關於 eiq auto 2.0 文件

I Have delivered the question to NXP AE ,and here is the replay:

The ONNX official doesn't provide model split tool like Tensorflow transform_graph tool, so we did it in eIQ Auto. 

you can do it by using airunner::LoadSubGraphFromONNX API as follow:

const std::string aGraphName3 = "data/airunner/yolov3-tiny.onnx";
std::vector<std::string> aInputName30 = {"input_1"};
std::vector<std::string> aOutputName30 = {"convolution_output1", "convolution_output"};
const std::string asubGraphName30 = "tiny_yolov3_subgraph1.onnx";
Status_t lStatus = LoadSubGraphFromONNX(aGraphName3, aInputName30, aOutputName30, asubGraphName30);

The tiny_yolov3_subgraph1.onnx will export to your board. Thanks

你好

想順便問說graph split 是eiq auto有提供特殊工具還是說模型自訓練時就要寫成兩個在各自轉成onnx當作graph split

謝謝