AI Inference SDK Integration Guide

An onboarding guide and model conversion reference for running AI inference on embedded devices.

Supported Frameworks

• TensorFlow Lite → Rockchip NPU (RV1126B)
• ONNX Runtime → DRP-AI (RZ/V2H)
• OpenCV DNN → CPU fallback

Model Conversion Flow

1. Export the trained model (TensorFlow / PyTorch)
2. Optimize model size via quantization (INT8/FP16)
3. Compile for NPU/DRP-AI target
4. Load the model via the SDK API and run inference

API Reference

int csun_ai_load_model(const char *model_path, csun_model_handle_t *handle);
int csun_ai_infer(csun_model_handle_t *handle, csun_tensor_t *input, csun_tensor_t *output);

Supported Models

• Object Detection: YOLOv5/v8, MobileNet SSD
• Classification: ResNet, EfficientNet
• Segmentation: DeepLabV3