如何打印模型？
// print register_module
// auto Tiny_Net = std::make_shared<VGG9>();
// print_modules(Tiny_Net)
void print_modules(const std::shared_ptr<torch::nn::Module> &module, size_t level = 0) {
auto tabs = [&](size_t num) {
for (size_t i = 0; i < num; i++) {
std::cout << "\t";
}
};
std::cout << module->name() << " (\n";
for (const auto& parameter : module->named_parameters()) {
tabs(level + 1);
std::cout << parameter.key() << '\t';
std::cout << parameter.value().sizes() << '\n';
}
tabs(level);
std::cout << ")\n";
}
//输入32x32 3通道图片
auto input = torch::rand({ 1,3,32,32 });
//输出
auto output_bilinear = torch::upsample_bilinear2d(input, { 8,8 }, false);
auto output_nearest = torch::upsample_nearest2d(input, { 5,5 });
auto output_avg = torch::adaptive_avg_pool2d(input, { 3,9 });
std::cout << output_bilinear << std::endl;
std::cout << output_nearest << std::endl;
std::cout << output_avg << std::endl;
libtorch 加载 pytorch 模块进行预测示例
void mat2tensor(const char * path, torch::Tensor &output)
{
//读取图片
cv::Mat img = cv::imread(path);
if (img.empty()) {
printf("load image failed!");
system("pause");
}
//调整大小
cv::resize(img, img, { 224,224 });
cv::cvtColor(img, img, cv::COLOR_BGR2RGB);
//浮点
img.convertTo(img, CV_32F, 1.0 / 255.0);
torch::TensorOptions option(torch::kFloat32);
auto img_tensor = torch::from_blob(img.data, { 1,img.rows,img.cols,img.channels() }, option);// opencv H x W x C torch C x H x W
img_tensor = img_tensor.permute({ 0,3,1,2 });// 调整 opencv 矩阵的维度使其和 torch 维度一致
//均值归一化
img_tensor[0][0] = img_tensor[0][0].sub_(0.485).div_(0.229);
img_tensor[0][1] = img_tensor[0][1].sub_(0.456).div_(0.224);
img_tensor[0][2] = img_tensor[0][2].sub_(0.406).div_(0.225);
output = img_tensor.clone();
}
int main()
{
torch::Tensor dog;
mat2tensor("dog.png", dog);
// Load model.
std::shared_ptr<torch::jit::script::Module> module = torch::jit::load("model.pt");
assert(module != nullptr);
std::cout << "ok\r\n" << std::endl;
// Create a vector of inputs.
std::vector<torch::jit::IValue> inputs;
torch::Tensor tensor = torch::rand({ 1, 3, 224, 224 });
inputs.push_back(dog);
// Execute the model and turn its output into a tensor.
at::Tensor output = module->forward(inputs).toTensor();
//加载标签文件
std::string label_file = "synset_words.txt";
std::fstream fs(label_file, std::ios::in);
if (!fs.is_open()) {
printf("label open failed!\r\n");
system("pause");
}
std::string line;
std::vector<std::string> labels;
while (std::getline(fs,line))
{
labels.push_back(line);
}
//排序 {1,1000} 矩阵取前10个元素(预测值),返回一个矩阵和一个矩阵的下标索引
std::tuple<torch::Tensor,torch::Tensor> result = output.topk(10, -1);
auto top_scores = std::get<0>(result).view(-1);//{1,10} 变成 {10}
auto top_idxs = std::get<1>(result).view(-1);
std::cout << top_scores << "\r\n" << top_idxs << std::endl;
//打印结果
for (int i = 0; i < 10; ++i) {
std::cout << "score: " << top_scores[i].item().toFloat() << "\t" << "label: " << labels[top_idxs[i].item().toInt()] << std::endl;
}
system("pause");
return 0;
]
torch::sort
torch::Tensor x = torch::rand({ 3,3 });
std::cout << x << std::endl;
//排序操作 true 大到小排序，false 小到大排序
auto out = x.sort(-1, true);
std::cout << std::get<0>(out) << "\r\n" << std::get<1>(out) << std::endl;
输出如下：
0.0855 0.4925 0.4323
0.8314 0.8954 0.0709
0.0996 0.3108 0.6845
[ Variable[CPUFloatType]{3,3} ]
0.4925 0.4323 0.0855
0.8954 0.8314 0.0709
0.6845 0.3108 0.0996
[ Variable[CPUFloatType]{3,3} ]
1 2 0
1 0 2
2 1 0
[ Variable[CPULongType]{3,3} ]