[Bug Fix] change reused_input_tensors&&reused_output_tensors name (Pa…

…ddlePaddle#534) * add paddle_trt in benchmark * update benchmark in device * update benchmark * update result doc * fixed for CI * update python api_docs * update index.rst * add runtime cpp examples * deal with comments * Update infer_paddle_tensorrt.py * Add runtime quick start * deal with comments * fixed reused_input_tensors&&reused_output_tensors Co-authored-by: Jason <[email protected]>
fhfeishi · Nov 8, 2022 · d259952 · d259952
1 parent 6962921
commit d259952
Show file tree

Hide file tree

Showing 17 changed files with 247 additions and 64 deletions.
diff --git a/examples/runtime/README.md b/examples/runtime/README.md
@@ -1,5 +1,9 @@
 # FastDeploy Runtime examples
 
+FastDeploy Runtime C++ 推理示例如下
+
+## Python 示例
+
 | Example Code | Program Language | Description |
 | :------- | :------- | :---- |
 | python/infer_paddle_paddle_inference.py | Python | Deploy Paddle model with Paddle Inference(CPU/GPU) |
@@ -8,9 +12,19 @@
 | python/infer_paddle_onnxruntime.py | Python | Deploy Paddle model with ONNX Runtime(CPU/GPU)  |
 | python/infer_onnx_openvino.py | Python | Deploy ONNX model with OpenVINO(CPU) |
 | python/infer_onnx_tensorrt.py | Python | Deploy ONNX model with TensorRT(GPU) |
+
+## C++ 示例
+
+| Example Code | Program Language | Description |
+| :------- | :------- | :---- |
 | cpp/infer_paddle_paddle_inference.cc | C++ | Deploy Paddle model with Paddle Inference(CPU/GPU) |
 | cpp/infer_paddle_tensorrt.cc | C++ | Deploy Paddle model with TensorRT(GPU) |
 | cpp/infer_paddle_openvino.cc | C++ | Deploy Paddle model with OpenVINO(CPU |
 | cpp/infer_paddle_onnxruntime.cc | C++ | Deploy Paddle model with ONNX Runtime(CPU/GPU) |
 | cpp/infer_onnx_openvino.cc | C++ | Deploy ONNX model with OpenVINO(CPU) |
 | cpp/infer_onnx_tensorrt.cc | C++ | Deploy ONNX model with TensorRT(GPU) |
+
+## 详细部署文档
+
+- [Python部署](python)
+- [C++部署](cpp)
diff --git a/examples/runtime/cpp/README.md b/examples/runtime/cpp/README.md
@@ -0,0 +1,121 @@
+# C++推理
+
+在运行demo前，需确认以下两个步骤
+
+- 1. 软硬件环境满足要求，参考[FastDeploy环境要求](../../../../../docs/cn/build_and_install/download_prebuilt_libraries.md)  
+- 2. 根据开发环境，下载预编译部署库和samples代码，参考[FastDeploy预编译库](../../../../../docs/cn/build_and_install/download_prebuilt_libraries.md)
+
+本文档以 PaddleClas 分类模型 MobileNetV2 为例展示CPU上的推理示例
+
+## 1. 获取模型
+
+```bash
+wget https://bj.bcebos.com/fastdeploy/models/mobilenetv2.tgz
+tar xvf mobilenetv2.tgz
+```
+
+## 2. 配置后端
+
+如下C++代码保存为`infer_paddle_onnxruntime.cc`
+
+``` c++
+#include "fastdeploy/runtime.h"
+
+namespace fd = fastdeploy;
+
+int main(int argc, char* argv[]) {
+  std::string model_file = "mobilenetv2/inference.pdmodel";
+  std::string params_file = "mobilenetv2/inference.pdiparams";
+
+  // setup option
+  fd::RuntimeOption runtime_option;
+  runtime_option.SetModelPath(model_file, params_file, fd::ModelFormat::PADDLE);
+  runtime_option.UseOrtBackend();
+  runtime_option.SetCpuThreadNum(12);
+  // init runtime
+  std::unique_ptr<fd::Runtime> runtime =
+      std::unique_ptr<fd::Runtime>(new fd::Runtime());
+  if (!runtime->Init(runtime_option)) {
+    std::cerr << "--- Init FastDeploy Runitme Failed! "
+              << "\n--- Model:  " << model_file << std::endl;
+    return -1;
+  } else {
+    std::cout << "--- Init FastDeploy Runitme Done! "
+              << "\n--- Model:  " << model_file << std::endl;
+  }
+  // init input tensor shape
+  fd::TensorInfo info = runtime->GetInputInfo(0);
+  info.shape = {1, 3, 224, 224};
+
+  std::vector<fd::FDTensor> input_tensors(1);
+  std::vector<fd::FDTensor> output_tensors(1);
+
