split cpp inference

deploy/cpp_infer/src_rec/preprocess_op.cpp

+// Copyright (c) 2020 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 "opencv2/core.hpp"
+#include "opencv2/imgcodecs.hpp"
+#include "opencv2/imgproc.hpp"
+#include "paddle_api.h"
+#include "paddle_inference_api.h"
+#include <chrono>
+#include <iomanip>
+#include <iostream>
+#include <ostream>
+#include <vector>
+
+#include <cstring>
+#include <fstream>
+#include <numeric>
+
+#include <include/preprocess_op.h>
+
+namespace PaddleOCR {
+
+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,
+                    const std::vector<float> &scale, const bool is_scale) {
+  double e = 1.0;
+  if (is_scale) {
+    e /= 255.0;
+  }
+  (*im).convertTo(*im, CV_32FC3, e);
+  std::vector<cv::Mat> bgr_channels(3);
+  cv::split(*im, bgr_channels);
+  for (auto i = 0; i < bgr_channels.size(); i++) {
+    bgr_channels[i].convertTo(bgr_channels[i], CV_32FC1, 1.0 * scale[i],
+                              (0.0 - mean[i]) * scale[i]);
+  }
+  cv::merge(bgr_channels, *im);
+}
+
+void ResizeImgType0::Run(const cv::Mat &img, cv::Mat &resize_img,
+                         int max_size_len, float &ratio_h, float &ratio_w,
+                         bool use_tensorrt) {
+  int w = img.cols;
+  int h = img.rows;
+
+  float ratio = 1.f;
+  int max_wh = w >= h ? w : h;
+  if (max_wh > max_size_len) {
+    if (h > w) {
+      ratio = float(max_size_len) / float(h);
+    } else {
+      ratio = float(max_size_len) / float(w);
+    }
+  }
+
+  int resize_h = int(float(h) * ratio);
+  int resize_w = int(float(w) * ratio);
+
+  resize_h = max(int(round(float(resize_h) / 32) * 32), 32);
+  resize_w = max(int(round(float(resize_w) / 32) * 32), 32);
+
+  cv::resize(img, resize_img, cv::Size(resize_w, resize_h));
+  ratio_h = float(resize_h) / float(h);
+  ratio_w = float(resize_w) / float(w);
+}
+
+void CrnnResizeImg::Run(const cv::Mat &img, cv::Mat &resize_img, float wh_ratio,
+                        bool use_tensorrt,
+                        const std::vector<int> &rec_image_shape) {
+  int imgC, imgH, imgW;
+  imgC = rec_image_shape[0];
+  imgH = rec_image_shape[1];
+  imgW = rec_image_shape[2];
+
+  imgW = int(32 * wh_ratio);
+
+  float ratio = float(img.cols) / float(img.rows);
+  int resize_w, resize_h;
+  if (ceilf(imgH * ratio) > imgW)
+    resize_w = imgW;
+  else
+    resize_w = int(ceilf(imgH * ratio));
+
+  cv::resize(img, resize_img, cv::Size(resize_w, imgH), 0.f, 0.f,
+             cv::INTER_LINEAR);
+  cv::copyMakeBorder(resize_img, resize_img, 0, 0, 0,
+                     int(imgW - resize_img.cols), cv::BORDER_CONSTANT,
+                     {127, 127, 127});
+}
+
+void ClsResizeImg::Run(const cv::Mat &img, cv::Mat &resize_img,
+                       bool use_tensorrt,
+                       const std::vector<int> &rec_image_shape) {
+  int imgC, imgH, imgW;
+  imgC = rec_image_shape[0];
+  imgH = rec_image_shape[1];
+  imgW = rec_image_shape[2];
+
+  float ratio = float(img.cols) / float(img.rows);
+  int resize_w, resize_h;
+  if (ceilf(imgH * ratio) > imgW)
+    resize_w = imgW;
+  else
+    resize_w = int(ceilf(imgH * ratio));
+
+  cv::resize(img, resize_img, cv::Size(resize_w, imgH), 0.f, 0.f,
+             cv::INTER_LINEAR);
+  if (resize_w < imgW) {
+    cv::copyMakeBorder(resize_img, resize_img, 0, 0, 0, imgW - resize_w,
+                       cv::BORDER_CONSTANT, cv::Scalar(0, 0, 0));
+  }
+}
+
+} // namespace PaddleOCR

deploy/cpp_infer/src_rec/utility.cpp

+// Copyright (c) 2020 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 <dirent.h>
+#include <include/utility.h>
+#include <iostream>
+#include <ostream>
+#include <sys/stat.h>
+#include <sys/types.h>
+#include <vector>
+
+namespace PaddleOCR {
+
+std::vector<std::string> Utility::ReadDict(const std::string &path) {
+  std::ifstream in(path);
+  std::string line;
+  std::vector<std::string> m_vec;
+  if (in) {
+    while (getline(in, line)) {
+      m_vec.push_back(line);
+    }
+  } else {
+    std::cout << "no such label file: " << path << ", exit the program..."
