关于mxnet的cpp的整理

MXNet/Gluon 中网络和参数的存取方式

mxnet-ssd

get a image // resize image image.resize(height_, width_); int size = image.channels() * image.cols() * image.rows(); std::vector<float> in_data(size); // de-interleave and minus means unsigned char *ptr = image.ptr(); float *data_ptr = in_data.data(); for (int i = 0; i < size; i +=3) { *(data_ptr++) = static_cast<float>(ptr[i]) - mean_r_; } for (int i = 1; i < size; i +=3) { *(data_ptr++) = static_cast<float>(ptr[i]) - mean_g_; } for (int i = 2; i < size; i +=3) { *(data_ptr++) = static_cast<float>(ptr[i]) - mean_b_; }

feature_extract

feature_extract.cpp WaitAll is need when we copy data between GPU and the main memory // 1 map<string, NDArray> paramters; NDArray::Load("./model/Inception-BN-0126.params", 0, ¶mters); ..... NDArray::WaitAll(); // 2 mean_img.SyncCopyFromCPU( NDArray::LoadToMap("./model/mean_224.nd")["mean_img"].GetData(),1 * 3 * 224 * 224); NDArray::WaitAll(); // 3 auto array = executor->outputs[0].Copy(Context(kCPU, 0)); NDArray::WaitAll();

image-classification/predict-cpp/