cpp shitu code format

deploy/cpp_shitu/include/feature_extracter.h

@@ -35,8 +35,8 @@ using namespace paddle_infer;
 
 namespace Feature {
 
-class FeatureExtracter {
-public:
+    class FeatureExtracter {
+    public:
         explicit FeatureExtracter(const YAML::Node &config_file) {
             this->use_gpu_ = config_file["Global"]["use_gpu"].as<bool>();
             if (config_file["Global"]["gpu_id"].IsDefined())
@@ -59,15 +59,13 @@ public:
             this->resize_size_ =
                     config_file["RecPreProcess"]["transform_ops"][0]["ResizeImage"]["size"]
                             .as<int>();
-    this->scale_ = config_file["RecPreProcess"]["transform_ops"][1]
-                              ["NormalizeImage"]["scale"]
-                                  .as<float>();
+            this->scale_ = config_file["RecPreProcess"]["transform_ops"][1]["NormalizeImage"]["scale"].as<float>();
             this->mean_ = config_file["RecPreProcess"]["transform_ops"][1]
                           ["NormalizeImage"]["mean"]
-                                 .as<std::vector<float>>();
+                                  .as < std::vector < float >> ();
             this->std_ = config_file["RecPreProcess"]["transform_ops"][1]
                          ["NormalizeImage"]["std"]
-                                .as<std::vector<float>>();
+                                 .as < std::vector < float >> ();
             if (config_file["Global"]["rec_feature_normlize"].IsDefined())
                 this->feature_norm =
                         config_file["Global"]["rec_feature_normlize"].as<bool>();
@@ -81,11 +79,12 @@ public:
         // Run predictor
         void Run(cv::Mat &img, std::vector<float> &out_data,
                  std::vector<double> &times);
+
         void FeatureNorm(std::vector<float> &feature);
 
-  std::shared_ptr<Predictor> predictor_;
+        std::shared_ptr <Predictor> predictor_;
 
-private:
+    private:
         bool use_gpu_ = false;
         int gpu_id_ = 0;
         int gpu_mem_ = 4000;
@@ -106,6 +105,6 @@ private:
         ResizeImg resize_op_;
         Normalize normalize_op_;
         Permute permute_op_;
-};
+    };
 
 } // namespace Feature

deploy/cpp_shitu/include/nms.h

@@ -17,7 +17,7 @@
 #include <algorithm>
 #include <include/object_detector.h>
 
-template <typename T>
+template<typename T>
 static inline bool SortScorePairDescend(const std::pair<float, T> &pair1,
                                         const std::pair<float, T> &pair2) {
     return pair1.first > pair2.first;
@@ -40,9 +40,9 @@ float RectOverlap(const Detection::ObjectResult &a,
 //    top_k: if -1, keep all; otherwise, keep at most top_k.
 //    score_index_vec: store the sorted (score, index) pair.
 inline void
-GetMaxScoreIndex(const std::vector<Detection::ObjectResult> &det_result,
+GetMaxScoreIndex(const std::vector <Detection::ObjectResult> &det_result,
                  const float threshold,
-                 std::vector<std::pair<float, int>> &score_index_vec) {
+                 std::vector <std::pair<float, int>> &score_index_vec) {
     // Generate index score pairs.
     for (size_t i = 0; i < det_result.size(); ++i) {
         if (det_result[i].confidence > threshold) {
@@ -61,12 +61,12 @@ GetMaxScoreIndex(const std::vector<Detection::ObjectResult> &det_result,
     // }
 }
 
-void NMSBoxes(const std::vector<Detection::ObjectResult> det_result,
+void NMSBoxes(const std::vector <Detection::ObjectResult> det_result,
               const float score_threshold, const float nms_threshold,
               std::vector<int> &indices) {
     int a = 1;
     // Get top_k scores (with corresponding indices).
-  std::vector<std::pair<float, int>> score_index_vec;
+    std::vector <std::pair<float, int>> score_index_vec;
     GetMaxScoreIndex(det_result, score_threshold, score_index_vec);
 
     // Do nms
@@ -74,7 +74,7 @@ void NMSBoxes(const std::vector<Detection::ObjectResult> det_result,
     for (size_t i = 0; i < score_index_vec.size(); ++i) {
         const int idx = score_index_vec[i].second;
         bool keep = true;
-    for (int k = 0; k < (int)indices.size() && keep; ++k) {
+        for (int k = 0; k < (int) indices.size() && keep; ++k) {
             const int kept_idx = indices[k];
             float overlap = RectOverlap(det_result[idx], det_result[kept_idx]);
             keep = overlap <= nms_threshold;

deploy/cpp_shitu/include/object_detector.h

@@ -33,26 +33,26 @@ using namespace paddle_infer;
 
 namespace Detection {
 // Object Detection Result
-struct ObjectResult {
+    struct ObjectResult {
         // Rectangle coordinates of detected object: left, right, top, down
         std::vector<int> rect;
         // Class id of detected object
         int class_id;
         // Confidence of detected object
         float confidence;
-};
+    };
 
 // Generate visualization colormap for each class
-std::vector<int> GenerateColorMap(int num_class);
+    std::vector<int> GenerateColorMap(int num_class);
 
 // Visualiztion Detection Result
-cv::Mat VisualizeResult(const cv::Mat &img,
-                        const std::vector<ObjectResult> &results,
-                        const std::vector<std::string> &lables,
+    cv::Mat VisualizeResult(const cv::Mat &img,
+                            const std::vector <ObjectResult> &results,
+                            const std::vector <std::string> &lables,
                             const std::vector<int> &colormap, const bool is_rbox);
 
