SpireCV-Pro 1.0 Officially Released: Free x86, Browser Debugging, and Simulation Evaluation in One
Developers who have worked on vision for drones, unmanned vehicles, or robots have probably run into these situations:
· An algorithm works well on a PC, but has to be adapted again once it is deployed to an onboard computer.
· Everything looks normal in simulation, but the data pipeline has to be debugged again when a real camera is connected.
· A remote server has no desktop environment, so even viewing a single frame requires a stack of forwarding workarounds.
· Model metrics, real-time video, and raw data are scattered across different tools, and the full picture only appears after switching back and forth.
· When something goes wrong, it is hard to know where the chain is blocked, so teams often have to guess from experience.
SpireCV-Pro 1.0 is designed to solve exactly this kind of last-mile problem: moving from an algorithm to a working system.
This release brings one very direct change: the x86 platform is permanently free. In other words, you do not have to put all development onto an edge computing board from day one. You can first run the full chain on a lab PC, workstation, or server, then migrate it to Jetson, RK3588, Ascend, and other edge platforms.
01. What Is SpireCV-Pro?
SpireCV-Pro can be understood as an edge real-time perception development platform for intelligent unmanned systems.
It does not simply provide a detection model or a visualization demo. Instead, it breaks the everyday perception development workflow into nodes that can run independently, be debugged independently, and be combined flexibly:
· Data input: supports cameras, video files, image directories, COCO, nuScenes, LiDAR point clouds, and other sources.
· Algorithm inference: covers YOLOv5, YOLOv8, YOLOv11, YOLO-World, RT-DETR, MMDetection, MobileSAM, pose estimation, instance segmentation, oriented bounding-box detection, 3D point-cloud detection, and more.
· Target processing: includes single-object tracking, multi-object tracking, click tracking, target-size statistics, and related functions.
· Engineering output: provides result visualization, video saving, RTSP/RTMP streaming, ROS2 forwarding, and Foxglove/PlotJuggler visualization.
· Evaluation stack: supports standardized evaluation for detection, tracking, single-object tracking, and related tasks.
These capabilities rely on SpireMS as the communication foundation. SpireMS is a lightweight ROS-like publish-subscribe messaging system. It supports C++ and Python, works on Windows and Linux, and helps keep development methods relatively consistent across x86, Jetson, RK3588, and other platforms. On Linux, image messages can also use shared memory to reduce the extra overhead of large-image transmission.
Put more plainly, SpireCV-Pro is not about showing one algorithm result. It helps you build a complete perception chain covering data, algorithms, debugging, evaluation, and deployment.
02. Core Feature 1: x86 Is Permanently Free
Why does free x86 support matter?
For most perception algorithm development, the first step is not immediately flying a drone or driving a vehicle. The first step is to clarify questions such as:
· Whether video input is stable and smooth.
· Whether the model can detect targets and classify them accurately.
· Whether tracking IDs jump unexpectedly.
· Whether the visualization output matches expectations.
· Whether saved results can be replayed and reproduced.
· Whether ROS2 or another external system can receive the correct messages.
· Whether the full chain of model, data, post-processing parameters, and nodes can run end to end.
These tasks are well suited to an x86 PC or workstation. SpireCV-Pro 1.0 turns x86 into a free starting line: you can complete validation on a more familiar, easier-to-debug, and more powerful computer before moving the mature chain to the edge.
For university labs, robotics teams, drone teams, and algorithm engineers, this noticeably lowers the barrier to experimentation.
03. Core Feature 2: smsvt Browser Visualization Platform
One new experience in SpireCV-Pro 1.0 that is especially worth noting is smsvt.
It can be understood as a SpireMS visualization terminal running in the browser. After the service starts, it listens on 0.0.0.0:9092 by default. As long as the network is reachable, you can view video streams, replay results, and raw messages from a remote machine in the browser.
This is friendly for headless servers, remote development machines, and edge-device debugging. In many cases you do not need a remote desktop; opening a browser is enough to see how the system is running.
