探索全新的 mydSPACE 服务: 许可证管理、全新的 dSPACE 帮助以及用于用户文档的 dSPACE 聊天机器人。请访问我们的专用支持页面,获取更多信息和支持。 链接了解更多并获取支持

Turning neural networks into deterministic production code

Neural Net Coder simplifies the deployment of trained neural networks to embedded targets by generating deterministic, standard-compliant C code directly from ONNX models. The resulting production code is suitable for microcontrollers, enabling the integration of neural network functionality into embedded software without requiring dedicated neural network hardware support. Advanced optimization capabilities help reduce memory footprint through configurable post-training optimizations. This balances resource usage and prediction accuracy without retraining the network. To support verification and implementation decisions, Neural Net Coder combines back-to-back testing, configurable validation metrics, and graphical analysis capabilities in a streamlined workflow. In addition, host measurements provide early insights into expected RAM and ROM consumption, stack usage, and runtime characteristics, supporting early target evaluation and comparing optimization strategies before deployment.