+  std::vector<float> inputs_data;
+  inputs_data.resize(1 * 3 * 224 * 224);
+  for (size_t i = 0; i < inputs_data.size(); ++i) {
+    inputs_data[i] = std::rand() % 1000 / 1000.0f;
+  }
+  input_tensors[0].SetExternalData({1, 3, 224, 224}, fd::FDDataType::FP32, inputs_data.data());
+
+  //get input name
+  input_tensors[0].name = info.name;
+
+  runtime->Infer(input_tensors, &output_tensors);
+
+  output_tensors[0].PrintInfo();
+  return 0;
+}
+```
+加载完成，会输出提示如下，说明初始化的后端，以及运行的硬件设备
+```
+[INFO] fastdeploy/fastdeploy_runtime.cc(283)::Init	Runtime initialized with Backend::OrtBackend in device Device::CPU.
+```
+
+## 3. 准备CMakeLists.txt
+
+FastDeploy中包含多个依赖库，直接采用`g++`或编译器编译较为繁杂，推荐使用cmake进行编译配置。示例配置如下，
+
+```cmake
+PROJECT(runtime_demo C CXX)
+CMAKE_MINIMUM_REQUIRED (VERSION 3.12)
+
+# 指定下载解压后的fastdeploy库路径
+option(FASTDEPLOY_INSTALL_DIR "Path of downloaded fastdeploy sdk.")
+
+include(${FASTDEPLOY_INSTALL_DIR}/FastDeploy.cmake)
+
+# 添加FastDeploy依赖头文件
+include_directories(${FASTDEPLOY_INCS})
+
+add_executable(runtime_demo ${PROJECT_SOURCE_DIR}/infer_onnx_openvino.cc)
+# 添加FastDeploy库依赖
+target_link_libraries(runtime_demo ${FASTDEPLOY_LIBS})
+```
+
+## 4. 编译可执行程序
+
+打开命令行终端，进入`infer_paddle_onnxruntime.cc`和`CMakeLists.txt`所在的目录，执行如下命令
+
+```bash
+mkdir build & cd build
+cmake .. -DFASTDEPLOY_INSTALL_DIR=$fastdeploy_cpp_sdk
+make -j
+```
+
+```fastdeploy_cpp_sdk``` 为FastDeploy C++部署库路径
+
+编译完成后，使用如下命令执行可得到预测结果
+```bash
+./runtime_demo
+```
+执行时如提示`error while loading shared libraries: libxxx.so: cannot open shared object file: No such file...`，说明程序执行时没有找到FastDeploy的库路径，可通过执行如下命令，将FastDeploy的库路径添加到环境变量之后，重新执行二进制程序。
+```bash
+source /Path/to/fastdeploy_cpp_sdk/fastdeploy_init.sh
+```
+
+本示例代码在各平台(Windows/Linux/Mac)上通用，但编译过程仅支持(Linux/Mac)，Windows上使用msbuild进行编译，具体使用方式参考[Windows平台使用FastDeploy C++ SDK](../../../../../docs/cn/faq/use_sdk_on_windows.md)
+
+## 其它文档
+
+- [Runtime Python 示例](../python)
+- [切换模型推理的硬件和后端](../../../../../docs/cn/faq/how_to_change_backend.md)
diff --git a/examples/runtime/python/README.md b/examples/runtime/python/README.md
@@ -0,0 +1,53 @@
+# Python推理
+
+在运行demo前，需确认以下两个步骤
+
+- 1. 软硬件环境满足要求，参考[FastDeploy环境要求](../../../../../docs/cn/build_and_install/download_prebuilt_libraries.md)  
+- 2. FastDeploy Python whl包安装，参考[FastDeploy Python安装](../../../../../docs/cn/build_and_install/download_prebuilt_libraries.md)
+
+本文档以 PaddleClas 分类模型 MobileNetV2 为例展示 CPU 上的推理示例
+
+## 1. 获取模型
+
+``` python
+import fastdeploy as fd
+
+model_url = "https://bj.bcebos.com/fastdeploy/models/mobilenetv2.tgz"
+fd.download_and_decompress(model_url, path=".")
+```
+
+## 2. 配置后端
+
+``` python
+option = fd.RuntimeOption()
+
+option.set_model_path("mobilenetv2/inference.pdmodel",
+                      "mobilenetv2/inference.pdiparams")
+
+# **** CPU 配置 ****
+option.use_cpu()
+option.use_ort_backend()
+option.set_cpu_thread_num(12)
+
+# 初始化构造runtime
+runtime = fd.Runtime(option)
+
+# 获取模型输入名
+input_name = runtime.get_input_info(0).name
+
+# 构造随机数据进行推理
+results = runtime.infer({
+    input_name: np.random.rand(1, 3, 224, 224).astype("float32")
+})
+
+print(results[0].shape)
+```
+加载完成，会输出提示如下，说明初始化的后端，以及运行的硬件设备
+```
+[INFO] fastdeploy/fastdeploy_runtime.cc(283)::Init	Runtime initialized with Backend::OrtBackend in device Device::CPU.
+```
+
+## 其它文档
+
+- [Runtime C++ 示例](../cpp)
+- [切换模型推理的硬件和后端](../../../../../docs/cn/faq/how_to_change_backend.md)
diff --git a/examples/vision/classification/yolov5cls/README.md b/examples/vision/classification/yolov5cls/README.md
@@ -17,8 +17,6 @@
 | [YOLOv5x-cls](https://bj.bcebos.com/paddlehub/fastdeploy/yolov5x-cls.onnx) | 184MB | 79.0% | 94.4% |
 
