English | 简体中文
NCNN is an AI visual inference acceleration toolkit based on a high-performance lightweight inference framework. We have integrated mainstream face detection models SCRFD and RetinaFace, as well as object detection YOLOv8. This product provides an out-of-the-box NCNN based on the Huawei Cloud EulerOS 2.0 64-bit system of Kunpeng servers.
- Extremely lightweight, no third-party dependency: Pure C++implementation, does not rely on any external libraries, has an extremely small compiled size, suitable for embedded devices, mobile devices, and resource constrained environments, truly achieving "zero dependency" deployment
- Optimized for mobile terminals and edge computing: Deeply adapted to ARM architecture, assembly level optimization is carried out for CPU instruction sets (such as NEON), significantly improving inference speed while balancing performance and power consumption
- High performance inference engine: Provide optimization strategies such as operator fusion, memory reuse, and multi-threaded parallelism to achieve millisecond level response while maintaining high precision, meeting high real-time requirements for application scenarios
The open-source image product NCNN Lightweight AI Visual Reasoning Engine provided by this project has pre-installed the the 20250503 version of NCNN and its related runtime environment, and provides deployment templates. Come and refer to the usage guide to easily start an efficient "out-of-the-box" experience!
System requirements are as follows:
- CPU: 2vCPUs or higher
- RAM: 4GB or larger
- Disk: At least 40GB
Register a Huawei account and activate Huawei Cloud
| Image Specification | Feature Description | Remarks |
|---|---|---|
| NCNN-20250503-kunpeng | Installed and deployed based on Kunpeng servers + Huawei Cloud EulerOS 2.0 64-bit |
- For more questions, you can contact us through issues or the service support of the specified product in the Huawei Cloud Marketplace.
- For other open-source images, please refer to open-source-image-repos.
- Fork this repository and submit a merge request.
- Synchronously update README.md based on your open-source image information.