libtorch部署的常用操作,Mat与Tensor转换，多输出、多输出

通过幸福的路唯有奋斗

已于 2022-08-06 09:54:15 修改

阅读量1.1k

点赞数

分类专栏：深度学习文章标签：深度学习人工智能 libtorch c++ opencv

于 2022-07-23 15:20:12 首次发布

本文链接：https://blog.csdn.net/X_windy/article/details/125948020

版权

深度学习专栏收录该内容

15 篇文章 0 订阅

订阅专栏

1.多输入

// 多数入, input1 和 input2已经是tensor std::vector<torch::jit::IValue> inputs; 
inputs.push_back(input1); 
inputs.push_back(input2); 
torch::Tensor result = net_head.forward(inputs).toTensor();

2.MAT转Tensor

# 输入mat数据为z_crop, tensor为tensor_image_T 
torch::Tensor tensor_image_T = torch::from_blob(z_crop.data, {1,z_crop.rows, z_crop.cols,3},torch::kByte); 

tensor_image_T = tensor_image_T.permute({0,3,1,2}); 
tensor_image_T = tensor_image_T.toType(torch::kFloat);

3.Tensor转MAT

cv::Mat output_mat(cv::Size{ height, width }, CV_8UC3, tensor.data_ptr<uchar>());

4.Load模型

torch::jit::script::Module module = torch::jit::load(T_model_backbone);

5.多输出

// 2个输出
auto result = net_head.forward(inputs).toTuple();
torch::Tensor cls_score = result->elements()[0].toTensor();
torch::Tensor bbox_pred = result->elements()[1].toTensor();