[feature][vision] Add YOLOv7 End2End model with TRT NMS (#157)

* [feature][vision] Add YOLOv7 End2End model with TRT NMS * [docs] update yolov7end2end_trt examples docs Co-authored-by: Jason <jiangjiajun@baidu.com>
2025-10-30 11:26:39 +08:00 · 2022-08-30 15:02:48 +08:00
parent 30bb233db8
commit 3c1330e896
14 changed files with 902 additions and 2 deletions
--- a/examples/vision/detection/yolov7end2end_trt/cpp/CMakeLists.txt
+++ b/examples/vision/detection/yolov7end2end_trt/cpp/CMakeLists.txt
@@ -0,0 +1,14 @@
+PROJECT(infer_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(infer_demo ${PROJECT_SOURCE_DIR}/infer.cc)
+# 添加FastDeploy库依赖
+target_link_libraries(infer_demo ${FASTDEPLOY_LIBS})
--- a/examples/vision/detection/yolov7end2end_trt/cpp/README.md
+++ b/examples/vision/detection/yolov7end2end_trt/cpp/README.md
@@ -0,0 +1,88 @@
+# YOLOv7End2EndTRT C++部署示例
+
+本目录下提供`infer.cc`快速完成GPU上通过TensorRT加速部署的示例，该类只支持TensorRT部署。
+
+在部署前，需确认以下两个步骤
+
+- 1. 软硬件环境满足要求，参考[FastDeploy环境要求](../../../../../docs/the%20software%20and%20hardware%20requirements.md)  
+- 2. 根据开发环境，下载预编译部署库和samples代码，参考[FastDeploy预编译库](../../../../../docs/quick_start)
+
+以Linux上推理为例，在本目录执行如下命令即可完成编译测试
+
+```bash
+mkdir build
+cd build
+wget https://bj.bcebos.com/fastdeploy/release/cpp/fastdeploy-linux-x64-gpu-0.2.0.tgz
+tar xvf fastdeploy-linux-x64-0.2.0.tgz
+cmake .. -DFASTDEPLOY_INSTALL_DIR=${PWD}/fastdeploy-linux-x64-gpu-0.2.0
+make -j
+# 若预编译库没有支持该类 则请自行从源码develop分支编译最新的FastDeploy C++ SDK
+
+#下载官方转换好的yolov7模型文件和测试图片
+wget https://bj.bcebos.com/paddlehub/fastdeploy/yolov7-end2end-trt-nms.onnx
+wget https://gitee.com/paddlepaddle/PaddleDetection/raw/release/2.4/demo/000000014439.jpg
+
+# TensorRT GPU推理
+./infer_demo yolov7-end2end-trt-nms.onnx 000000014439.jpg 2
+```
+
+运行完成可视化结果如下图所示
+
+<div align='center'>
+  <img width="640" alt="image" src="https://user-images.githubusercontent.com/31974251/186605967-ad0c53f2-3ce8-4032-a90f-6f5c1238e7f4.png">
+</div>
+
+以上命令只适用于Linux或MacOS, Windows下SDK的使用方式请参考:  
+- [如何在Windows中使用FastDeploy C++ SDK](../../../../../docs/compile/how_to_use_sdk_on_windows.md)
+
+注意，YOLOv7End2EndTRT 是专门用于推理YOLOv7中导出模型带[TRT_NMS](https://github.com/WongKinYiu/yolov7/blob/main/models/experimental.py#L111) 版本的End2End模型，不带nms的模型推理请使用YOLOv7类，而 [ORT_NMS](https://github.com/WongKinYiu/yolov7/blob/main/models/experimental.py#L87) 版本的End2End模型请使用YOLOv7End2EndORT进行推理。
+
+## YOLOv7End2EndTRT C++接口
+
+### YOLOv7End2EndTRT 类
+
+```c++
+fastdeploy::vision::detection::YOLOv7End2EndTRT(
+        const string& model_file,
+        const string& params_file = "",
+        const RuntimeOption& runtime_option = RuntimeOption(),
+        const Frontend& model_format = Frontend::ONNX)
+```
+
+YOLOv7End2EndTRT 模型加载和初始化，其中model_file为导出的ONNX模型格式。
+
+**参数**
+
+> * **model_file**(str): 模型文件路径
+> * **params_file**(str): 参数文件路径，当模型格式为ONNX时，此参数传入空字符串即可
+> * **runtime_option**(RuntimeOption): 后端推理配置，默认为None，即采用默认配置
+> * **model_format**(Frontend): 模型格式，默认为ONNX格式
+
+#### Predict函数
+
+> ```c++
+> YOLOv7End2EndTRT::Predict(cv::Mat* im, DetectionResult* result,
+>                           float conf_threshold = 0.25)
+> ```
+>
+> 模型预测接口，输入图像直接输出检测结果。
+>
+> **参数**
+>
+> > * **im**: 输入图像，注意需为HWC，BGR格式
+> > * **result**: 检测结果，包括检测框，各个框的置信度, DetectionResult说明参考[视觉模型预测结果](../../../../../docs/api/vision_results/)
+> > * **conf_threshold**: 检测框置信度过滤阈值，但由于YOLOv7 End2End的模型在导出成ONNX时已经指定了score阈值，因此该参数只有在大于已经指定的阈值时才会有效。
+
+### 类成员变量
+#### 预处理参数
+用户可按照自己的实际需求，修改下列预处理参数，从而影响最终的推理和部署效果
+
+> > * **size**(vector&lt;int&gt;): 通过此参数修改预处理过程中resize的大小，包含两个整型元素，表示[width, height], 默认值为[640, 640]
+> > * **padding_value**(vector&lt;float&gt;): 通过此参数可以修改图片在resize时候做填充(padding)的值, 包含三个浮点型元素, 分别表示三个通道的值, 默认值为[114, 114, 114]
