libtorch 哪些函數比較經常使用？

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";
}

//輸入32x32 3通道圖片
		auto input = torch::rand({ 1,3,32,32 });

		//輸出
		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 加載 pytorch 模塊進行預測示例

void mat2tensor(const char * path, torch::Tensor &output)
{
	//讀取圖片
	cv::Mat img = cv::imread(path);
	if (img.empty()) {
		printf("load image failed!");
		system("pause");
	}

	//調整大小
	cv::resize(img, img, { 224,224 });
	cv::cvtColor(img, img, cv::COLOR_BGR2RGB);
	//浮點
	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 });// 調整 opencv 矩陣的維度使其和 torch 維度一致

	//均值歸一化
	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();

	//加載標籤文件
	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} 矩陣取前10個元素(預測值),返回一個矩陣和一個矩陣的下標索引
	std::tuple<torch::Tensor,torch::Tensor> result = output.topk(10, -1);
	auto top_scores = std::get<0>(result).view(-1);//{1,10} 變成 {10}
	auto top_idxs = std::get<1>(result).view(-1);
	std::cout << top_scores << "\r\n" << top_idxs << std::endl;

	//打印結果
	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;

	//排序操做 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;

輸出以下：

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} ]