Google Research

MediaPipe: A Framework for Building Perception Pipelines

MediaPipe functions as a versatile, cross-platform framework engineered to streamline the creation and deployment of sophisticated perception pipelines.

• MediaPipe is a cross-platform framework that uses a modular, directed-graph architecture to build and deploy real-time perception pipelines like object detection and face landmark estimation.
• The framework utilizes a deterministic synchronization system based on settled timestamps to manage the streaming of time-series data across diverse hardware environments.
• Performance is optimized through cross-platform GPU acceleration that uses opaque buffer types and automated synchronization to minimize CPU-GPU overhead.