paddle表情识别部署

1.环境部署

1.1同样采用fastDeploy库

可以参考百度Paddle中PP-Mattingv2的部署并将之封装并调用一个C++静态库，部署过程大致一样，只是核心的代码进行了改动。

1.2相关模型

模型使用的自训练resnet50模型，其中输出的标签为：

• 0.angry
• 1.disgust
• 2.fear
• 3.happy
• 4.neutral
• 5.sad
• 6.surprise
  模型需要三个文件：model.pdmodel，model.pdiparams，model.yml

2.封装成静态库

2.1参考百度Paddle中PP-Mattingv2的部署并将之封装并调用一个C++静态库

framework.h代码：

#pragma once#define WIN32_LEAN_AND_MEAN             // 从 Windows 头文件中排除极少使用的内容
#include "fastdeploy/vision.h"std::string emotion_CpuInfer(const std::string& model_dir, const cv::Mat& image_file);std::string emotion_GpuInfer(const std::string& model_dir, const cv::Mat& image_file);int emotion_infer_by_camera(const std::string& device, const std::string& model_dir, const std::string& window_name);

emotion_StaticLib.cpp代码为：

// emotion_StaticLib.cpp : 定义静态库的函数。
//#include "pch.h"#include "framework.h"#ifdef WIN32
const char sep = '\\';
#else
const char sep = '/';
#endifstd::string emotion_CpuInfer(const std::string& model_dir, const cv::Mat& image_file) {auto model_file = model_dir + sep + "model.pdmodel";auto params_file = model_dir + sep + "model.pdiparams";auto config_file = model_dir + sep + "inference.yml";auto option = fastdeploy::RuntimeOption();option.UseCpu();auto model = fastdeploy::vision::classification::PaddleClasModel(model_file, params_file, config_file, option);std::string result;if (!model.Initialized()) {std::cerr << "Failed to initialize." << std::endl;result = "Failed to initialize.";return result;}auto im = image_file;fastdeploy::vision::ClassifyResult res;if (!model.Predict(im, &res)) {std::cerr << "Failed to predict." << std::endl;result = "Failed to initialize.";return result;}// print resstd::cout << res.Str() << std::endl;return res.Str();
}std::string emotion_GpuInfer(const std::string& model_dir, const cv::Mat& image_file) {auto model_file = model_dir + sep + "model.pdmodel";auto params_file = model_dir + sep + "model.pdiparams";auto config_file = model_dir + sep + "inference.yml";auto option = fastdeploy::RuntimeOption();option.UseGpu();auto model = fastdeploy::vision::classification::PaddleClasModel(model_file, params_file, config_file, option);std::string result;if (!model.Initialized()) {std::cerr << "Failed to initialize." << std::endl;result = "Failed to initialize.";return result;}auto im = image_file;fastdeploy::vision::ClassifyResult res;if (!model.Predict(im, &res)) {std::cerr << "Failed to predict." << std::endl;result = "Failed to initialize.";return result;}// print resstd::cout << res.Str() << std::endl;return res.Str();
}int emotion_infer_by_camera(const std::string& device, const std::string& model_dir, const std::string& window_name = "video") {cv::VideoCapture cap;cap.open(0);std::string result;if (!cap.isOpened()) {std::cout << "open camera failed!" << std::endl;return 0;}cv::namedWindow(window_name, 1);while (1) {time_t t_now = time(0);cv::Mat frame;cap >> frame;if (frame.empty()) {return 0;}cv::imshow(window_name, frame);emotion_CpuInfer(model_dir, frame);if (device == "gpu") {cv::imshow(window_name, frame);result = emotion_GpuInfer(model_dir, frame);}else {cv::imshow(window_name, frame);result = emotion_CpuInfer(model_dir, frame);}std::cout << "emotion此帧共消耗" << (time(0) - t_now) << "秒" << std::endl;if (cv::waitKey(30) >= 0) break;}cap.release();return 1;
}

所有的环境部署步骤与百度Paddle中PP-Mattingv2的部署并将之封装并调用一个C++静态库一致，在该部署过程中，只进行了cpu环境的部署

2.2项目依赖添加

注意：所有的环境必须是Release X64

2.3生成成功

到此为止封装已经超过了，在项目里面即可部署使用。

3.test

3.1创建emotion_test项目

emotion_test.cpp

#include <vector>
#include <iostream>
#include <string>
#include "C:/Users/44869/Desktop/emotion_StaticLib/emotion_StaticLib/pch.h"int main() {emotion_infer_by_camera("cpu", "A:/emotion/resnet50", "emotion");return 0;
}

3.2进行项目配置

3.3解决dll文件缺失的问题

运行C:\Users\44869\Desktop\emotion_StaticLib\fastdeploy-win-x64-1.0.3下的fastdeploy_init.bat
生成的所有dll文件复制到C:\Users\44869\Desktop\emotion_StaticLib\emotion_test\x64\Release下即可

3.4运行结果