 
-
-
 ## 详细部署文档
 
 - [Python部署](python)

diff --git a/fastdeploy/fastdeploy_model.cc b/fastdeploy/fastdeploy_model.cc
@@ -239,7 +239,7 @@ bool FastDeployModel::Infer(std::vector<FDTensor>& input_tensors,
 }
 
 bool FastDeployModel::Infer() {
-  return Infer(reused_input_tensors, &reused_output_tensors);
+  return Infer(reused_input_tensors_, &reused_output_tensors_);
 }
 
 std::map<std::string, float> FastDeployModel::PrintStatisInfoOfRuntime() {

diff --git a/fastdeploy/fastdeploy_model.h b/fastdeploy/fastdeploy_model.h
@@ -28,7 +28,7 @@ class FASTDEPLOY_DECL FastDeployModel {
   virtual bool Infer(std::vector<FDTensor>& input_tensors,
                      std::vector<FDTensor>* output_tensors);
 
-  /** \brief Inference the model by the runtime. This interface is using class member reused_input_tensors to do inference and writing results to reused_output_tensors
+  /** \brief Inference the model by the runtime. This interface is using class member reused_input_tensors_ to do inference and writing results to reused_output_tensors_
   */
   virtual bool Infer();
 
@@ -107,17 +107,10 @@ class FASTDEPLOY_DECL FastDeployModel {
   /** \brief Release reused input/output buffers
   */
   virtual void ReleaseReusedBuffer() {
-    std::vector<FDTensor>().swap(reused_input_tensors);
-    std::vector<FDTensor>().swap(reused_output_tensors);
+    std::vector<FDTensor>().swap(reused_input_tensors_);
+    std::vector<FDTensor>().swap(reused_output_tensors_);
   }
 
