SpireCV-Pro Multi-Platform Visual Perception Software Upgrade for More Efficient AI Vision Implementation

In the current rapid development of intelligent unmanned systems, mobile robot visual perception is still constrained by core pain points such as “performance bottlenecks, cross-platform barriers, and low development efficiency.”

In October 2023, we released SpireCV 1.0. The English word “Spire” means the top or peak, and “CV” stands for computer vision. SpireCV 1.0 provides users with basic functions such as camera reading, pod control, video saving, and streaming, as well as perception algorithms like object detection, instance segmentation, single-object tracking, and multi-object tracking. Users can combine basic functions with perception algorithms to achieve their own visual perception tasks.

In November 2025, we officially launched SpireCV-Pro, which is characterized by the integration of our self-developed data distribution system SpireMS, achieving “cross-platform, cross-system, and cross-language” through node-based programming.

Compared to SpireCV 1.0, all functions of SpireCV-Pro are implemented in a node-based manner, allowing users to freely combine these nodes to achieve their own perception tasks. Nodes can be independently compiled and run. Nodes can operate across platforms, allowing usage on different hardware platforms; they can work across systems, enabling workflow combinations between different systems; and they can communicate across languages, allowing nodes implemented in different languages to interact.

SpireCV-Pro currently supports multiple hardware platforms, including X86 PC, Nvidia Jetson, Rockchip, etc.; it allows users to build their own perception task workflows between Windows and Linux systems; and it supports users in implementing their custom nodes using Python or C++.

Additionally, SpireCV-Pro connects with third-party systems such as ROS1, ROS2, PX4-Fast RTPS, etc.

SpireCV

AMOVLAB

1. Software Download

SpireCV-Pro:

https://www.wolai.com/iGncpTPWRZNcz7Y3gmUqXB

SpireMS:

https://www.wolai.com/duBWkKUCYQJCVSGPwueCrx

2. Source Code Repository

SpireCV-Pro Source Code:

https://gitee.com/spirecv/spirecv-pro.git

SpireMS Source Code:

https://gitee.com/spirecv/spirems.git

3. Quick Start Guide

SpireCV-Pro User Manual:

https://www.wolai.com/6fPCyNvcJemWuL7ygH5DXd

4. Project Community Group

SpireCV Visual Perception SDK Community Group, contact assistant Jiayue (jiayue199506) for group entry inquiries.

Selected as GITEE’s Most Valuable Open Source Project

Software Patent

Function Demonstration

AMOVLAB

3D Object Detection

Multi-Object Tracking

Underwater Image Enhancement

MobileSAM Segmentation

Keypoint Detection

Oriented Bounding Box Object Detection

Multi-Modal Fusion Detection

Software Architecture

AMOVLAB

Core Advantages of the New PRO Version

AMOVLAB

Equipped with the self-developed data distribution system SpireMS, it achieves node-based programming and “cross-platform, cross-system, and cross-language” capabilities.

Node-Based Programming

All functions of SpireCV-Pro are implemented in a node-based manner, allowing users to freely combine these nodes to achieve their own perception tasks. Nodes can be independently compiled and run. Nodes can operate across platforms, allowing usage on different hardware platforms; they can work across systems, enabling workflow combinations between different systems; and they can communicate across languages, allowing nodes implemented in different languages to interact.

Cross-Platform, Cross-System, Cross-Language

SpireCV-Pro currently supports multiple hardware platforms, including X86 PC, Nvidia Jetson, Rockchip RK series, etc.; it allows users to build their own perception task workflows between Windows and Linux systems; and it supports users in implementing their custom nodes using Python or C++.

Support for Third-Party System Connections

SpireCV-Pro also connects with third-party systems such as ROS1, ROS2, PX4-Fast RTPS, etc.

Integrated Basic Toolchain

SpireCV-Pro has integrated commonly used toolchain nodes for visual and radar perception, such as various interface camera reading nodes, video, image, point cloud file reading nodes, message diagnostic nodes, video streaming nodes, visualization nodes, etc., and works with Foxglove visualization software to display all message content.

Three Steps to Achieve a Visual Task

AMOVLAB

To lower the entry barrier, SpireCV-Pro has designed a “Capture-Process-Visualize” workflow. Taking the object detection task on the Rockchip platform as an example, it can be completed in just three steps.

Step 1: Acquire Data

Collect image data using “smsrun pcam” (camera reading node);;

Step 2: Perform Detection

Conduct object detection using “smsrun pyolo11rk” (Rockchip platform YOLO11 detector node);

Step 3: View Results

Visualize the detection results using “smsrun pdetvis”.

Note:

• You can use smsrun pcam -p to view the available parameters in the node, and follow with parameters to modify them during node execution.

• You can use sms ls to view all current topics.

Workflow Example

AMOVLAB

General Object Detection and Visualization

Scroll down to read more

1 Overview

Read video files or camera feeds, detect general/aerial targets (using COCO general object and VisDrone aerial target datasets), and visualize detection results.