The Stream View page in smsvt automatically discovers image topics in SpireMS. Images can be placed in a single window, a 4-grid view, or a 9-grid view, and you can select different platforms, encoding methods, and display channels.
· Supports WebRTC, HLS, RTSP, RTMP, and other viewing methods.
· Uses WebRTC by default for lower-latency preview.
· Can connect to RTSP/RTMP streaming nodes for remote viewing of algorithm results.
· Can display raw images, detection visualization images, and tracking-result images at the same time.
· The page clearly shows which image topic is active, along with status, streaming, and viewing state, reducing guesswork.
You can run a detection node while viewing both the raw video and recognition results in the browser. For remote servers, headless industrial PCs, and edge boxes, this makes debugging much lighter.
Many algorithm problems are hard to locate through a quick real-time glance. For example, one frame may have a missed detection, a target ID may jump, or a flight or driving mission may need to be reviewed after it ends.
The Video Playback page in smsvt is designed for exactly these scenarios.
· It automatically scans video files under the Home directory and supports common formats such as .mp4, .mkv, and .avi. If a matching .svj file exists in the same directory, it also reads frame numbers, field of view, ROI, bounding boxes, categories, tracking IDs, and other result information.
· During playback, you see not only the video frame but also the boxes, center cross, IDs, categories, and other information output by the algorithm at that time. The page provides play, pause, fast forward, rewind, frame stepping, playback-speed control, and overlay toggles, making it suitable for troubleshooting reviews and result demonstrations.
· More importantly, it forms a closed loop with the saving node: images and results in SpireMS can be saved as video plus .svj files at runtime, then replayed later through the browser. An experiment, an inspection run, or a simulation round can be reopened, observed, and analyzed again at any time.
When debugging a system, the most frustrating problem is often not an error message, but not knowing whether a message was actually sent.
· The Raw Messages page in smsvt lists all message topics in SpireMS. After subscribing, you can directly view raw JSON content and message frequency. It is useful for quickly checking detection results, tracking results, camera parameters, Boolean control values, text logs, and many other message types.
· The right side of the page also provides message publishing. It supports common types such as std_msgs::Number, String, Boolean, and arrays. That means you can temporarily send a control message, test a switch value, or simulate a trigger directly in the browser without writing a separate script.
For engineering debugging, this is critical: it makes key questions such as whether the chain is connected, whether fields are correct, and whether frequency is normal immediately visible.
04. More Than a Tool: A Complete Development Chain
The value of SpireCV-Pro 1.0 is not limited to one node or one page. Its value is that it connects common development actions into a chain.
In SpireCV-Pro, smsrun is a core command-line entry point.
You can use it to view which nodes and launches are available on the current system, start a single node, start a group of nodes, view status, view logs, and stop or restart tasks. The node list in the source code is filtered according to architecture, operating system, CUDA, TensorRT, RKNN, CANN, and other capabilities, preventing unsupported tasks from being launched on the wrong platform.
For example, you can start with the simplest commands:
|
smsrun ls |
This is friendly to new users: instead of understanding every low-level parameter first, you can run a video detection and visualization chain, then gradually replace the input, model, visualization, and output nodes.
Real unmanned-system development has both risk and cost. The SpireCV-Pro wiki already provides multiple Gazebo scenario examples, including air-to-air target detection, road multi-object tracking, underground target detection, agricultural inspection, factory inspection, and ground strike.
The point of these scenarios is not merely that they look impressive. They allow you to validate the task workflow in simulation first:
· Whether the camera angle is suitable.
· Whether target detection is stable.
· Whether the control strategy behaves as expected.
· Whether data saving and playback can reproduce the process.
Once the chain is stable, switching to real equipment becomes much cheaper. The following scenarios cover common task types in unmanned-system perception development.
· Air-to-air target detection: simulates aerial target recognition and helps observe the detection stability of small and moving targets under complex viewpoints.