-class ObjectDetector {
-public:
+    class ObjectDetector {
+    public:
         explicit ObjectDetector(const YAML::Node &config_file) {
             this->use_gpu_ = config_file["Global"]["use_gpu"].as<bool>();
             if (config_file["Global"]["gpu_id"].IsDefined())
@@ -68,9 +68,9 @@ public:
             this->threshold_ = config_file["Global"]["threshold"].as<float>();
             this->max_det_results_ = config_file["Global"]["max_det_results"].as<int>();
             this->image_shape_ =
-        config_file["Global"]["image_shape"].as<std::vector<int>>();
+                    config_file["Global"]["image_shape"].as < std::vector < int >> ();
             this->label_list_ =
-        config_file["Global"]["labe_list"].as<std::vector<std::string>>();
+                    config_file["Global"]["labe_list"].as < std::vector < std::string >> ();
             this->ir_optim_ = config_file["Global"]["ir_optim"].as<bool>();
             this->batch_size_ = config_file["Global"]["batch_size"].as<int>();
 
@@ -83,17 +83,19 @@ public:
                        const std::string &run_mode = "fluid");
 
         // Run predictor
-  void Predict(const std::vector<cv::Mat> imgs, const int warmup = 0,
+        void Predict(const std::vector <cv::Mat> imgs, const int warmup = 0,
                      const int repeats = 1,
-               std::vector<ObjectResult> *result = nullptr,
+                     std::vector <ObjectResult> *result = nullptr,
                      std::vector<int> *bbox_num = nullptr,
                      std::vector<double> *times = nullptr);
-  const std::vector<std::string> &GetLabelList() const {
+
+        const std::vector <std::string> &GetLabelList() const {
             return this->label_list_;
         }
+
         const float &GetThreshold() const { return this->threshold_; }
 
-private:
+    private:
         bool use_gpu_ = true;
         int gpu_id_ = 0;
         int gpu_mem_ = 800;
@@ -107,7 +109,7 @@ private:
         float threshold_ = 0.5;
         float max_det_results_ = 5;
         std::vector<int> image_shape_ = {3, 640, 640};
-  std::vector<std::string> label_list_;
+        std::vector <std::string> label_list_;
         bool ir_optim_ = true;
         bool det_permute_ = true;
         bool det_postprocess_ = true;
@@ -120,16 +122,17 @@ private:
 
         // Preprocess image and copy data to input buffer
         void Preprocess(const cv::Mat &image_mat);
+
         // Postprocess result
-  void Postprocess(const std::vector<cv::Mat> mats,
-                   std::vector<ObjectResult> *result, std::vector<int> bbox_num,
+        void Postprocess(const std::vector <cv::Mat> mats,
+                         std::vector <ObjectResult> *result, std::vector<int> bbox_num,
                          bool is_rbox);
 
-  std::shared_ptr<Predictor> predictor_;
+        std::shared_ptr <Predictor> predictor_;
         Preprocessor preprocessor_;
         ImageBlob inputs_;
         std::vector<float> output_data_;
         std::vector<int> out_bbox_num_data_;
-};
+    };
 
 } // namespace Detection

deploy/cpp_shitu/include/preprocess_op.h

@@ -31,27 +31,27 @@ using namespace std;
 
 namespace Feature {
 
-class Normalize {
-public:
+    class Normalize {
+    public:
         virtual void Run(cv::Mat *im, const std::vector<float> &mean,
                          const std::vector<float> &std, float scale);
-};
+    };
 
 // RGB -> CHW
-class Permute {
-public:
+    class Permute {
+    public:
         virtual void Run(const cv::Mat *im, float *data);
-};
+    };
 
-class CenterCropImg {
-public:
+    class CenterCropImg {
+    public:
         virtual void Run(cv::Mat &im, const int crop_size = 224);
-};
+    };
 
-class ResizeImg {
-public:
+    class ResizeImg {
+    public:
         virtual void Run(const cv::Mat &img, cv::Mat &resize_img, int max_size_len,
                          int size = 0);
-};
+    };
 
 } // namespace Feature

deploy/cpp_shitu/include/preprocess_op_det.h

@@ -31,8 +31,8 @@
 namespace Detection {
 
 // Object for storing all preprocessed data
-class ImageBlob {
-public:
+    class ImageBlob {
+    public:
         // image width and height
         std::vector<float> im_shape_;
         // Buffer for image data after preprocessing
@@ -43,51 +43,54 @@ public:
         // std::vector<float>  eval_im_size_f_;
         // Scale factor for image size to origin image size
         std::vector<float> scale_factor_;
-};
+    };
 
 // Abstraction of preprocessing opration class
-class PreprocessOp {
-public:
+    class PreprocessOp {
+    public:
         virtual void Init(const YAML::Node &item) = 0;
+
         virtual void Run(cv::Mat *im, ImageBlob *data) = 0;
-};
+    };
 
-class InitInfo : public PreprocessOp {
-public:
+    class InitInfo : public PreprocessOp {
+    public:
         virtual void Init(const YAML::Node &item) {}
+
         virtual void Run(cv::Mat *im, ImageBlob *data);
-};
+    };
 
-class NormalizeImage : public PreprocessOp {
-public:
+    class NormalizeImage : public PreprocessOp {
+    public:
         virtual void Init(const YAML::Node &item) {
-    mean_ = item["mean"].as<std::vector<float>>();
-    scale_ = item["std"].as<std::vector<float>>();
+            mean_ = item["mean"].as < std::vector < float >> ();
+            scale_ = item["std"].as < std::vector < float >> ();
             is_scale_ = item["is_scale"].as<bool>();
         }
 
         virtual void Run(cv::Mat *im, ImageBlob *data);
 
