DepthAI 3.9.0 Is Here
DepthAI 3.9.0 is now available, bringing a new home for experimental features, important reliability improvements for RVC2 PoE devices, better image processing and video encoding on RVC4, a new depth colorization utility, and a range of fixes across the platform.
This release contains plenty of technical improvements under the hood, but many of them translate into something simple: more reliable devices, more flexibility for developers, and easier access to new DepthAI capabilities.
A full changelog together with links to examples can be found on our DepthAI documentation page.
Here’s what’s new.
🧪 Introducing Beta Features
DepthAI 3.9.0 introduces a dedicated Beta features area.
Think of it as a staging area for experimental DepthAI nodes and features. It allows us to make new capabilities available sooner, so developers can try them before their public APIs are fully finalized.
Because these features are still in beta, their public APIs may change without prior notice.
There’s already quite a lot available:
- All Beta nodes can run natively on RVC4 and on the host when using RVC2.
- All parsers and unique messages previously available in depthai_nodes have been ported over and now have native support.
- All parsers expose runtime configuration changes through config messages, meaning their behavior can be adjusted while an application is running.
- Python and C++ examples are available to make getting started easier.
For developers, this means new DepthAI functionality can reach your hands faster while we continue refining it for a stable public API.
📷 Better Control Over RVC4 Image Undistortion
DepthAI 3.9.0 also improves image undistortion on RVC4.
The StereoDepth node now properly handles the alpha parameter when scaling undistorted images.
In simpler terms, this gives developers better control over the trade-off between cropping the image and keeping as much of the camera’s original field of view as possible.
Camera output requests now accept an optional alphaScaling value for controlling this behavior.
🌈 New Depth Colorization Utility
Visualizing raw depth information is now easier with a new built-in depth colorization utility.
It is available as:
- C++: dai::utility::colorizeDepthFrame
- Python: dai.utility.colorizeDepthFrame
The utility converts depth information into an easier-to-understand colored visualization.
It supports ImgFrame as well as OpenCV matrices, and developers can configure:
- the depth range to visualize,
- the color map,
- linear or logarithmic scaling.
It can also automatically determine useful visualization boundaries from the valid depth pixels in a frame.
Pixels containing invalid depth information can be rendered as black, making it easier to distinguish areas where no valid depth measurement is available.
🎥 Faster RVC4 Video Encoding
Video encoding on RVC4 gets several improvements in this release.
We’ve improved latency when encoding 4000 × 3000 video using:
We also fixed an issue where setKeyframeFrequency had no effect on RVC4. The setting now works as intended.
⚡ RVC4 480 FPS Mode Fix
DepthAI 3.9.0 fixes a regression introduced in 3.7.1 that affected the 480 FPS high-frame-rate mode on RVC4.
The problem was caused by the ImageManip output pool being too small to reliably handle frames at such high speeds.
ImageManip can now grow its output pool automatically when it starts running low on available frames.
For applications that need tighter control over memory usage, we’ve also added setMaxPoolSize, allowing developers to define the maximum size that the output pool can reach.
🔧 More Fixes and Improvements
DepthAI 3.9.0 includes several additional fixes across RVC2, RVC4, image processing, neural-network decoding, depth processing, and video replay.
RVC2 projector power protection: We added a total power limit for the DOT and FLOOD projectors to prevent power resets when using the RVC2 ToF 63D sensor.
SpatialLocationCalculator: On RVC2, SpatialLocationCalculator now correctly remaps regions of interest when its depth input is not aligned.
YOLOv6 decoding: We fixed decoding for YOLOv6 and YOLOv6-R1, including the reconstruction of bounding boxes based on stride values and the handling of confidence scores.
Rectification: Rectification now uses the transformations carried by its input frames. If those transformations change, rectification will reinitialize automatically so that it continues using the correct geometry.
ImageManip four-point transforms: We fixed an issue where the ImageManip four-point transform did not correctly account for normalized coordinates.
ToF examples: We fixed colorization flickering and inconsistent frame rates in the Time-of-Flight examples.
ReplayVideo: If FFmpeg cannot open a video, ReplayVideo now reports a proper runtime error instead of relying on an assertion. This makes failures clearer and easier to diagnose.
💻 RVC4 Luxonis OS Compatibility
DepthAI 3.9.0 integration has been tested with the following Luxonis OS releases:
- Luxonis OS 1.27.1
- Luxonis OS 1.30.1
- Luxonis OS 1.33.0
📦 Install DepthAI 3.9.0
Ready to try the release? Install it with pip:
python3 -m pip install depthai==3.9.0