+              << std::endl;
+    exit(1);
+  }
+  return m_vec;
+}
+
+void Utility::VisualizeBboxes(
+    const cv::Mat &srcimg,
+    const std::vector<std::vector<std::vector<int>>> &boxes) {
+  cv::Mat img_vis;
+  srcimg.copyTo(img_vis);
+  for (int n = 0; n < boxes.size(); n++) {
+    cv::Point rook_points[4];
+    for (int m = 0; m < boxes[n].size(); m++) {
+      rook_points[m] = cv::Point(int(boxes[n][m][0]), int(boxes[n][m][1]));
+    }
+
+    const cv::Point *ppt[1] = {rook_points};
+    int npt[] = {4};
+    cv::polylines(img_vis, ppt, npt, 1, 1, CV_RGB(0, 255, 0), 2, 8, 0);
+  }
+
+  cv::imwrite("./ocr_vis.png", img_vis);
+  std::cout << "The detection visualized image saved in ./ocr_vis.png"
+            << std::endl;
+}
+
+// list all files under a directory
+void Utility::GetAllFiles(const char *dir_name,
+                          std::vector<std::string> &all_inputs) {
+  if (NULL == dir_name) {
+    std::cout << " dir_name is null ! " << std::endl;
+    return;
+  }
+  struct stat s;
+  lstat(dir_name, &s);
+  if (!S_ISDIR(s.st_mode)) {
+    std::cout << "dir_name is not a valid directory !" << std::endl;
+    all_inputs.push_back(dir_name);
+    return;
+  } else {
+    struct dirent *filename; // return value for readdir()
+    DIR *dir;                // return value for opendir()
+    dir = opendir(dir_name);
+    if (NULL == dir) {
+      std::cout << "Can not open dir " << dir_name << std::endl;
+      return;
+    }
+    std::cout << "Successfully opened the dir !" << std::endl;
+    while ((filename = readdir(dir)) != NULL) {
+      if (strcmp(filename->d_name, ".") == 0 ||
+          strcmp(filename->d_name, "..") == 0)
+        continue;
+      // img_dir + std::string("/") + all_inputs[0];
+      all_inputs.push_back(dir_name + std::string("/") +
+                           std::string(filename->d_name));
+    }
+  }
+}
+
+} // namespace PaddleOCR
\ No newline at end of file

deploy/cpp_infer/src_system/main.cpp

+// Copyright (c) 2020 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 "glog/logging.h"
+#include "omp.h"
+#include "opencv2/core.hpp"
+#include "opencv2/imgcodecs.hpp"
+#include "opencv2/imgproc.hpp"
+#include <chrono>
+#include <iomanip>
+#include <iostream>
+#include <ostream>
+#include <vector>
+
+#include <cstring>
+#include <fstream>
+#include <numeric>
+
+#include <glog/logging.h>
+// #include <include/config.h>
+#include <include/ocr_det.h>
+#include <include/ocr_rec.h>
+// #include <include/utility.h>
+#include <sys/stat.h>
+
+#include <gflags/gflags.h>
+
+DEFINE_bool(use_gpu, false, "Infering with GPU or CPU.");
+DEFINE_int32(gpu_id, 0, "Device id of GPU to execute.");
+DEFINE_int32(gpu_mem, 4000, "GPU id when infering with GPU.");
+DEFINE_int32(cpu_math_library_num_threads, 10, "Num of threads with CPU.");
+DEFINE_bool(use_mkldnn, false, "Whether use mkldnn with CPU.");
+
+DEFINE_string(image_dir, "", "Dir of input image.");
+DEFINE_string(det_model_dir, "", "Path of det inference model.");
+DEFINE_int32(max_side_len, 960, "max_side_len of input image.");
+DEFINE_double(det_db_thresh, 0.3, "Threshold of det_db_thresh.");
+DEFINE_double(det_db_box_thresh, 0.5, "Threshold of det_db_box_thresh.");
+DEFINE_double(det_db_unclip_ratio, 1.6, "Threshold of det_db_unclip_ratio.");
+DEFINE_bool(use_polygon_score, false, "Whether use polygon score.");
+DEFINE_bool(visualize, true, "Whether show the detection results.");
+
+DEFINE_bool(use_angle_cls, false, "Whether use use_angle_cls.");
+DEFINE_string(cls_model_dir, "", "Path of cls inference model.");
+DEFINE_double(cls_thresh, 0.9, "Threshold of cls_thresh.");
+
+DEFINE_string(rec_model_dir, "", "Path of rec inference model.");
+DEFINE_string(char_list_file, "../../ppocr/utils/ppocr_keys_v1.txt", "Path of dictionary.");
+
+DEFINE_bool(use_tensorrt, false, "Whether use tensorrt.");
+DEFINE_bool(use_fp16, false, "Whether use fp16 when use tensorrt.");
+
+using namespace std;
+using namespace cv;
+using namespace PaddleOCR;
+
+
+static bool PathExists(const std::string& path){
+#ifdef _WIN32
+  struct _stat buffer;
+  return (_stat(path.c_str(), &buffer) == 0);
+#else
+  struct stat buffer;
+  return (stat(path.c_str(), &buffer) == 0);
+#endif  // !_WIN32
+}
+
+
+cv::Mat GetRotateCropImage(const cv::Mat &srcimage,
+                            std::vector<std::vector<int>> box) {
+  cv::Mat image;
+  srcimage.copyTo(image);
+  std::vector<std::vector<int>> points = box;
+
+  int x_collect[4] = {box[0][0], box[1][0], box[2][0], box[3][0]};
+  int y_collect[4] = {box[0][1], box[1][1], box[2][1], box[3][1]};