-private:
+    private:
         // CHW or HWC
         std::vector<float> mean_;
         std::vector<float> scale_;
         bool is_scale_;
-};
+    };
 
-class Permute : public PreprocessOp {
-public:
+    class Permute : public PreprocessOp {
+    public:
         virtual void Init(const YAML::Node &item) {}
+
         virtual void Run(cv::Mat *im, ImageBlob *data);
-};
+    };
 
-class Resize : public PreprocessOp {
-public:
+    class Resize : public PreprocessOp {
+    public:
         virtual void Init(const YAML::Node &item) {
             interp_ = item["interp"].as<int>();
             // max_size_ = item["target_size"].as<int>();
             keep_ratio_ = item["keep_ratio"].as<bool>();
-    target_size_ = item["target_size"].as<std::vector<int>>();
+            target_size_ = item["target_size"].as < std::vector < int >> ();
         }
 
         // Compute best resize scale for x-dimension, y-dimension
@@ -95,28 +98,28 @@ public:
 
         virtual void Run(cv::Mat *im, ImageBlob *data);
 
-private:
+    private:
         int interp_ = 2;
         bool keep_ratio_;
         std::vector<int> target_size_;
         std::vector<int> in_net_shape_;
-};
+    };
 
 // Models with FPN need input shape % stride == 0
-class PadStride : public PreprocessOp {
-public:
+    class PadStride : public PreprocessOp {
+    public:
         virtual void Init(const YAML::Node &item) {
             stride_ = item["stride"].as<int>();
         }
 
         virtual void Run(cv::Mat *im, ImageBlob *data);
 
-private:
+    private:
         int stride_;
-};
+    };
 
-class Preprocessor {
-public:
+    class Preprocessor {
+    public:
         void Init(const YAML::Node &config_node) {
             // initialize image info at first
             ops_["InitInfo"] = std::make_shared<InitInfo>();
@@ -145,11 +148,11 @@ public:
 
         void Run(cv::Mat *im, ImageBlob *data);
 
-public:
-  static const std::vector<std::string> RUN_ORDER;
+    public:
+        static const std::vector <std::string> RUN_ORDER;
 
-private:
-  std::unordered_map<std::string, std::shared_ptr<PreprocessOp>> ops_;
-};
+    private:
+        std::unordered_map <std::string, std::shared_ptr<PreprocessOp>> ops_;
+    };
 
 } // namespace Detection

deploy/cpp_shitu/include/vector_search.h

@@ -26,7 +26,7 @@
 #include <map>
 
 struct SearchResult {
-  std::vector<faiss::Index::idx_t> I;
+    std::vector <faiss::Index::idx_t> I;
     std::vector<float> D;
     int return_k;
 };
@@ -46,10 +46,15 @@ public:
         this->I.resize(this->return_k * this->max_query_number);
         this->D.resize(this->return_k * this->max_query_number);
     };
+
     void LoadIdMap();
+
     void LoadIndexFile();
+
     const SearchResult &Search(float *feature, int query_number);
+
     const std::string &GetLabel(faiss::Index::idx_t ind);
+
     const float &GetThreshold() { return this->score_thres; }
 
 private:
@@ -60,6 +65,6 @@ private:
     faiss::Index *index;
     int max_query_number = 6;
     std::vector<float> D;
-  std::vector<faiss::Index::idx_t> I;
+    std::vector <faiss::Index::idx_t> I;
     SearchResult sr;
 };

deploy/cpp_shitu/include/yaml_config.h

@@ -45,9 +45,14 @@ public:
     explicit YamlConfig(const std::string &path) {
         config_file = ReadYamlConfig(path);
     }
-  static std::vector<std::string> ReadDict(const std::string &path);
+
+    static std::vector <std::string> ReadDict(const std::string &path);
+
     static std::map<int, std::string> ReadIndexId(const std::string &path);
+
     static YAML::Node ReadYamlConfig(const std::string &path);
+
     void PrintConfigInfo();
+
     YAML::Node config_file;
 };

deploy/cpp_shitu/readme.md

@@ -6,10 +6,7 @@
 ## 1. 准备环境
 
 ### 运行准备
-- Linux环境，推荐使用docker。
-- Windows环境，目前支持基于`Visual Studio 2019 Community`进行编译；此外，如果您希望通过生成`sln解决方案`的方式进行编译，可以参考该文档：[https://zhuanlan.zhihu.com/p/145446681](https://zhuanlan.zhihu.com/p/145446681)
-
-* 该文档主要介绍基于Linux环境下的PaddleClas C++预测流程，如果需要在Windows环境下使用预测库进行C++预测，具体编译方法请参考[Windows下编译教程](./docs/windows_vs2019_build.md)。
+- Linux环境，推荐使用ubuntu docker。
 
 ### 1.1 编译opencv库
 
@@ -103,7 +100,7 @@ make -j
 make inference_lib_dist
 ```
 
-更多编译参数选项可以参考Paddle C++预测库官网：[https://www.paddlepaddle.org.cn/documentation/docs/zh/develop/guides/05_inference_deployment/inference/build_and_install_lib_cn.html#id16](https://www.paddlepaddle.org.cn/documentation/docs/zh/develop/guides/05_inference_deployment/inference/build_and_install_lib_cn.html#id16)。
+更多编译参数选项可以参考[Paddle C++预测库官网](https://www.paddlepaddle.org.cn/documentation/docs/zh/develop/guides/05_inference_deployment/inference/build_and_install_lib_cn.html#id16)。
 
 
 * 编译完成之后，可以在`build/paddle_inference_install_dir/`文件下看到生成了以下文件及文件夹。
@@ -137,6 +134,7 @@ tar -xvf paddle_inference.tgz
 ### 1.3 安装faiss库
 