2 Information Flow: Involved Nodes and Toolchain

2.1 On PC (x86_64) Platform

2.1.1 Read Video Files, Detection, Visualization

2.1.2 Read Camera, Detection, Visualization

2.2 On Rockchip (RK3588) Platform

2.2.1 Read Video Files, Detection, Visualization

2.2.2 Read Camera, Detection, Visualization

3 Startup Commands

3.1 On PC (x86_64) Platform

To read video files, the startup command is as follows:

smsrun viddet_yolo11_vis_cuda

To read real-time camera data, the startup command is as follows:

smsrun camdet_yolo11_vis_cuda

3.2 On Rockchip (RK3588) Platform

To read video files, the startup command is as follows:

smsrun viddet_yolo11_vis_rknn

To read real-time camera data, the startup command is as follows:

smsrun camdet_yolo11_vis_rknn

4 Effect Demonstration

Run smsrun viddet_yolo11_vis_cuda

Followed Follow Replay Share Like <!– Just take a look –> CloseWatch MoreMoreExit Full ScreenSwitch to Vertical Full ScreenExit Full ScreenAMOVLABFollowedShare VideoDuration00:13

0/0

00:00/00:13 Switch to Landscape Mode Continue PlayingProgress Bar, 0 percentPlay00:00/00:1300:13 SpeedFull Screen Speed playback at 0.5x 0.75x 1.0x 1.5x 2.0x Ultra Clear Smooth

Continue Watching

SpireCV-Pro Multi-Platform Visual Perception Software Upgrade for More Efficient AI Vision Implementation

Watch More Reprint, SpireCV-Pro Multi-Platform Visual Perception Software Upgrade for More Efficient AI Vision ImplementationAMOVLABFollowedShareLikeWowAdded to Top Stories Enter comment Video Details

Supported Datasets

AMOVLAB

1. nuscenes (Autonomous Driving Multi-Modal Object Detection): The world’s first publicly available dataset covering the full set of sensors for autonomous driving, it has become a benchmark in the field due to its multi-modal data, complex scene coverage, and fine annotations.

2. coco_detection (COCO General Object Detection): An internationally mainstream general object detection dataset covering various everyday objects and complex scenes, it is a core benchmark test set in the field of object detection due to its fine annotations and rich tasks.

3. voc_detection (VOC General Object Detection): An early classic general object detection dataset containing common object categories, it was one of the key benchmarks for comparing the performance of object detection algorithms.

4. visdrone2019_det (Aerial Remote Sensing Image Detection): A dataset for object detection from a drone’s perspective, covering ground targets in various scenes, suitable for low-altitude remote sensing detection tasks.

5. dota_v1.5 (Satellite Remote Sensing Image Detection): A dataset for object detection from a satellite’s perspective, containing large-scale ground object targets, supporting the detection needs of high-resolution remote sensing images.

6. air2air_det_db (Air-to-Air Detection): A dataset focused on object detection in air-to-air scenarios, targeting aerial vehicle object annotations, suitable for air-to-air perspective detection tasks.

7. air2air_det_mfb (Air-to-Air Detection): A dataset for object detection in air-to-air scenarios, supplementing aerial target data from different scenes to assist in the training and validation of related detection algorithms.

8. homeobjects_3k_det (Indoor Object Detection): A dataset for object detection in indoor scenes, covering various indoor items, suitable for smart home and indoor robot detection scenarios.

9. amovcar_det (AMOV Car Detection): A dataset for object detection targeting a specific small car platform, focusing on the annotation of surrounding environmental targets, suitable for perception tasks on that platform.

10. coco_pose (Human Pose Estimation): A subset of COCO for human pose estimation, containing annotations for human keypoints, it is a mainstream benchmark dataset in the field of human pose estimation.

11. hand_keypoints (Gesture Recognition/Keypoint Detection): A dataset focused on hand keypoint detection and gesture recognition, containing annotations for hand joints, suitable for gesture interaction scenario development.

12. carparts_seg (Vehicle Part Segmentation): A semantic segmentation dataset for vehicle components, with fine annotations for different vehicle parts, supporting vehicle fine-grained analysis tasks.

13. crack_seg (Crack Segmentation): A semantic segmentation dataset for cracks, focusing on the annotation of crack areas in various scenes, suitable for infrastructure defect detection tasks.

*More datasets will continue to be added, stay tuned!

Supporting Hardware and Serial Number Instructions

AMOVLAB

As an open-source project, the source code of SpireCV-Pro is fully public, and developers can freely access and develop it through the Gitee repository, while also supporting project contributions through PR submissions to co-build the open-source ecosystem. For commercial use and learning needs, the project provides flexible licensing options:

1. Supporting Hardware (Coming Soon)

SpireCV-TS1 Training Machine

SpireCV-3588 Domestic Intelligent Inference Board

. Serial Number Authorization

2. Serial Number Authorization

Supporting Hardware: Already pre-installed with a serial number, can be used directly.

Self-Installation: Requires separate purchase of serial number authorization, each 1000 yuan, contact AMOVLAB for purchase.

Short-Term Learning Testing: Can send an email to the SpireCV official email to obtain a serial number for free, with the following requirements.

• Official email address: [email protected]

• Send the following content:

🔹Your name, organization

🔹Your contact number

🔹Your device (Device) ID

🔹Your specific usage description

How to Contribute Open Source Code

AMOVLAB

Submit PR (Pull Request) to the open-source repository on Gitee to participate in project contributions, detailed steps can be found at:

https://www.wolai.com/cu5EyMMTmEaqUrYiz9jtWr