+  int left = int(*std::min_element(x_collect, x_collect + 4));
+  int right = int(*std::max_element(x_collect, x_collect + 4));
+  int top = int(*std::min_element(y_collect, y_collect + 4));
+  int bottom = int(*std::max_element(y_collect, y_collect + 4));
+
+  cv::Mat img_crop;
+  image(cv::Rect(left, top, right - left, bottom - top)).copyTo(img_crop);
+
+  for (int i = 0; i < points.size(); i++) {
+    points[i][0] -= left;
+    points[i][1] -= top;
+  }
+
+  int img_crop_width = int(sqrt(pow(points[0][0] - points[1][0], 2) +
+                                pow(points[0][1] - points[1][1], 2)));
+  int img_crop_height = int(sqrt(pow(points[0][0] - points[3][0], 2) +
+                                 pow(points[0][1] - points[3][1], 2)));
+
+  cv::Point2f pts_std[4];
+  pts_std[0] = cv::Point2f(0., 0.);
+  pts_std[1] = cv::Point2f(img_crop_width, 0.);
+  pts_std[2] = cv::Point2f(img_crop_width, img_crop_height);
+  pts_std[3] = cv::Point2f(0.f, img_crop_height);
+
+  cv::Point2f pointsf[4];
+  pointsf[0] = cv::Point2f(points[0][0], points[0][1]);
+  pointsf[1] = cv::Point2f(points[1][0], points[1][1]);
+  pointsf[2] = cv::Point2f(points[2][0], points[2][1]);
+  pointsf[3] = cv::Point2f(points[3][0], points[3][1]);
+
+  cv::Mat M = cv::getPerspectiveTransform(pointsf, pts_std);
+
+  cv::Mat dst_img;
+  cv::warpPerspective(img_crop, dst_img, M,
+                      cv::Size(img_crop_width, img_crop_height),
+                      cv::BORDER_REPLICATE);
+
+  if (float(dst_img.rows) >= float(dst_img.cols) * 1.5) {
+    cv::Mat srcCopy = cv::Mat(dst_img.rows, dst_img.cols, dst_img.depth());
+    cv::transpose(dst_img, srcCopy);
+    cv::flip(srcCopy, srcCopy, 0);
+    return srcCopy;
+  } else {
+    return dst_img;
+  }
+}
+
+
+int main(int argc, char **argv) {
+  // Parsing command-line
+  google::ParseCommandLineFlags(&argc, &argv, true);
+  if ((FLAGS_det_model_dir.empty() || FLAGS_rec_model_dir.empty() || FLAGS_image_dir.empty()) ||
+     (FLAGS_use_angle_cls && FLAGS_cls_model_dir.empty())) {
+    std::cout << "Usage[default]: ./ocr_system --det_model_dir=/PATH/TO/DET_INFERENCE_MODEL/ "
+                << "--rec_model_dir=/PATH/TO/REC_INFERENCE_MODEL/ "
+                << "--image_dir=/PATH/TO/INPUT/IMAGE/" << std::endl;
+    std::cout << "Usage[use angle cls]: ./ocr_system --det_model_dir=/PATH/TO/DET_INFERENCE_MODEL/ "
+                << "--use_angle_cls=true "
+                << "--cls_model_dir=/PATH/TO/CLS_INFERENCE_MODEL/ "
+                << "--rec_model_dir=/PATH/TO/REC_INFERENCE_MODEL/ "
+                << "--image_dir=/PATH/TO/INPUT/IMAGE/" << std::endl;
+    return -1;
+  }
+
+  if (!PathExists(FLAGS_image_dir)) {
+      std::cerr << "[ERROR] image path not exist! image_dir: " << FLAGS_image_dir << endl;
+      exit(1);      
+  }
+  std::vector<cv::String> cv_all_img_names;
+  cv::glob(FLAGS_image_dir, cv_all_img_names);
+  std::cout << "total images num: " << cv_all_img_names.size() << endl;
+
+  DBDetector det(FLAGS_det_model_dir, FLAGS_use_gpu, FLAGS_gpu_id,
+                 FLAGS_gpu_mem, FLAGS_cpu_math_library_num_threads, 
+                 FLAGS_use_mkldnn, FLAGS_max_side_len, FLAGS_det_db_thresh,
+                 FLAGS_det_db_box_thresh, FLAGS_det_db_unclip_ratio,
+                 FLAGS_use_polygon_score, FLAGS_visualize,
+                 FLAGS_use_tensorrt, FLAGS_use_fp16);
+
+  Classifier *cls = nullptr;
+  if (FLAGS_use_angle_cls) {
+    cls = new Classifier(FLAGS_cls_model_dir, FLAGS_use_gpu, FLAGS_gpu_id,
+                         FLAGS_gpu_mem, FLAGS_cpu_math_library_num_threads,
+                         FLAGS_use_mkldnn, FLAGS_cls_thresh,
+                         FLAGS_use_tensorrt, FLAGS_use_fp16);
+  }
+
+  CRNNRecognizer rec(FLAGS_rec_model_dir, FLAGS_use_gpu, FLAGS_gpu_id,
+                     FLAGS_gpu_mem, FLAGS_cpu_math_library_num_threads,
+                     FLAGS_use_mkldnn, FLAGS_char_list_file,
+                     FLAGS_use_tensorrt, FLAGS_use_fp16);
+
+  auto start = std::chrono::system_clock::now();
+
+  for (int i = 0; i < cv_all_img_names.size(); ++i) {
+    LOG(INFO) << "The predict img: " << cv_all_img_names[i];
+
+    cv::Mat srcimg = cv::imread(FLAGS_image_dir, cv::IMREAD_COLOR);
+    if (!srcimg.data) {
+      std::cerr << "[ERROR] image read failed! image path: " << cv_all_img_names[i] << endl;
+      exit(1);
+    }
+    std::vector<std::vector<std::vector<int>>> boxes;
+
+    det.Run(srcimg, boxes);
+  
+    cv::Mat crop_img;
+    for (int j = 0; j < boxes.size(); j++) {