 ```shell
+ # 下载 faiss
  git clone https://github.com/facebookresearch/faiss.git
  cd faiss 
  cmake -B build . -DFAISS_ENABLE_PYTHON=OFF  -DCMAKE_INSTALL_PREFIX=${faiss_install_path}
@@ -144,22 +142,19 @@ tar -xvf paddle_inference.tgz
  make -C build install
 ```
 
-## 2 开始运行
+在安装`faiss`前，请安装`openblas`，`ubuntu`系统中安装命令如下：
 
-### 2.1 将模型导出为inference model
+```shell
+apt-get install libopenblas-dev
+```
 
-* 可以参考[模型导出](../../tools/export_model.py)，导出`inference model`，用于模型预测。得到预测模型后，假设模型文件放在`inference`目录下，则目录结构如下。
+注意本教程以安装faiss cpu版本为例，安装时请参考[faiss](https://github.com/facebookresearch/faiss)官网文档，根据需求自行安装。
 
-```
-inference/
-|--cls_infer.pdmodel
-|--cls_infer.pdiparams
-```
-**注意**：上述文件中，`cls_infer.pdmodel`文件存储了模型结构信息，`cls_infer.pdiparams`文件存储了模型参数信息。注意两个文件的路径需要与配置文件`tools/config.txt`中的`cls_model_path`和`cls_params_path`参数对应一致。
+## 2 代码编译
 
 ### 2.2 编译PaddleClas C++预测demo
 
-* 编译命令如下，其中Paddle C++预测库、opencv等其他依赖库的地址需要换成自己机器上的实际地址。
+编译命令如下，其中Paddle C++预测库、opencv等其他依赖库的地址需要换成自己机器上的实际地址。同时，编译过程中需要下载编译`yaml-cpp`等C++库，请保持联网环境。
 
 
 ```shell
@@ -169,11 +164,12 @@ sh tools/build.sh
 具体地，`tools/build.sh`中内容如下。
 
 ```shell
-OPENCV_DIR=your_opencv_dir
-LIB_DIR=your_paddle_inference_dir
-CUDA_LIB_DIR=your_cuda_lib_dir
-CUDNN_LIB_DIR=your_cudnn_lib_dir
-TENSORRT_DIR=your_tensorrt_lib_dir
+OPENCV_DIR=${opencv_install_dir}
+LIB_DIR=${paddle_inference_dir}
+CUDA_LIB_DIR=/usr/local/cuda/lib64
+CUDNN_LIB_DIR=/usr/lib/x86_64-linux-gnu/
+FAISS_DIR=${faiss_install_dir}
+FAISS_WITH_MKL=OFF
 
 BUILD_DIR=build
 rm -rf ${BUILD_DIR}
@@ -182,14 +178,14 @@ cd ${BUILD_DIR}
 cmake .. \
     -DPADDLE_LIB=${LIB_DIR} \
     -DWITH_MKL=ON \
-    -DDEMO_NAME=clas_system \
     -DWITH_GPU=OFF \
     -DWITH_STATIC_LIB=OFF \
-    -DWITH_TENSORRT=OFF \
-    -DTENSORRT_DIR=${TENSORRT_DIR} \
+    -DUSE_TENSORRT=OFF \
     -DOPENCV_DIR=${OPENCV_DIR} \
     -DCUDNN_LIB=${CUDNN_LIB_DIR} \
     -DCUDA_LIB=${CUDA_LIB_DIR} \
+    -DFAISS_DIR=${FAISS_DIR} \
+    -DFAISS_WITH_MKL=${FAISS_WITH_MKL}
 
 make -j
 ```
@@ -197,47 +193,75 @@ make -j
 上述命令中，
 
 * `OPENCV_DIR`为opencv编译安装的地址（本例中为`opencv-3.4.7/opencv3`文件夹的路径）；
-
 * `LIB_DIR`为下载的Paddle预测库（`paddle_inference`文件夹），或编译生成的Paddle预测库（`build/paddle_inference_install_dir`文件夹）的路径；
-
 * `CUDA_LIB_DIR`为cuda库文件地址，在docker中为`/usr/local/cuda/lib64`；
-
 * `CUDNN_LIB_DIR`为cudnn库文件地址，在docker中为`/usr/lib/x86_64-linux-gnu/`。
-
 * `TENSORRT_DIR`是tensorrt库文件地址，在dokcer中为`/usr/local/TensorRT6-cuda10.0-cudnn7/`，TensorRT需要结合GPU使用。
+* `FAISS_DIR`是faiss的安装地址
+  * `FAISS_WITH_MKL`是指在编译faiss的过程中，是否使用了mkldnn，本文档中编译faiss，没有使用，而使用了openblas，故设置为`OFF`，若使用了mkldnn，则为`ON`.
 
