Real-IAD: Industrial Anomaly Detection with Multi-View Image Data

Real-IAD is an industrial anomaly detection dataset released in 2024 for research purposes. It provides a collection of multi-view images of industrial objects designed for benchmarking defect detection algorithms using a dedicated evaluation tool.

Use Cases

• Detecting manufacturing defects using the anomaly labels encoded in the sample keys
• Aggregating multi-angle perspectives for 3D anomaly localization using the C[0-9] camera indices
• Benchmarking industrial inspection models using the ADEval command-line interface

Strengths

• Includes the ADEval Python package for standardized performance metrics
• Multi-view coverage with up to 10 distinct camera angles (C0-C9) per object
• Standardized regex-based naming convention for automated parsing of anomaly types

Limitations

• Restricted to non-commercial research use under CC BY-NC-SA 4.0
• Requires specific .pkl result formatting for compatibility with the evaluation tool

Provenance

Source

Real-IAD Research Team

Freshness

Last updated October 2024.

Geography

United States