+      crop_img = GetRotateCropImage(srcimg, boxes[j]);
+
+      if (cls != nullptr) {
+        crop_img = cls->Run(crop_img);
+      }
+      rec.Run(crop_img);
+    }
+      
+    auto end = std::chrono::system_clock::now();
+    auto duration =
+        std::chrono::duration_cast<std::chrono::microseconds>(end - start);
+    std::cout << "Cost  "
+              << double(duration.count()) *
+                     std::chrono::microseconds::period::num /
+                     std::chrono::microseconds::period::den
+              << "s" << std::endl;
+  }
+
+  return 0;
+}

deploy/cpp_infer/src_system/ocr_det.cpp

+// Copyright (c) 2020 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 <include/ocr_det.h>
+
+namespace PaddleOCR {
+
+void DBDetector::LoadModel(const std::string &model_dir) {
+  //   AnalysisConfig config;
+  paddle_infer::Config config;
+  config.SetModel(model_dir + "/inference.pdmodel",
+                  model_dir + "/inference.pdiparams");
+
+  if (this->use_gpu_) {
+    config.EnableUseGpu(this->gpu_mem_, this->gpu_id_);
+    if (this->use_tensorrt_) {
+      config.EnableTensorRtEngine(
+          1 << 20, 10, 3,
+          this->use_fp16_ ? paddle_infer::Config::Precision::kHalf
+                          : paddle_infer::Config::Precision::kFloat32,
+          false, false);
+      std::map<std::string, std::vector<int>> min_input_shape = {
+          {"x", {1, 3, 50, 50}},
+          {"conv2d_92.tmp_0", {1, 96, 20, 20}},
+          {"conv2d_91.tmp_0", {1, 96, 10, 10}},
+          {"nearest_interp_v2_1.tmp_0", {1, 96, 10, 10}},
+          {"nearest_interp_v2_2.tmp_0", {1, 96, 20, 20}},
+          {"nearest_interp_v2_3.tmp_0", {1, 24, 20, 20}},
+          {"nearest_interp_v2_4.tmp_0", {1, 24, 20, 20}},
+          {"nearest_interp_v2_5.tmp_0", {1, 24, 20, 20}},
+          {"elementwise_add_7", {1, 56, 2, 2}},
+          {"nearest_interp_v2_0.tmp_0", {1, 96, 2, 2}}};
+      std::map<std::string, std::vector<int>> max_input_shape = {
+          {"x", {1, 3, this->max_side_len_, this->max_side_len_}},
+          {"conv2d_92.tmp_0", {1, 96, 400, 400}},
+          {"conv2d_91.tmp_0", {1, 96, 200, 200}},
+          {"nearest_interp_v2_1.tmp_0", {1, 96, 200, 200}},
+          {"nearest_interp_v2_2.tmp_0", {1, 96, 400, 400}},
+          {"nearest_interp_v2_3.tmp_0", {1, 24, 400, 400}},
+          {"nearest_interp_v2_4.tmp_0", {1, 24, 400, 400}},
+          {"nearest_interp_v2_5.tmp_0", {1, 24, 400, 400}},
+          {"elementwise_add_7", {1, 56, 400, 400}},
+          {"nearest_interp_v2_0.tmp_0", {1, 96, 400, 400}}};
+      std::map<std::string, std::vector<int>> opt_input_shape = {
+          {"x", {1, 3, 640, 640}},
+          {"conv2d_92.tmp_0", {1, 96, 160, 160}},
+          {"conv2d_91.tmp_0", {1, 96, 80, 80}},
+          {"nearest_interp_v2_1.tmp_0", {1, 96, 80, 80}},
+          {"nearest_interp_v2_2.tmp_0", {1, 96, 160, 160}},
+          {"nearest_interp_v2_3.tmp_0", {1, 24, 160, 160}},
+          {"nearest_interp_v2_4.tmp_0", {1, 24, 160, 160}},
+          {"nearest_interp_v2_5.tmp_0", {1, 24, 160, 160}},
+          {"elementwise_add_7", {1, 56, 40, 40}},
+          {"nearest_interp_v2_0.tmp_0", {1, 96, 40, 40}}};
+
+      config.SetTRTDynamicShapeInfo(min_input_shape, max_input_shape,
+                                    opt_input_shape);
+    }
+  } else {
+    config.DisableGpu();
+    if (this->use_mkldnn_) {
+      config.EnableMKLDNN();
+      // cache 10 different shapes for mkldnn to avoid memory leak
+      config.SetMkldnnCacheCapacity(10);
+    }
+    config.SetCpuMathLibraryNumThreads(this->cpu_math_library_num_threads_);
+  }
+  // use zero_copy_run as default
+  config.SwitchUseFeedFetchOps(false);
+  // true for multiple input
+  config.SwitchSpecifyInputNames(true);
+
+  config.SwitchIrOptim(true);
+
+  config.EnableMemoryOptim();
+  // config.DisableGlogInfo();
+
+  this->predictor_ = CreatePredictor(config);
+}
+
+void DBDetector::Run(cv::Mat &img,
+                     std::vector<std::vector<std::vector<int>>> &boxes) {
+  float ratio_h{};
+  float ratio_w{};
+
+  cv::Mat srcimg;
+  cv::Mat resize_img;
+  img.copyTo(srcimg);
+  this->resize_op_.Run(img, resize_img, this->max_side_len_, ratio_h, ratio_w,
+                       this->use_tensorrt_);
+
+  this->normalize_op_.Run(&resize_img, this->mean_, this->scale_,
+                          this->is_scale_);
+
+  std::vector<float> input(1 * 3 * resize_img.rows * resize_img.cols, 0.0f);
+  this->permute_op_.Run(&resize_img, input.data());
+
+  // Inference.