-在执行上述命令，编译完成之后，会在当前路径下生成`build`文件夹，其中生成一个名为`clas_system`的可执行文件。
 
+在执行上述命令，编译完成之后，会在当前路径下生成`build`文件夹，其中生成一个名为`pp_shitu`的可执行文件。
 
-### 运行demo
-* 首先修改`tools/config.txt`中对应字段：
-  * use_gpu：是否使用GPU；
-  * gpu_id：使用的GPU卡号；
-  * gpu_mem：显存；
-  * cpu_math_library_num_threads：底层科学计算库所用线程的数量；
-  * use_mkldnn：是否使用MKLDNN加速；
-  * use_tensorrt: 是否使用tensorRT进行加速；
-  * use_fp16：是否使用半精度浮点数进行计算，该选项仅在use_tensorrt为true时有效；
-  * cls_model_path：预测模型结构文件路径；
-  * cls_params_path：预测模型参数文件路径；
-  * resize_short_size：预处理时图像缩放大小；
-  * crop_size：预处理时图像裁剪后的大小。
+## 3 运行demo
 
-* 然后修改`tools/run.sh`：
-  * `./build/clas_system ./tools/config.txt ./docs/imgs/ILSVRC2012_val_00000666.JPEG`
-  * 上述命令中分别为：编译得到的可执行文件`clas_system`；运行时的配置文件`config.txt`；待预测的图像。
+- 请参考[识别快速开始文档](../../docs/zh_CN/quick_start/quick_start_recognition.md)，下载好相应的 轻量级通用主体检测模型、轻量级通用识别模型及瓶装饮料测试数据并解压。
 
-* 最后执行以下命令，完成对一幅图像的分类。
+  ```shell
+  mkdir models
+  cd models
+  wget https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/rec/models/inference/picodet_PPLCNet_x2_5_mainbody_lite_v1.0_infer.tar
+  tar -xf picodet_PPLCNet_x2_5_mainbody_lite_v1.0_infer.tar
+  wget https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/rec/models/inference/general_PPLCNet_x2_5_lite_v1.0_infer.tar
+  tar -xf general_PPLCNet_x2_5_lite_v1.0_infer.tar
+  cd ..
   
-```shell
-sh tools/run.sh
-```
+  mkdir data 
+  cd data
+  wget https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/rec/data/drink_dataset_v1.0.tar
+  tar -xf drink_dataset_v1.0.tar
+  cd ..
+  ```
+
+- 将相应的yaml文件拷到`test`文件夹下
+
+  ```shell
+  cp ../configs/inference_drink.yaml .
+  ```
+
+- 将`inference_drink.yaml`中的相对路径，改成基于本目录的路径或者绝对路径。涉及到的参数有
+
+  - Global.infer_imgs ：此参数可以是具体的图像地址，也可以是图像集所在的目录
+  - Global.det_inference_model_dir ： 检测模型存储目录
+  - Global.rec_inference_model_dir ： 识别模型存储目录
+  - IndexProcess.index_dir ： 检索库的存储目录，在示例中，检索库在下载的demo数据中。
+
+- 字典转换
+
+  由于python的检索库的字典，使用`pickle`进行的序列化存储，导致C++不方便读取，因此进行转换
+
+  ```shell
+  python tools/transform_id_map.py -c inference_drink.yaml
+  ```
+
+  转换成功后，在`IndexProcess.index_dir`目录下生成`id_map.txt`，方便c++ 读取。
+
+- 执行程序
+
+  ```shell
+  ./build/pp_shitu -c inference_drink.yaml
+  # or
+  ./build/pp_shitu -config inference_drink.yaml
+  ```
+
+  若对图像集进行检索，则可能得到，如下结果。注意，此结果只做展示，具体以实际运行结果为准。
+
+  同时，需注意的是，由于opencv 版本问题，会导致图像在预处理的过程中，resize产生细微差别，导致python 和c++结果，轻微不同，如bbox相差几个像素，检索结果小数点后3位diff等。但不会改变最终检索label。
 
-* 最终屏幕上会输出结果，如下图所示。
+  ![](../../docs/images/quick_start/shitu_c++_result.png)
 
-<div align="center">
-    <img src="./docs/imgs/cpp_infer_result.png" width="600">
-</div>
+## 4  使用自己模型
 
+使用自己训练的模型，可以参考[模型导出](../../docs/zh_CN/inference_deployment/export_model.md)，导出`inference model`，用于模型预测。
 
-其中`class id`表示置信度最高的类别对应的id，score表示图片属于该类别的概率。
+同时注意修改`yaml`文件中具体参数。

deploy/cpp_shitu/src/feature_extracter.cpp

@@ -18,7 +18,7 @@
 
 namespace Feature {
 
-void FeatureExtracter::LoadModel(const std::string &model_path,
+    void FeatureExtracter::LoadModel(const std::string &model_path,
                                      const std::string &params_path) {
         paddle_infer::Config config;
         config.SetModel(model_path, params_path);
@@ -52,9 +52,9 @@ void FeatureExtracter::LoadModel(const std::string &model_path,
         config.DisableGlogInfo();
 
         this->predictor_ = CreatePredictor(config);
-}
+    }
 
-void FeatureExtracter::Run(cv::Mat &img, std::vector<float> &out_data,
+    void FeatureExtracter::Run(cv::Mat &img, std::vector<float> &out_data,
                                std::vector<double> &times) {
         cv::Mat resize_img;
         std::vector<double> time;
@@ -108,12 +108,12 @@ void FeatureExtracter::Run(cv::Mat &img, std::vector<float> &out_data,
         times[0] += time[0];
         times[1] += time[1];
         times[2] += time[2];
-}
+    }
 