· Road multi-object tracking: continuously detects vehicles in road traffic scenes and preserves target IDs, making it suitable for validating multi-object tracking, visualization, and streaming chains.
· Underground target detection: targets scenarios with limited lighting, complex textures, and narrow spaces, and helps validate model robustness in non-ideal environments.
· Agricultural inspection: places perception algorithms in fields, crop, and fruit scenarios, making it suitable for agricultural robots and low-altitude inspection tasks.
· Factory inspection: focuses on indoor industrial scenarios such as assembly, production lines, and equipment inspection, and is suitable for validating target recognition and task workflows under stable viewpoints.
· Ground strike: places target discovery, continuous tracking, task triggering, and related steps into one simulation workflow, making it suitable for validating a more complete unmanned-system mission chain.
Algorithm quality cannot be judged only by whether the result looks acceptable.
SpireCV-Pro provides evaluation workflows for detection, multi-object tracking, single-object tracking, and related tasks. Detection tasks can output metrics such as mAP, AP50, and AP75. Multi-object tracking can be evaluated with HOTA, MOTA, IDF1, IDSW, and related metrics. Single-object tracking can be evaluated with Success AUC, Precision, and NormPrecision.
This is especially important for team collaboration. After changing the model, data, or post-processing parameters, everyone can compare results using the same standard rather than relying on a few screenshots.
SpireCV-Pro does not require you to abandon your existing robotics system.
· Through smsrt --ros2, SpireMS and ROS2 can perform bidirectional routing of standard messages. Images, camera parameters, IMU, point clouds, GPS, odometry, paths, TF, detection results, and other messages can be converted between SpireMS and ROS2.
· By default, ROS2-to-SpireMS uses the /ros2 prefix, while SpireMS-to-ROS2 uses the /sms prefix. Reverse prefixes are automatically avoided to reduce the risk of message loops.
This means you can use SpireCV-Pro as the perception-side development and deployment tool, then connect the results to an existing ROS2 system.
05. Who Is It For?
For students and researchers, it can reduce a large amount of engineering glue work. For engineering teams, it can put algorithm results and system deployment into the same workflow.
· Drone, unmanned-vehicle, and robotics perception development.
· Real-time detection, tracking, segmentation, and pose estimation on edge devices.
· Algorithm migration between Jetson, RK3588, and x86 workstations.
· Projects that need to connect AI perception results into ROS2.
· Teams that want a unified evaluation tool for collaboration.
· Scenarios where the complete chain should be validated in simulation before real-world deployment.
06. Fastest Way to Get Started
If installation is already complete, you can start with these steps:
|
# Start the SpireMS service |
If you need to connect to ROS2, you can also start:
|
smsrt --ros2 |
If you want to see results first, start with three things:
· Run a video detection and target visualization example with smsrun.
· Open smsvt in the browser to view streams, playback, and raw messages.
· Use the benchmark workflow to compare results with standard metrics.
After these three steps work, the design logic of SpireCV-Pro becomes easier to understand: it is not a wrapper around one model. It makes every link in the perception system visible, replaceable, measurable, and portable.
Closing Notes
The most important part of the SpireCV-Pro 1.0 release is not that it adds a few more algorithm nodes. It is that unmanned-system perception development is becoming easier to start, easier to debug, and easier to reproduce.
· The x86 platform is permanently free, allowing more developers to run it first on their own computers.
· The smsvt browser visualization terminal lets remote headless development show the field situation clearly in a browser.
· smsrun, Gazebo, Benchmark, and ROS2 routing connect the path from simulation, algorithms, visualization, and evaluation to system integration.
If you are working on drones, unmanned vehicles, or robots, or if you simply want to put a vision algorithm into a real-time system and run it properly, SpireCV-Pro 1.0 is worth opening the documentation and trying yourself.
Project Gitee address:
https://gitee.com/spirecv/spirecv-pro
Documentation:
https://spireai.top/wiki
