Anomalib

Anomalib is an open-source deep learning library focused on visual anomaly detection, the task of finding and localizing defects or irregularities in images and video. First released on PyPI in December 2021, it is maintained under Intel's Open Edge Platform GitHub organization (formerly the OpenVINO Toolkit org) and ships a large collection of state-of-the-art anomaly detection algorithms alongside tools for experiment management, hyperparameter optimization, and edge inference.

Core Features

• State-of-the-art algorithm collection — implements a wide range of visual anomaly detection approaches in a single, benchmarkable codebase.
• Multiple inference backends — trained models export and run through Torch, Lightning, Gradio, or OpenVINO Intermediate Representation.
• Anomalib Studio — a low-code and no-code web application for training and deploying anomaly detection models without writing Python.
• Edge-ready export — built-in export to OpenVINO IR for accelerated inference on Intel CPUs, GPUs, and NPUs.

Tools and Resources

• Anomalib Documentation — anomalib.readthedocs.io, getting started guide and API reference.
• GitHub repository — open-edge-platform/anomalib, source code, examples, and release notes.
• Defect Detection with Anomalib and OpenVINO — Intel walkthrough tutorial.

Ecosystem and Integrations

• Exports directly to OpenVINO for accelerated inference on Intel hardware, including Intel Core Ultra processors.
• Used alongside Intel Physical AI Studio for visual anomaly detection in robotics defect-detection and response pipelines.

Start with the Anomalib documentation or the GitHub repository.