+  auto input_names = this->predictor_->GetInputNames();
+  auto input_t = this->predictor_->GetInputHandle(input_names[0]);
+  input_t->Reshape({1, 3, resize_img.rows, resize_img.cols});
+  input_t->CopyFromCpu(input.data());
+  this->predictor_->Run();
+
+  std::vector<float> out_data;
+  auto output_names = this->predictor_->GetOutputNames();
+  auto output_t = this->predictor_->GetOutputHandle(output_names[0]);
+  std::vector<int> output_shape = output_t->shape();
+  int out_num = std::accumulate(output_shape.begin(), output_shape.end(), 1,
+                                std::multiplies<int>());
+
+  out_data.resize(out_num);
+  output_t->CopyToCpu(out_data.data());
+
+  int n2 = output_shape[2];
+  int n3 = output_shape[3];
+  int n = n2 * n3;
+
+  std::vector<float> pred(n, 0.0);
+  std::vector<unsigned char> cbuf(n, ' ');
+
+  for (int i = 0; i < n; i++) {
+    pred[i] = float(out_data[i]);
+    cbuf[i] = (unsigned char)((out_data[i]) * 255);
+  }
+
+  cv::Mat cbuf_map(n2, n3, CV_8UC1, (unsigned char *)cbuf.data());
+  cv::Mat pred_map(n2, n3, CV_32F, (float *)pred.data());
+
+  const double threshold = this->det_db_thresh_ * 255;
+  const double maxvalue = 255;
+  cv::Mat bit_map;
+  cv::threshold(cbuf_map, bit_map, threshold, maxvalue, cv::THRESH_BINARY);
+  cv::Mat dilation_map;
+  cv::Mat dila_ele = cv::getStructuringElement(cv::MORPH_RECT, cv::Size(2, 2));
+  cv::dilate(bit_map, dilation_map, dila_ele);
+  boxes = post_processor_.BoxesFromBitmap(
+      pred_map, dilation_map, this->det_db_box_thresh_,
+      this->det_db_unclip_ratio_, this->use_polygon_score_);
+
+  boxes = post_processor_.FilterTagDetRes(boxes, ratio_h, ratio_w, srcimg);
+  std::cout << "Detected boxes num: " << boxes.size() << endl;
+    
+  //// visualization
+  if (this->visualize_) {
+    Utility::VisualizeBboxes(srcimg, boxes);
+  }
+}
+
+} // namespace PaddleOCR

deploy/cpp_infer/src_system/ocr_rec.cpp

+// Copyright (c) 2020 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 <include/ocr_rec.h>
+
+namespace PaddleOCR {
+
+void CRNNRecognizer::Run(cv::Mat &img) {
+  cv::Mat srcimg;
+  img.copyTo(srcimg);
+  cv::Mat resize_img;
+
+  float wh_ratio = float(srcimg.cols) / float(srcimg.rows);
+
+  this->resize_op_.Run(srcimg, resize_img, wh_ratio, this->use_tensorrt_);
+
+  this->normalize_op_.Run(&resize_img, this->mean_, this->scale_,
+                          this->is_scale_);
+
+  std::vector<float> input(1 * 3 * resize_img.rows * resize_img.cols, 0.0f);
+
+  this->permute_op_.Run(&resize_img, input.data());
+
+  // Inference.
+  auto input_names = this->predictor_->GetInputNames();
+  auto input_t = this->predictor_->GetInputHandle(input_names[0]);
+  input_t->Reshape({1, 3, resize_img.rows, resize_img.cols});
+  input_t->CopyFromCpu(input.data());
+  this->predictor_->Run();
+
+  std::vector<float> predict_batch;
+  auto output_names = this->predictor_->GetOutputNames();
+  auto output_t = this->predictor_->GetOutputHandle(output_names[0]);
+  auto predict_shape = output_t->shape();
+
+  int out_num = std::accumulate(predict_shape.begin(), predict_shape.end(), 1,
+                                std::multiplies<int>());
+  predict_batch.resize(out_num);
+
+  output_t->CopyToCpu(predict_batch.data());
+
+  // ctc decode
+  std::vector<std::string> str_res;
+  int argmax_idx;
+  int last_index = 0;
+  float score = 0.f;
+  int count = 0;
+  float max_value = 0.0f;
+
+  for (int n = 0; n < predict_shape[1]; n++) {
+    argmax_idx =
+        int(Utility::argmax(&predict_batch[n * predict_shape[2]],
+                            &predict_batch[(n + 1) * predict_shape[2]]));
+    max_value =
+        float(*std::max_element(&predict_batch[n * predict_shape[2]],
+                                &predict_batch[(n + 1) * predict_shape[2]]));
+
+    if (argmax_idx > 0 && (!(n > 0 && argmax_idx == last_index))) {
+      score += max_value;
+      count += 1;
+      str_res.push_back(label_list_[argmax_idx]);
+    }
+    last_index = argmax_idx;
+  }
+  score /= count;
+  for (int i = 0; i < str_res.size(); i++) {
+    std::cout << str_res[i];
+  }
+  std::cout << "\tscore: " << score << std::endl;
+}
+
+void CRNNRecognizer::LoadModel(const std::string &model_dir) {
+  //   AnalysisConfig config;
+  paddle_infer::Config config;
+  config.SetModel(model_dir + "/inference.pdmodel",
+                  model_dir + "/inference.pdiparams");