-  /** \brief Reused input tensors
-  */
-  std::vector<FDTensor> reused_input_tensors;
-  /** \brief Reused output tensors
-  */
-  std::vector<FDTensor> reused_output_tensors;
-
  protected:
   virtual bool InitRuntime();
   virtual bool CreateCpuBackend();
@@ -126,7 +119,11 @@ class FASTDEPLOY_DECL FastDeployModel {
   virtual bool CreateRKNPUBackend();
 
   bool initialized = false;
-  std::vector<Backend> valid_external_backends;
+  std::vector<Backend> valid_external_backends_;
+  // Reused input tensors
+  std::vector<FDTensor> reused_input_tensors_;
+  // Reused output tensors
+  std::vector<FDTensor> reused_output_tensors_;
 
  private:
   std::shared_ptr<Runtime> runtime_;

diff --git a/fastdeploy/vision/classification/ppcls/model.cc b/fastdeploy/vision/classification/ppcls/model.cc
@@ -60,18 +60,18 @@ bool PaddleClasModel::Predict(const cv::Mat& im, ClassifyResult* result) {
 
 bool PaddleClasModel::BatchPredict(const std::vector<cv::Mat>& images, std::vector<ClassifyResult>* results) {
   std::vector<FDMat> fd_images = WrapMat(images);
-  if (!preprocessor_.Run(&fd_images, &reused_input_tensors)) {
+  if (!preprocessor_.Run(&fd_images, &reused_input_tensors_)) {
     FDERROR << "Failed to preprocess the input image." << std::endl;
     return false;
   }
 
-  reused_input_tensors[0].name = InputInfoOfRuntime(0).name;
-  if (!Infer(reused_input_tensors, &reused_output_tensors)) {
+  reused_input_tensors_[0].name = InputInfoOfRuntime(0).name;
+  if (!Infer(reused_input_tensors_, &reused_output_tensors_)) {
     FDERROR << "Failed to inference by runtime." << std::endl;
     return false;
   }
 
-  if (!postprocessor_.Run(reused_output_tensors, results)) {
+  if (!postprocessor_.Run(reused_output_tensors_, results)) {
     FDERROR << "Failed to postprocess the inference results by runtime." << std::endl;
     return false;
   }

diff --git a/fastdeploy/vision/detection/contrib/scaledyolov4.cc b/fastdeploy/vision/detection/contrib/scaledyolov4.cc
@@ -84,7 +84,7 @@ bool ScaledYOLOv4::Initialize() {
   is_scale_up = false;
   stride = 32;
   max_wh = 7680.0;
-  reused_input_tensors.resize(1);
+  reused_input_tensors_.resize(1);
 
   if (!InitRuntime()) {
     FDERROR << "Failed to initialize fastdeploy backend." << std::endl;
@@ -230,17 +230,17 @@ bool ScaledYOLOv4::Predict(cv::Mat* im, DetectionResult* result,
   im_info["output_shape"] = {static_cast<float>(mat.Height()),
                              static_cast<float>(mat.Width())};
 
-  if (!Preprocess(&mat, &reused_input_tensors[0], &im_info)) {
+  if (!Preprocess(&mat, &reused_input_tensors_[0], &im_info)) {
     FDERROR << "Failed to preprocess input image." << std::endl;
     return false;
   }
 
-  reused_input_tensors[0].name = InputInfoOfRuntime(0).name;
+  reused_input_tensors_[0].name = InputInfoOfRuntime(0).name;
   if (!Infer()) {
     FDERROR << "Failed to inference." << std::endl;
     return false;
   }
-  if (!Postprocess(reused_output_tensors[0], result, im_info, conf_threshold,
+  if (!Postprocess(reused_output_tensors_[0], result, im_info, conf_threshold,
                    nms_iou_threshold)) {
     FDERROR << "Failed to post process." << std::endl;
     return false;

