DepthAI Nodes

🌟 Overview

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.

📜 Table of Contents

• 🌟 Overview
• 🛠️ Installation
• 📦 Content
  • 📨 Message
  • 🧩 Node
  • Runtime
• 🤝 Contributing

🛠️ Installation

Install from PyPI:

pip install depthai-nodes

To use the box measurement nodes, install the box-measurement extra, which adds open3d:

pip install "depthai-nodes[box-measurement]"

On Debian or Ubuntu, open3d also needs system libraries that minimal images such as Docker containers and CI runners often lack:

sudo apt-get install libegl1 libgl1 libgomp1

open3d 0.20 publishes Linux wheels only for glibc 2.35 or newer, such as Debian 12+ or Ubuntu 22.04+. On older systems, including Debian 11, Ubuntu 20.04 and JetPack 5, pip cannot find a matching open3d and the install fails.

Or install from source:

git clone https://github.com/luxonis/depthai-nodes.git
cd depthai-nodes
pip install .

📦 Content

📨 Message

The message module provides Collection, GatheredData, and SnapData messages, box measurement messages such as BoxDetections, BoxMeasurements, and SupportPlane, 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.

🧩 Node

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 ParsingNeuralNetwork and ParserGenerator, create and connect inference and parser nodes.
• Utility nodes handle detection filtering, image overlays, colormaps, and message collection.
• Box measurement nodes detect, track, and measure boxes from color and a point cloud. BoxDetector and BoxMeasurer run on RVC4 only. BoxDimensioner requires open3d and has no RVC4-only check.

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.

Runtime

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")

🤝 Contributing

See CONTRIBUTING.md for development setup, parser guidelines, and testing instructions. Feedback and bug reports are welcome in GitHub issues.