SpireCV 2.0 Launch: Cross-Platform, Multi-Language, Efficient Visual Perception

With the rapid development of intelligent unmanned systems, visual perception has entered a new stage of multi-scenario integration, large-scale expansion, and full-dimensional integration. The computing requirements are greater, the algorithm models are more complex, and the application fields are expanding. After a year of polishing, the visual perception open-source artifact SpireCV 2.0 is officially launched! Facing the industry wave of multi-scenario integration, in order to improve industry efficiency and empower users, we use the open-source model as the core and launch the open-source visual perception SDK to help users solve problems such as lack of development experience, long development cycles, high costs, and data confidentiality in visual perception tasks; we adhere to open-source collaboration to promote the rapid implementation of visual perception tasks.

SpireCV 2.0

In SpireCV 2.0, we introduce the software design pattern of node-based programming, by splitting the logic of the program into independent nodes, each node is responsible for completing a specific function. With the self-developed SpireMS messaging system, developers can build visual workflows like building blocks. Currently, it supports a variety of popular visual algorithms, video storage, streaming, streaming media reading, control protocols and other related nodes. Users can choose the corresponding nodes to form their own workflow according to their actual needs. Users can also develop their own function nodes according to their actual needs.

open-source resource express

SpireCV Wiki address:https://docs.amovlab.com/spire\_cv\_amov/#/
SpireCV source code address: Github：GitHub: https://github.com/amov-lab/SpireCV
Gitee: https://gitee.com/amovlab/SpireCV

SpireMS

SpireMS, also called Spire messaging system, is a ROS-like lightweight message publishing and subscription software package that supports image, radar and other sensor topics.

Why do SpireMS?

1. Cross-program data distribution is still difficult, and there are still various problems when installing ROS and ROS2 in different versions of the system Windows, Linux, program C++, Python, and Anaconda.

2. The data distribution system is not lightweight enough. We hope to install and deploy it with one click, but ROS and ROS2 often take half a day to deploy.

3. The data type of the data distribution system needs to be defined in advance (ROS message, IDL language of ROS2) and cannot be flexibly adapted.

4. ROS has a problem of excessive bandwidth usage, and nodes that are not subscribed still occupy bandwidth.

5. Programs that consume a lot of bandwidth, such as images and radar, require a shared memory mechanism.

Advantages of SpireMS

1. Perfectly adapted to Windows, Linux systems, and various versions of C++, Python, and Anaconda.

2. One-click deployment and installation, lightweight.

3. Basic data types can be predefined, and new types can also be flexibly adapted, which greatly facilitates multi-message synchronization.

4. There is a central mode with a built-in anti-crash mechanism, the communication is stable and reliable, and no subscribers occupy bandwidth.

5. Supported by default memory sharing mechanism, the cross-process image transmission performance is significantly better than ROS2 (Fast-DDS).

SpireCV 2.0 Advantages

1. Node programming to achieve one step ahead of others

Compared with SpireCV 1.0, the biggest feature of SpireCV 2.0 is that it is equipped with the self-developed data distribution system SpireMS, and all its functions are implemented in a node-based manner. Through node-based programming, "cross-platform, cross-system, and cross-language" are realized. Nodes are compiled and run independently. It can be cross-platform between nodes, and can be used between different hardware platforms; it can be cross-system, and workflows can be combined between different systems; it can be cross-language, and nodes implemented in different languages ​​can still communicate.

SpireCV 2.0 automatic attention object detection workflow (including pod control)

2. AI empowerment, automation of labeling training

With the "labeling-training" one-click pipeline, developers no longer have to stay up all night for data labeling, truly realizing "what you think is what you get."

SpireCV 2.0 target detection process diagram

🔹Supports new version of SAM2 automated annotation (no need to set up your own server)

🔹Supports new version of GroundingDINO automatic annotation (no need to set up your own server)

🔹Support YOLOv11 training management After users complete custom data annotation on SpireVIEW, they can use SpireVIEW's model training function, enter training instructions, and quickly train with one click.

3. Support shared memory, ultra-low latency image data transmission

The test performance is as follows:

🔹X86 platform, SpireMS shared memory transmits 4K resolution image delay 6 times faster than ROS2

🔹Jetson platform, SpireMS shared memory transfers 4K resolution images 75 times faster than ROS2

• SpireMS-JPG uses the default transmission method of sensor_msgs::CompressedImage type

• SpireMS-NVJPG uses sensor_msgs::CompressedImage type and turns on NV hardware accelerated transmission.

• SpireMS-ShareMem uses memory_msgs::RawImage type shared memory transfer

• The lower the latency, the better

🔹Multi-channel shared memory stress test(8 channels 1920×1080)X86 (5600G+RTX3090 platform) Native inter-process communication uses Shared memory mechanism, 8 processes send 8 channels of 1920×1080\@30Hzimage data, while another 8 processes receive and display these 8 images respectively.The comprehensive delay of the entire system is controlled within the frame, and the delay time does not exceed 30 milliseconds.

Welcome to try