DepthAI is officially out of the release candidate stage and fully released as v3.0.0!
🚀 What’s New in v3.0.0
This release brings a major update to DepthAI with a simplified API, refreshed nodes, new high-level functionality, and improved developer tools. Here are the highlights:
Simplified API
The DepthAI API is now cleaner and more intuitive. Core tasks like setting up camera streams, running neural networks, and building spatial pipelines require fewer lines of code. Deprecated functions have been removed, leaving a more consistent interface for both newcomers and experienced users.
Refreshed Core Nodes
Camera: Color and Mono cameras are unified into a single Camera node, supporting multiple outputs with independent format, size, and frame rate.
ImageManip: Redesigned for predictability and robustness, now supports relative crops and can run on host or device.
New High-Level Nodes
RGBD: Easily combine synchronized color and depth data, with automatic setup for pointclouds.
VSLAM: Early preview of SLAM and VIO as host nodes, enabling advanced spatial AI use cases.
DynamicCalibration: Automatically compensates for thermal and mechanical stress to keep depth accurate.
Host Node Support
Parts of your pipeline can now run on the host with custom logic, while others run on-device. This makes it possible to integrate external libraries or do heavier post-processing in a unified pipeline.
Visualizer
Debugging and development just got easier. The new visualizer lets you see camera streams, neural network outputs, overlays, and even your pipeline graph in real time.
Quality-of-Life Improvements
Auto-reconnect for more resilient applications.
Standardized RDF coordinate system across all sensors.
Native Luxonis Model Zoo support for easy model deployment.
Improved detection remapping across pipeline images.
RVC4 Support
DepthAI v3 runs seamlessly on both RVC2 and RVC4 devices with a unified codebase. Prototype on one device, deploy on another — no code changes required.
🔄 Changes Since v3.0.0-rc4
New Features
- DynamicCalibration node with Python & C++ examples.
- Support for OAK-Thermal R8.
- Host implementation of ImageAlign for higher FPS.
- ObjectTracker supports ImgTransformations tracking.
- Added
depthai-core-example reference in README.
Bug Fixes & Stability
Fixed crashes on partially calibrated RVC2.
Resolved pink streams (OV9728 on OAK-D[-Pro]-Wide).
Fixed replay regression (invalid first frames).
Corrected IMU rotation vector on RVC2.
Fixed RVC4 calibration/startup issues.
Resolved ImageManip resize edge case on RVC4.
Misc Improvements
.setSensorType() on RVC4 Camera (quick color/mono switching).
Faster ToF decoding and examples now run at 30 FPS.
Combined Python wheels → smaller PyPi footprint.
Python bindings more responsive to interrupts.
Python utilities moved to repo root.
Added tests for CalibrationHandler.
CI test coverage for OAK4-S.
Full changelog in detail can be found on our depthai-core Release notes.
🖥️ OS Compatibility
Integration tested on Luxonis OS 1.6.0-debug, 1.14.1, and 1.15.0.
⚠️ Known Issue: On release flavors of OS for RVC4, cameras may fail to restart streaming.
→ We recommend using -debug versions of the OS until this is resolved.
📦 Installation
Upgrade or install the latest DepthAI release via pip:
python3 -m pip install depthai==3.0.0
Or simply:
python3 -m pip install depthai
(v3.0.0 installs by default)
🚀 Get Started
DepthAI v3.0.0 brings cutting-edge calibration, performance boosts, and an improved developer experience. Whether you’re building AI pipelines on OAK devices or experimenting with new features, this release provides a solid foundation for the future of spatial AI.