+
+  if (this->use_gpu_) {
+    config.EnableUseGpu(this->gpu_mem_, this->gpu_id_);
+    if (this->use_tensorrt_) {
+      config.EnableTensorRtEngine(
+          1 << 20, 10, 3,
+          this->use_fp16_ ? paddle_infer::Config::Precision::kHalf
+                          : paddle_infer::Config::Precision::kFloat32,
+          false, false);
+      std::map<std::string, std::vector<int>> min_input_shape = {
+          {"x", {1, 3, 32, 10}}};
+      std::map<std::string, std::vector<int>> max_input_shape = {
+          {"x", {1, 3, 32, 2000}}};
+      std::map<std::string, std::vector<int>> opt_input_shape = {
+          {"x", {1, 3, 32, 320}}};
+
+      config.SetTRTDynamicShapeInfo(min_input_shape, max_input_shape,
+                                    opt_input_shape);
+    }
+  } else {
+    config.DisableGpu();
+    if (this->use_mkldnn_) {
+      config.EnableMKLDNN();
+      // cache 10 different shapes for mkldnn to avoid memory leak
+      config.SetMkldnnCacheCapacity(10);
+    }
+    config.SetCpuMathLibraryNumThreads(this->cpu_math_library_num_threads_);
+  }
+
+  config.SwitchUseFeedFetchOps(false);
+  // true for multiple input
+  config.SwitchSpecifyInputNames(true);
+
+  config.SwitchIrOptim(true);
+
+  config.EnableMemoryOptim();
+  config.DisableGlogInfo();
+
+  this->predictor_ = CreatePredictor(config);
+}
+
+cv::Mat CRNNRecognizer::GetRotateCropImage(const cv::Mat &srcimage,
+                                           std::vector<std::vector<int>> box) {
+  cv::Mat image;
+  srcimage.copyTo(image);
+  std::vector<std::vector<int>> points = box;
+
+  int x_collect[4] = {box[0][0], box[1][0], box[2][0], box[3][0]};
+  int y_collect[4] = {box[0][1], box[1][1], box[2][1], box[3][1]};
+  int left = int(*std::min_element(x_collect, x_collect + 4));
+  int right = int(*std::max_element(x_collect, x_collect + 4));
+  int top = int(*std::min_element(y_collect, y_collect + 4));
+  int bottom = int(*std::max_element(y_collect, y_collect + 4));
+
+  cv::Mat img_crop;
+  image(cv::Rect(left, top, right - left, bottom - top)).copyTo(img_crop);
+
+  for (int i = 0; i < points.size(); i++) {
+    points[i][0] -= left;
+    points[i][1] -= top;
+  }
+
+  int img_crop_width = int(sqrt(pow(points[0][0] - points[1][0], 2) +
+                                pow(points[0][1] - points[1][1], 2)));
+  int img_crop_height = int(sqrt(pow(points[0][0] - points[3][0], 2) +
+                                 pow(points[0][1] - points[3][1], 2)));
+
+  cv::Point2f pts_std[4];
+  pts_std[0] = cv::Point2f(0., 0.);
+  pts_std[1] = cv::Point2f(img_crop_width, 0.);
+  pts_std[2] = cv::Point2f(img_crop_width, img_crop_height);
+  pts_std[3] = cv::Point2f(0.f, img_crop_height);
+
+  cv::Point2f pointsf[4];
+  pointsf[0] = cv::Point2f(points[0][0], points[0][1]);
+  pointsf[1] = cv::Point2f(points[1][0], points[1][1]);
+  pointsf[2] = cv::Point2f(points[2][0], points[2][1]);
+  pointsf[3] = cv::Point2f(points[3][0], points[3][1]);
+
+  cv::Mat M = cv::getPerspectiveTransform(pointsf, pts_std);
+
+  cv::Mat dst_img;
+  cv::warpPerspective(img_crop, dst_img, M,
+                      cv::Size(img_crop_width, img_crop_height),
+                      cv::BORDER_REPLICATE);
+
+  if (float(dst_img.rows) >= float(dst_img.cols) * 1.5) {
+    cv::Mat srcCopy = cv::Mat(dst_img.rows, dst_img.cols, dst_img.depth());
+    cv::transpose(dst_img, srcCopy);
+    cv::flip(srcCopy, srcCopy, 0);
+    return srcCopy;
+  } else {
+    return dst_img;
+  }
+}
+
+} // namespace PaddleOCR

deploy/cpp_infer/src_system/postprocess_op.cpp

+// Copyright (c) 2020 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 <include/postprocess_op.h>
+
+namespace PaddleOCR {
+
+void PostProcessor::GetContourArea(const std::vector<std::vector<float>> &box,
+                                   float unclip_ratio, float &distance) {
+  int pts_num = 4;
+  float area = 0.0f;
+  float dist = 0.0f;
+  for (int i = 0; i < pts_num; i++) {
+    area += box[i][0] * box[(i + 1) % pts_num][1] -
+            box[i][1] * box[(i + 1) % pts_num][0];
+    dist += sqrtf((box[i][0] - box[(i + 1) % pts_num][0]) *
+                      (box[i][0] - box[(i + 1) % pts_num][0]) +
+                  (box[i][1] - box[(i + 1) % pts_num][1]) *
+                      (box[i][1] - box[(i + 1) % pts_num][1]));
+  }
+  area = fabs(float(area / 2.0));
+
+  distance = area * unclip_ratio / dist;