-void FeatureExtracter::FeatureNorm(std::vector<float> &featuer) {
+    void FeatureExtracter::FeatureNorm(std::vector<float> &featuer) {
         float featuer_sqrt = std::sqrt(std::inner_product(
                 featuer.begin(), featuer.end(), featuer.begin(), 0.0f));
         for (int i = 0; i < featuer.size(); ++i)
             featuer[i] /= featuer_sqrt;
-}
+    }
 } // namespace Feature

deploy/cpp_shitu/src/main.cpp

@@ -37,13 +37,15 @@
 using namespace std;
 using namespace cv;
 
-DEFINE_string(config, "", "Path of yaml file");
-DEFINE_string(c, "", "Path of yaml file");
+DEFINE_string(config,
+"", "Path of yaml file");
+DEFINE_string(c,
+"", "Path of yaml file");
 
-void DetPredictImage(const std::vector<cv::Mat> &batch_imgs,
-                     const std::vector<std::string> &all_img_paths,
+void DetPredictImage(const std::vector <cv::Mat> &batch_imgs,
+                     const std::vector <std::string> &all_img_paths,
                      const int batch_size, Detection::ObjectDetector *det,
-                     std::vector<Detection::ObjectResult> &im_result,
+                     std::vector <Detection::ObjectResult> &im_result,
                      std::vector<int> &im_bbox_num, std::vector<double> &det_t,
                      const bool visual_det = false,
                      const bool run_benchmark = false,
@@ -63,7 +65,7 @@ void DetPredictImage(const std::vector<cv::Mat> &batch_imgs,
         // }
 
         // Store all detected result
-    std::vector<Detection::ObjectResult> result;
+        std::vector <Detection::ObjectResult> result;
         std::vector<int> bbox_num;
         std::vector<double> det_times;
         bool is_rbox = false;
@@ -134,7 +136,7 @@ void DetPredictImage(const std::vector<cv::Mat> &batch_imgs,
 }
 
 void PrintResult(std::string &img_path,
-                 std::vector<Detection::ObjectResult> &det_result,
+                 std::vector <Detection::ObjectResult> &det_result,
                  std::vector<int> &indeices, VectorSearch &vector_search,
                  SearchResult &search_result) {
     printf("%s:\n", img_path.c_str());
@@ -194,9 +196,9 @@ int main(int argc, char **argv) {
     // load image_file_path
     std::string path =
             config.config_file["Global"]["infer_imgs"].as<std::string>();
-  std::vector<std::string> img_files_list;
+    std::vector <std::string> img_files_list;
     if (cv::utils::fs::isDirectory(path)) {
-    std::vector<cv::String> filenames;
+        std::vector <cv::String> filenames;
         cv::glob(path, filenames);
         for (auto f : filenames) {
             img_files_list.push_back(f);
@@ -209,10 +211,10 @@ int main(int argc, char **argv) {
     std::vector<double> cls_times = {0, 0, 0};
     std::vector<double> det_times = {0, 0, 0};
     // for read images
-  std::vector<cv::Mat> batch_imgs;
-  std::vector<std::string> img_paths;
+    std::vector <cv::Mat> batch_imgs;
+    std::vector <std::string> img_paths;
     // for detection
-  std::vector<Detection::ObjectResult> det_result;
+    std::vector <Detection::ObjectResult> det_result;
     std::vector<int> det_bbox_num;
     // for vector search
     std::vector<float> features;

deploy/cpp_shitu/src/object_detector.cpp

@@ -22,7 +22,7 @@ using namespace paddle_infer;
 namespace Detection {
 
 // Load Model and create model predictor
-void ObjectDetector::LoadModel(const std::string &model_dir,
+    void ObjectDetector::LoadModel(const std::string &model_dir,
                                    const int batch_size,
                                    const std::string &run_mode) {
         paddle_infer::Config config;
@@ -64,11 +64,11 @@ void ObjectDetector::LoadModel(const std::string &model_dir,
                                                               this->trt_max_shape_};
                     const std::vector<int> opt_input_shape = {1, 3, this->trt_opt_shape_,
                                                               this->trt_opt_shape_};
-        const std::map<std::string, std::vector<int>> map_min_input_shape = {
+                    const std::map <std::string, std::vector<int>> map_min_input_shape = {
                             {"image", min_input_shape}};
-        const std::map<std::string, std::vector<int>> map_max_input_shape = {
+                    const std::map <std::string, std::vector<int>> map_max_input_shape = {
                             {"image", max_input_shape}};
-        const std::map<std::string, std::vector<int>> map_opt_input_shape = {
+                    const std::map <std::string, std::vector<int>> map_opt_input_shape = {
                             {"image", opt_input_shape}};
 
                     config.SetTRTDynamicShapeInfo(map_min_input_shape, map_max_input_shape,
@@ -94,12 +94,12 @@ void ObjectDetector::LoadModel(const std::string &model_dir,
         // Memory optimization
         config.EnableMemoryOptim();
         predictor_ = std::move(CreatePredictor(config));
-}
+    }
 
 // Visualiztion MaskDetector results
-cv::Mat VisualizeResult(const cv::Mat &img,
-                        const std::vector<ObjectResult> &results,
-                        const std::vector<std::string> &lables,
+    cv::Mat VisualizeResult(const cv::Mat &img,
+                            const std::vector <ObjectResult> &results,
+                            const std::vector <std::string> &lables,
                             const std::vector<int> &colormap,
                             const bool is_rbox = false) {
         cv::Mat vis_img = img.clone();
@@ -151,17 +151,17 @@ cv::Mat VisualizeResult(const cv::Mat &img,
                         cv::Scalar(255, 255, 255), thickness);
         }
         return vis_img;
-}
+    }
 