diff --git a/fastdeploy/vision/detection/contrib/yolor.cc b/fastdeploy/vision/detection/contrib/yolor.cc
@@ -83,7 +83,7 @@ bool YOLOR::Initialize() {
   is_scale_up = false;
   stride = 32;
   max_wh = 7680.0;
-  reused_input_tensors.resize(1);
+  reused_input_tensors_.resize(1);
 
   if (!InitRuntime()) {
     FDERROR << "Failed to initialize fastdeploy backend." << std::endl;
@@ -227,18 +227,18 @@ bool YOLOR::Predict(cv::Mat* im, DetectionResult* result, float conf_threshold,
   im_info["output_shape"] = {static_cast<float>(mat.Height()),
                              static_cast<float>(mat.Width())};
 
-  if (!Preprocess(&mat, &reused_input_tensors[0], &im_info)) {
+  if (!Preprocess(&mat, &reused_input_tensors_[0], &im_info)) {
     FDERROR << "Failed to preprocess input image." << std::endl;
     return false;
   }
 
-  reused_input_tensors[0].name = InputInfoOfRuntime(0).name;
+  reused_input_tensors_[0].name = InputInfoOfRuntime(0).name;
   if (!Infer()) {
     FDERROR << "Failed to inference." << std::endl;
     return false;
   }
 
-  if (!Postprocess(reused_output_tensors[0], result, im_info, conf_threshold,
+  if (!Postprocess(reused_output_tensors_[0], result, im_info, conf_threshold,
                    nms_iou_threshold)) {
     FDERROR << "Failed to post process." << std::endl;
     return false;

diff --git a/fastdeploy/vision/detection/contrib/yolov5.cc b/fastdeploy/vision/detection/contrib/yolov5.cc
@@ -93,7 +93,7 @@ bool YOLOv5::Initialize() {
   stride_ = 32;
   max_wh_ = 7680.0;
   multi_label_ = true;
-  reused_input_tensors.resize(1);
+  reused_input_tensors_.resize(1);
 
   if (!InitRuntime()) {
     FDERROR << "Failed to initialize fastdeploy backend." << std::endl;
@@ -350,28 +350,28 @@ bool YOLOv5::Predict(cv::Mat* im, DetectionResult* result, float conf_threshold,
   std::map<std::string, std::array<float, 2>> im_info;
 
   if (use_cuda_preprocessing_) {
-    if (!CudaPreprocess(&mat, &reused_input_tensors[0], &im_info, size_,
+    if (!CudaPreprocess(&mat, &reused_input_tensors_[0], &im_info, size_,
                         padding_value_, is_mini_pad_, is_no_pad_, is_scale_up_,
                         stride_, max_wh_, multi_label_)) {
       FDERROR << "Failed to preprocess input image." << std::endl;
       return false;
     }
   } else {
-    if (!Preprocess(&mat, &reused_input_tensors[0], &im_info, size_,
+    if (!Preprocess(&mat, &reused_input_tensors_[0], &im_info, size_,
                     padding_value_, is_mini_pad_, is_no_pad_, is_scale_up_,
                     stride_, max_wh_, multi_label_)) {
       FDERROR << "Failed to preprocess input image." << std::endl;
       return false;
     }
   }
 
-  reused_input_tensors[0].name = InputInfoOfRuntime(0).name;
+  reused_input_tensors_[0].name = InputInfoOfRuntime(0).name;
   if (!Infer()) {
     FDERROR << "Failed to inference." << std::endl;
     return false;
   }
 
-  if (!Postprocess(reused_output_tensors, result, im_info, conf_threshold,
+  if (!Postprocess(reused_output_tensors_, result, im_info, conf_threshold,
                    nms_iou_threshold, multi_label_)) {
     FDERROR << "Failed to post process." << std::endl;
     return false;
Original file line number	Diff line number	Diff line change
Expand Up		@@ -17,8 +17,6 @@
		\| [YOLOv5x-cls](https://bj.bcebos.com/paddlehub/fastdeploy/yolov5x-cls.onnx) \| 184MB \| 79.0% \| 94.4% \|




		## 详细部署文档

		- [Python部署](python)
Expand Down