DepthAI Nodes provides reusable nodes and helpers for DepthAI v3 pipelines, including neural network post-processing, image utilities, message handling, and runtime integrations. Inference helpers support both native DepthAI parsers and Python host parsers.
For complete Python API documentation, see the DepthAINodes documentation and API reference.
Install from PyPI:
pip install depthai-nodesOr install from source:
git clone https://github.com/luxonis/depthai-nodes.git
cd depthai-nodes
pip install .The message module provides Collection, GatheredData, and SnapData
messages, plus creator functions for native DepthAI parser messages. Creators
cover detections, segmentation, classification, keypoints, maps, and other model
outputs. See the message package documentation for the available
types and their roles.
The node module provides parsers and pipeline helpers:
- Parser nodes handle model post-processing for architectures such as YOLO, MediaPipe, and YuNet.
- Inference helpers, including
ParsingNeuralNetworkandParserGenerator, create and connect inference and parser nodes. - Utility nodes handle detection filtering, image overlays, colormaps, and message collection.
ParsingNeuralNetwork and ParserGenerator use native DepthAI parsers by default. Use HostParsingNeuralNetwork or pass hostOnly=True to ParserGenerator.build() to select this package's host parsers.
See the node package documentation for available nodes and examples.
The runtime module contains runtime integrations. Its OAK4 QNN helper
creates ONNX Runtime sessions on the Hexagon DSP when used with the
onnxruntime variant of the oakapp-base image:
from depthai_nodes.runtime import onnx_qnn_session
session = onnx_qnn_session("model.onnx")See CONTRIBUTING.md for development setup, parser guidelines, and testing instructions. Feedback and bug reports are welcome in GitHub issues.