-void ObjectDetector::Preprocess(const cv::Mat &ori_im) {
+    void ObjectDetector::Preprocess(const cv::Mat &ori_im) {
         // Clone the image : keep the original mat for postprocess
         cv::Mat im = ori_im.clone();
         cv::cvtColor(im, im, cv::COLOR_BGR2RGB);
         preprocessor_.Run(&im, &inputs_);
-}
+    }
 
-void ObjectDetector::Postprocess(const std::vector<cv::Mat> mats,
-                                 std::vector<ObjectResult> *result,
+    void ObjectDetector::Postprocess(const std::vector <cv::Mat> mats,
+                                     std::vector <ObjectResult> *result,
                                      std::vector<int> bbox_num,
                                      bool is_rbox = false) {
         result->clear();
@@ -215,11 +215,11 @@ void ObjectDetector::Postprocess(const std::vector<cv::Mat> mats,
             }
             start_idx += bbox_num[im_id];
         }
-}
+    }
 
-void ObjectDetector::Predict(const std::vector<cv::Mat> imgs, const int warmup,
+    void ObjectDetector::Predict(const std::vector <cv::Mat> imgs, const int warmup,
                                  const int repeats,
-                             std::vector<ObjectResult> *result,
+                                 std::vector <ObjectResult> *result,
                                  std::vector<int> *bbox_num,
                                  std::vector<double> *times) {
         auto preprocess_start = std::chrono::steady_clock::now();
@@ -344,9 +344,9 @@ void ObjectDetector::Predict(const std::vector<cv::Mat> imgs, const int warmup,
         std::chrono::duration<float> postprocess_diff =
                 postprocess_end - postprocess_start;
         times->push_back(double(postprocess_diff.count() * 1000));
-}
+    }
 
-std::vector<int> GenerateColorMap(int num_class) {
+    std::vector<int> GenerateColorMap(int num_class) {
         auto colormap = std::vector<int>(3 * num_class, 0);
         for (int i = 0; i < num_class; ++i) {
             int j = 0;
@@ -360,6 +360,6 @@ std::vector<int> GenerateColorMap(int num_class) {
             }
         }
         return colormap;
-}
+    }
 
 } // namespace Detection

deploy/cpp_shitu/src/preprocess_op.cpp

@@ -32,16 +32,16 @@
 
 namespace Feature {
 
-void Permute::Run(const cv::Mat *im, float *data) {
+    void Permute::Run(const cv::Mat *im, float *data) {
         int rh = im->rows;
         int rw = im->cols;
         int rc = im->channels();
         for (int i = 0; i < rc; ++i) {
             cv::extractChannel(*im, cv::Mat(rh, rw, CV_32FC1, data + i * rh * rw), i);
         }
-}
+    }
 
-void Normalize::Run(cv::Mat *im, const std::vector<float> &mean,
+    void Normalize::Run(cv::Mat *im, const std::vector<float> &mean,
                         const std::vector<float> &std, float scale) {
         (*im).convertTo(*im, CV_32FC3, scale);
         for (int h = 0; h < im->rows; h++) {
@@ -54,18 +54,18 @@ void Normalize::Run(cv::Mat *im, const std::vector<float> &mean,
                         (im->at<cv::Vec3f>(h, w)[2] - mean[2]) / std[2];
             }
         }
-}
+    }
 
-void CenterCropImg::Run(cv::Mat &img, const int crop_size) {
+    void CenterCropImg::Run(cv::Mat &img, const int crop_size) {
         int resize_w = img.cols;
         int resize_h = img.rows;
         int w_start = int((resize_w - crop_size) / 2);
         int h_start = int((resize_h - crop_size) / 2);
         cv::Rect rect(w_start, h_start, crop_size, crop_size);
         img = img(rect);
-}
+    }
 
-void ResizeImg::Run(const cv::Mat &img, cv::Mat &resize_img,
+    void ResizeImg::Run(const cv::Mat &img, cv::Mat &resize_img,
                         int resize_short_size, int size) {
         int resize_h = 0;
         int resize_w = 0;
@@ -86,6 +86,6 @@ void ResizeImg::Run(const cv::Mat &img, cv::Mat &resize_img,
             resize_w = round(float(w) * ratio);
         }
         cv::resize(img, resize_img, cv::Size(resize_w, resize_h));
-}
+    }
 
 } // namespace Feature

deploy/cpp_shitu/src/preprocess_op_det.cpp

@@ -19,15 +19,15 @@
 
 namespace Detection {
 
-void InitInfo::Run(cv::Mat *im, ImageBlob *data) {
+    void InitInfo::Run(cv::Mat *im, ImageBlob *data) {
         data->im_shape_ = {static_cast<float>(im->rows),
                            static_cast<float>(im->cols)};
         data->scale_factor_ = {1., 1.};
         data->in_net_shape_ = {static_cast<float>(im->rows),
                                static_cast<float>(im->cols)};
-}
+    }
 
-void NormalizeImage::Run(cv::Mat *im, ImageBlob *data) {
+    void NormalizeImage::Run(cv::Mat *im, ImageBlob *data) {
         double e = 1.0;
         if (is_scale_) {
             e /= 255.0;
@@ -43,9 +43,9 @@ void NormalizeImage::Run(cv::Mat *im, ImageBlob *data) {
                         (im->at<cv::Vec3f>(h, w)[2] - mean_[2]) / scale_[2];
             }
         }
-}
+    }
 