+}
+
+cv::RotatedRect PostProcessor::UnClip(std::vector<std::vector<float>> box,
+                                      const float &unclip_ratio) {
+  float distance = 1.0;
+
+  GetContourArea(box, unclip_ratio, distance);
+
+  ClipperLib::ClipperOffset offset;
+  ClipperLib::Path p;
+  p << ClipperLib::IntPoint(int(box[0][0]), int(box[0][1]))
+    << ClipperLib::IntPoint(int(box[1][0]), int(box[1][1]))
+    << ClipperLib::IntPoint(int(box[2][0]), int(box[2][1]))
+    << ClipperLib::IntPoint(int(box[3][0]), int(box[3][1]));
+  offset.AddPath(p, ClipperLib::jtRound, ClipperLib::etClosedPolygon);
+
+  ClipperLib::Paths soln;
+  offset.Execute(soln, distance);
+  std::vector<cv::Point2f> points;
+
+  for (int j = 0; j < soln.size(); j++) {
+    for (int i = 0; i < soln[soln.size() - 1].size(); i++) {
+      points.emplace_back(soln[j][i].X, soln[j][i].Y);
+    }
+  }
+  cv::RotatedRect res;
+  if (points.size() <= 0) {
+    res = cv::RotatedRect(cv::Point2f(0, 0), cv::Size2f(1, 1), 0);
+  } else {
+    res = cv::minAreaRect(points);
+  }
+  return res;
+}
+
+float **PostProcessor::Mat2Vec(cv::Mat mat) {
+  auto **array = new float *[mat.rows];
+  for (int i = 0; i < mat.rows; ++i)
+    array[i] = new float[mat.cols];
+  for (int i = 0; i < mat.rows; ++i) {
+    for (int j = 0; j < mat.cols; ++j) {
+      array[i][j] = mat.at<float>(i, j);
+    }
+  }
+
+  return array;
+}
+
+std::vector<std::vector<int>>
+PostProcessor::OrderPointsClockwise(std::vector<std::vector<int>> pts) {
+  std::vector<std::vector<int>> box = pts;
+  std::sort(box.begin(), box.end(), XsortInt);
+
+  std::vector<std::vector<int>> leftmost = {box[0], box[1]};
+  std::vector<std::vector<int>> rightmost = {box[2], box[3]};
+
+  if (leftmost[0][1] > leftmost[1][1])
+    std::swap(leftmost[0], leftmost[1]);
+
+  if (rightmost[0][1] > rightmost[1][1])
+    std::swap(rightmost[0], rightmost[1]);
+
+  std::vector<std::vector<int>> rect = {leftmost[0], rightmost[0], rightmost[1],
+                                        leftmost[1]};
+  return rect;
+}
+
+std::vector<std::vector<float>> PostProcessor::Mat2Vector(cv::Mat mat) {
+  std::vector<std::vector<float>> img_vec;
+  std::vector<float> tmp;
+
+  for (int i = 0; i < mat.rows; ++i) {
+    tmp.clear();
+    for (int j = 0; j < mat.cols; ++j) {
+      tmp.push_back(mat.at<float>(i, j));
+    }
+    img_vec.push_back(tmp);
+  }
+  return img_vec;
+}
+
+bool PostProcessor::XsortFp32(std::vector<float> a, std::vector<float> b) {
+  if (a[0] != b[0])
+    return a[0] < b[0];
+  return false;
+}
+
+bool PostProcessor::XsortInt(std::vector<int> a, std::vector<int> b) {
+  if (a[0] != b[0])
+    return a[0] < b[0];
+  return false;
+}
+
+std::vector<std::vector<float>> PostProcessor::GetMiniBoxes(cv::RotatedRect box,
+                                                            float &ssid) {
+  ssid = std::max(box.size.width, box.size.height);
+
+  cv::Mat points;
+  cv::boxPoints(box, points);
+
+  auto array = Mat2Vector(points);
+  std::sort(array.begin(), array.end(), XsortFp32);
+
+  std::vector<float> idx1 = array[0], idx2 = array[1], idx3 = array[2],
+                     idx4 = array[3];
+  if (array[3][1] <= array[2][1]) {
+    idx2 = array[3];
+    idx3 = array[2];
+  } else {
+    idx2 = array[2];
+    idx3 = array[3];
+  }
+  if (array[1][1] <= array[0][1]) {
+    idx1 = array[1];
+    idx4 = array[0];
+  } else {
+    idx1 = array[0];
+    idx4 = array[1];
+  }
+
+  array[0] = idx1;
+  array[1] = idx2;
+  array[2] = idx3;
+  array[3] = idx4;
+
+  return array;
+}
+
+float PostProcessor::PolygonScoreAcc(std::vector<cv::Point> contour,
+                                     cv::Mat pred) {
+  int width = pred.cols;
+  int height = pred.rows;
+  std::vector<float> box_x;
+  std::vector<float> box_y;
+  for (int i = 0; i < contour.size(); ++i) {
+    box_x.push_back(contour[i].x);
+    box_y.push_back(contour[i].y);
+  }
+
+  int xmin =
+      clamp(int(std::floor(*(std::min_element(box_x.begin(), box_x.end())))), 0,
+            width - 1);
+  int xmax =
+      clamp(int(std::ceil(*(std::max_element(box_x.begin(), box_x.end())))), 0,
+            width - 1);
+  int ymin =
+      clamp(int(std::floor(*(std::min_element(box_y.begin(), box_y.end())))), 0,
+            height - 1);
+  int ymax =
+      clamp(int(std::ceil(*(std::max_element(box_y.begin(), box_y.end())))), 0,
+            height - 1);
