Introduction¶

Kenning provides an API for deploying deep learning applications on edge devices using various model training and compilation frameworks.

This documentation consists of the following chapters:

• Kenning section provides a project description, installation steps and a quick start guide,

• Deep Learning deployment stack section describes a typical model deployment flow on edge devices,

• Defining optimization pipelines in Kenning section describes a way to create advanced optimization scenarios with JSON config,

• Using Kenning via command-line arguments section describes executable scripts available in Kenning,

• Kenning gallery contains use cases with models optimization and integration with ROS2, Renode and Pipeline Manager,

• Kenning environment variables section describes specific variables which can influence how the Kenning works,

• Kenning measurements section describes data gathered during the compilation and evaluation process,

• Choosing optimal optimization pipeline section describes a way to run multiple pipelines and search for the one that achieves the best result

• Sample autogenerated report section provides a sample report generated using Kenning,

• Creating applications with Kenning section describes KenningFlow and its usage,

• Developing Kenning blocks section describes a way to develop new Kenning components,

• Kenning resources section describes how Kenning deals with its resources (pretrained models, datasets etc.),

• Kenning API section provides an in-depth description of the Kenning API.