DepthAI 3.10.0 is now available, bringing several improvements for multi-camera applications, spatial data processing, debugging, and RVC4 performance.
This release introduces a beta Stitching node, a flexible RVC4 Align node, message save/load APIs, improved ToF support, and a number of stability and performance fixes across the platform.
Stitching node arrives in beta
The new Stitching node can combine frames from two different sensors by identifying matching visual features in the overlapping areas of the images.
It supports two stitching modes:
- Panorama stitching, which creates a wider view from overlapping images.
- Planar projection, which projects image data onto a configurable plane.
This makes the node useful for applications such as wide-angle imaging, multi-camera visualization, robotics, and scene reconstruction.
Developers can explore the new functionality with the panorama stitching example and the planar projection example.
Because the node is currently in beta, behavior and APIs may continue to evolve in future releases.
More flexible alignment on RVC4
DepthAI 3.10.0 also introduces the RVC4 Align node, which provides a generic way to align Transformable messages and ImgFrame objects.
Unlike a fixed alignment pipeline, the node can align messages in different orders and can transform an image frame to any supported transformable message type. This is particularly useful when combining camera data with detections, depth information, or other spatial outputs.
The node also supports custom messages when they implement the transformTo function and the node is running on the host. Examples are available for both Python and C++.
For detection workflows, see the Python ImgDetections alignment example and the corresponding C++ example.
Save and reload DepthAI messages
A new message serialization API makes it easier to record and replay data during development and testing.
Supported messages can now be saved to a binary .dai file and loaded later:
msg.save("frame.dai")
msg.load("frame.dai")
The file uses a binary protobuf envelope and supports common message types including image frames, neural network data, detections, segmentation masks, point clouds, RGB-D data, IMU data, and image annotations.
This can simplify debugging by allowing developers to capture data from a live pipeline and reproduce processing behavior without needing to reconnect a device.
Improved RVC4 ToF support
For RVC4 time-of-flight workflows, setOutputUndistortion can now be used to undistort the ToF depth frame.
The release also improves extrinsics handling for multi-device setups. Extrinsics now include a toDeviceId field, which works together with toCameraSocket to make transformations more reliable when messages are associated with different devices.
Performance and reliability improvements
DepthAI 3.10.0 includes several fixes designed to make pipelines more stable and predictable.
On RVC4, neural network model loading has been optimized and the RAM footprint has been reduced by approximately half. Device discovery functions now release the Python GIL, helping prevent stuttering in applications that discover devices while a pipeline is running.
Other fixes improve normalized-coordinate transforms, rotated crops, camera intrinsics validation, device health checks, video encoder parameter validation, and detection parser behavior.
The embedded visualizer has also been updated to version 3.9.6, bringing performance and stability improvements.
RVC4 compatibility
This release has been integration tested with Luxonis OS versions:
There is one important downgrade issue to be aware of: downgrading from OS 1.40.0 directly to an OS version between 1.24.0 and 1.31.1 can cause an STM flash error and device crash. If this is required, downgrade to OS 1.35.0 first.
Install DepthAI 3.10.0
The Python package can be installed with:
python3 -m pip install depthai==3.10.0
For the complete list of changes, see the DepthAI 3.10.0 release notes.
With new multi-camera processing tools, more flexible message alignment, easier data replay, and significant RVC4 memory improvements, DepthAI 3.10.0 is a strong update for developers building spatial AI and computer vision applications.