-void Permute::Run(cv::Mat *im, ImageBlob *data) {
+    void Permute::Run(cv::Mat *im, ImageBlob *data) {
         int rh = im->rows;
         int rw = im->cols;
         int rc = im->channels();
@@ -54,9 +54,9 @@ void Permute::Run(cv::Mat *im, ImageBlob *data) {
         for (int i = 0; i < rc; ++i) {
             cv::extractChannel(*im, cv::Mat(rh, rw, CV_32FC1, base + i * rh * rw), i);
         }
-}
+    }
 
-void Resize::Run(cv::Mat *im, ImageBlob *data) {
+    void Resize::Run(cv::Mat *im, ImageBlob *data) {
         auto resize_scale = GenerateScale(*im);
         data->im_shape_ = {static_cast<float>(im->cols * resize_scale.first),
                            static_cast<float>(im->rows * resize_scale.second)};
@@ -70,9 +70,9 @@ void Resize::Run(cv::Mat *im, ImageBlob *data) {
         data->scale_factor_ = {
                 resize_scale.second, resize_scale.first,
         };
-}
+    }
 
-std::pair<double, double> Resize::GenerateScale(const cv::Mat &im) {
+    std::pair<double, double> Resize::GenerateScale(const cv::Mat &im) {
         std::pair<double, double> resize_scale;
         int origin_w = im.cols;
         int origin_h = im.rows;
@@ -97,9 +97,9 @@ std::pair<double, double> Resize::GenerateScale(const cv::Mat &im) {
                     static_cast<double>(target_size_[0]) / static_cast<double>(origin_h);
         }
         return resize_scale;
-}
+    }
 
-void PadStride::Run(cv::Mat *im, ImageBlob *data) {
+    void PadStride::Run(cv::Mat *im, ImageBlob *data) {
         if (stride_ <= 0) {
             return;
         }
@@ -113,18 +113,18 @@ void PadStride::Run(cv::Mat *im, ImageBlob *data) {
         data->in_net_shape_ = {
                 static_cast<float>(im->rows), static_cast<float>(im->cols),
         };
-}
+    }
 
 // Preprocessor op running order
-const std::vector<std::string> Preprocessor::RUN_ORDER = {
+    const std::vector <std::string> Preprocessor::RUN_ORDER = {
             "InitInfo", "Resize", "NormalizeImage", "PadStride", "Permute"};
 
-void Preprocessor::Run(cv::Mat *im, ImageBlob *data) {
+    void Preprocessor::Run(cv::Mat *im, ImageBlob *data) {
         for (const auto &name : RUN_ORDER) {
             if (ops_.find(name) != ops_.end()) {
                 ops_[name]->Run(im, data);
             }
         }
-}
+    }
 
 } // namespace Detection

deploy/cpp_shitu/src/vector_search.cpp

@@ -29,11 +29,11 @@ void VectorSearch::LoadIdMap() {
     std::string file_path = this->index_dir + OS_PATH_SEP + "id_map.txt";
     std::ifstream in(file_path);
     std::string line;
-  std::vector<std::string> m_vec;
+    std::vector <std::string> m_vec;
     if (in) {
         while (getline(in, line)) {
             std::regex ws_re("\\s+");
-      std::vector<std::string> v(
+            std::vector <std::string> v(
                     std::sregex_token_iterator(line.begin(), line.end(), ws_re, -1),
                     std::sregex_token_iterator());
             if (v.size() != 2) {

deploy/cpp_shitu/src/yaml_config.cpp

@@ -19,10 +19,10 @@
 #include <include/yaml_config.h>
 
 
-std::vector<std::string> YamlConfig::ReadDict(const std::string &path) {
+std::vector <std::string> YamlConfig::ReadDict(const std::string &path) {
     std::ifstream in(path);
     std::string line;
-  std::vector<std::string> m_vec;
+    std::vector <std::string> m_vec;
     if (in) {
         while (getline(in, line)) {
             m_vec.push_back(line);
@@ -42,7 +42,7 @@ std::map<int, std::string> YamlConfig::ReadIndexId(const std::string &path) {
     if (in) {
         while (getline(in, line)) {
             std::regex ws_re("\\s+");
-      std::vector<std::string> v(
+            std::vector <std::string> v(
                     std::sregex_token_iterator(line.begin(), line.end(), ws_re, -1),
                     std::sregex_token_iterator());
             if (v.size() != 3) {

deploy/cpp_shitu/tools/build.sh

-OPENCV_DIR=/work/project/project/cpp_infer/opencv-3.4.7/opencv3
-LIB_DIR=/work/project/project/cpp_infer/paddle_inference/
+OPENCV_DIR=${opencv_install_dir}
+LIB_DIR=${paddle_inference_dir}
 CUDA_LIB_DIR=/usr/local/cuda/lib64
 CUDNN_LIB_DIR=/usr/lib/x86_64-linux-gnu/
-FAISS_DIR=/work/project/project/cpp_infer/faiss/faiss_install
+FAISS_DIR=${faiss_install_dir}
 FAISS_WITH_MKL=OFF
 
 BUILD_DIR=build

deploy/cpp_shitu/tools/config.txt

-# model load config
-use_gpu  0
-gpu_id  0
-gpu_mem  4000
-cpu_threads  10
-use_mkldnn 1
-use_tensorrt 0
-use_fp16 0
-
-# cls config
-cls_model_path  /PaddleClas/inference/cls_infer.pdmodel
-cls_params_path /PaddleClas/inference/cls_infer.pdiparams
-resize_short_size 256
-crop_size 224
-
-# for log env info
-benchmark 0

deploy/cpp_shitu/tools/run.sh

-./build/clas_system ../configs/inference_rec.yaml