+> > * **is_no_pad**(bool): 通过此参数让图片是否通过填充的方式进行resize, `is_no_pad=ture` 表示不使用填充的方式，默认值为`is_no_pad=false`
+> > * **is_mini_pad**(bool): 通过此参数可以将resize之后图像的宽高这是为最接近`size`成员变量的值, 并且满足填充的像素大小是可以被`stride`成员变量整除的。默认值为`is_mini_pad=false`
+> > * **stride**(int): 配合`stris_mini_pad`成员变量使用, 默认值为`stride=32`
+
+- [模型介绍](../../)
+- [Python部署](../python)
+- [视觉模型预测结果](../../../../../docs/api/vision_results/)
--- a/examples/vision/detection/yolov7end2end_trt/cpp/infer.cc
+++ b/examples/vision/detection/yolov7end2end_trt/cpp/infer.cc
@@ -0,0 +1,110 @@
+// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "fastdeploy/vision.h"
+
+void CpuInfer(const std::string& model_file, const std::string& image_file) {
+  auto model = fastdeploy::vision::detection::YOLOv7End2EndTRT(model_file);
+  if (!model.Initialized()) {
+    std::cerr << "Failed to initialize." << std::endl;
+    return;
+  }
+
+  auto im = cv::imread(image_file);
+  auto im_bak = im.clone();
+
+  fastdeploy::vision::DetectionResult res;
+  if (!model.Predict(&im, &res)) {
+    std::cerr << "Failed to predict." << std::endl;
+    return;
+  }
+  std::cout << res.Str() << std::endl;
+
+  auto vis_im = fastdeploy::vision::Visualize::VisDetection(im_bak, res);
+  cv::imwrite("vis_result.jpg", vis_im);
+  std::cout << "Visualized result saved in ./vis_result.jpg" << std::endl;
+}
+
+void GpuInfer(const std::string& model_file, const std::string& image_file) {
+  auto option = fastdeploy::RuntimeOption();
+  option.UseGpu();
+  auto model =
+      fastdeploy::vision::detection::YOLOv7End2EndTRT(model_file, "", option);
+  if (!model.Initialized()) {
+    std::cerr << "Failed to initialize." << std::endl;
+    return;
+  }
+
+  auto im = cv::imread(image_file);
+  auto im_bak = im.clone();
+
+  fastdeploy::vision::DetectionResult res;
+  if (!model.Predict(&im, &res)) {
+    std::cerr << "Failed to predict." << std::endl;
+    return;
+  }
+  std::cout << res.Str() << std::endl;
+
+  auto vis_im = fastdeploy::vision::Visualize::VisDetection(im_bak, res);
+  cv::imwrite("vis_result.jpg", vis_im);
+  std::cout << "Visualized result saved in ./vis_result.jpg" << std::endl;
+}
+
+void TrtInfer(const std::string& model_file, const std::string& image_file) {
+  auto option = fastdeploy::RuntimeOption();
+  option.UseGpu();
+  option.UseTrtBackend();
+  option.SetTrtInputShape("images", {1, 3, 640, 640});
+  auto model =
+      fastdeploy::vision::detection::YOLOv7End2EndTRT(model_file, "", option);
+  if (!model.Initialized()) {
+    std::cerr << "Failed to initialize." << std::endl;
+    return;
+  }
+
+  auto im = cv::imread(image_file);
+  auto im_bak = im.clone();
+
+  fastdeploy::vision::DetectionResult res;
+  if (!model.Predict(&im, &res)) {
+    std::cerr << "Failed to predict." << std::endl;
+    return;
+  }
+  std::cout << res.Str() << std::endl;
+
+  auto vis_im = fastdeploy::vision::Visualize::VisDetection(im_bak, res);
+  cv::imwrite("vis_result.jpg", vis_im);
+  std::cout << "Visualized result saved in ./vis_result.jpg" << std::endl;
+}
+
+int main(int argc, char* argv[]) {
+  if (argc < 4) {
+    std::cout << "Usage: infer_demo path/to/model path/to/image run_option, "
+                 "e.g ./infer_model ./yolov7-end2end-trt-nms.onnx ./test.jpeg 0"
+              << std::endl;
+    std::cout << "The data type of run_option is int, 0: run with cpu; 1: run "
+                 "with gpu; 2: run with gpu and use tensorrt backend."
+              << std::endl;
+    return -1;
+  }
+
+  if (std::atoi(argv[3]) == 0) {
+    CpuInfer(argv[1], argv[2]);
+  } else if (std::atoi(argv[3]) == 1) {
+    GpuInfer(argv[1], argv[2]);
+  } else if (std::atoi(argv[3]) == 2) {
+    TrtInfer(argv[1], argv[2]);
+  }
+  return 0;
+}