worth bookmarking if you are building or researching visual inspection pipelines and need a map of what has actually been published — with code. Skip it if you are hunting for a single pip-installable framework; this is a reading list, not a runtime.