+
+  cv::Mat mask;
+  mask = cv::Mat::zeros(ymax - ymin + 1, xmax - xmin + 1, CV_8UC1);
+
+
+  cv::Point* rook_point = new cv::Point[contour.size()];
+   
+  for (int i = 0; i < contour.size(); ++i) {
+    rook_point[i] = cv::Point(int(box_x[i]) - xmin, int(box_y[i]) - ymin);
+  }
+  const cv::Point *ppt[1] = {rook_point};
+  int npt[] = {int(contour.size())};
+
+
+  cv::fillPoly(mask, ppt, npt, 1, cv::Scalar(1));
+
+  cv::Mat croppedImg;
+  pred(cv::Rect(xmin, ymin, xmax - xmin + 1, ymax - ymin + 1)).copyTo(croppedImg);
+  float score = cv::mean(croppedImg, mask)[0];
+
+  delete []rook_point;
+  return score;
+}
+
+float PostProcessor::BoxScoreFast(std::vector<std::vector<float>> box_array,
+                                  cv::Mat pred) {
+  auto array = box_array;
+  int width = pred.cols;
+  int height = pred.rows;
+
+  float box_x[4] = {array[0][0], array[1][0], array[2][0], array[3][0]};
+  float box_y[4] = {array[0][1], array[1][1], array[2][1], array[3][1]};
+
+  int xmin = clamp(int(std::floor(*(std::min_element(box_x, box_x + 4)))), 0,
+                   width - 1);
+  int xmax = clamp(int(std::ceil(*(std::max_element(box_x, box_x + 4)))), 0,
+                   width - 1);
+  int ymin = clamp(int(std::floor(*(std::min_element(box_y, box_y + 4)))), 0,
+                   height - 1);
+  int ymax = clamp(int(std::ceil(*(std::max_element(box_y, box_y + 4)))), 0,
+                   height - 1);
+
+  cv::Mat mask;
+  mask = cv::Mat::zeros(ymax - ymin + 1, xmax - xmin + 1, CV_8UC1);
+
+  cv::Point root_point[4];
+  root_point[0] = cv::Point(int(array[0][0]) - xmin, int(array[0][1]) - ymin);
+  root_point[1] = cv::Point(int(array[1][0]) - xmin, int(array[1][1]) - ymin);
+  root_point[2] = cv::Point(int(array[2][0]) - xmin, int(array[2][1]) - ymin);
+  root_point[3] = cv::Point(int(array[3][0]) - xmin, int(array[3][1]) - ymin);
+  const cv::Point *ppt[1] = {root_point};
+  int npt[] = {4};
+  cv::fillPoly(mask, ppt, npt, 1, cv::Scalar(1));
+
+  cv::Mat croppedImg;
+  pred(cv::Rect(xmin, ymin, xmax - xmin + 1, ymax - ymin + 1))
+      .copyTo(croppedImg);
+
+  auto score = cv::mean(croppedImg, mask)[0];
+  return score;
+}
+
+std::vector<std::vector<std::vector<int>>> PostProcessor::BoxesFromBitmap(
+    const cv::Mat pred, const cv::Mat bitmap, const float &box_thresh,
+    const float &det_db_unclip_ratio, const bool &use_polygon_score) {
+  const int min_size = 3;
+  const int max_candidates = 1000;
+
+  int width = bitmap.cols;
+  int height = bitmap.rows;
+
+  std::vector<std::vector<cv::Point>> contours;
+  std::vector<cv::Vec4i> hierarchy;
+
+  cv::findContours(bitmap, contours, hierarchy, cv::RETR_LIST,
+                   cv::CHAIN_APPROX_SIMPLE);
+
+  int num_contours =
+      contours.size() >= max_candidates ? max_candidates : contours.size();
+
+  std::vector<std::vector<std::vector<int>>> boxes;
+
+  for (int _i = 0; _i < num_contours; _i++) {
+    if (contours[_i].size() <= 2) {
+      continue;
+    }
+    float ssid;
+    cv::RotatedRect box = cv::minAreaRect(contours[_i]);
+    auto array = GetMiniBoxes(box, ssid);
+
+    auto box_for_unclip = array;
+    // end get_mini_box
+
+    if (ssid < min_size) {
+      continue;
+    }
+
+    float score;
+    if (use_polygon_score)
+      /* compute using polygon*/
+      score = PolygonScoreAcc(contours[_i], pred);
+    else
+      score = BoxScoreFast(array, pred);
+
+    if (score < box_thresh)
+      continue;
+
+    // start for unclip
+    cv::RotatedRect points = UnClip(box_for_unclip, det_db_unclip_ratio);
+    if (points.size.height < 1.001 && points.size.width < 1.001) {
+      continue;
+    }
+    // end for unclip
+
+    cv::RotatedRect clipbox = points;
+    auto cliparray = GetMiniBoxes(clipbox, ssid);
+
+    if (ssid < min_size + 2)
+      continue;
+
+    int dest_width = pred.cols;
+    int dest_height = pred.rows;
+    std::vector<std::vector<int>> intcliparray;
+
+    for (int num_pt = 0; num_pt < 4; num_pt++) {
+      std::vector<int> a{int(clampf(roundf(cliparray[num_pt][0] / float(width) *
+                                           float(dest_width)),
+                                    0, float(dest_width))),
+                         int(clampf(roundf(cliparray[num_pt][1] /
+                                           float(height) * float(dest_height)),
+                                    0, float(dest_height)))};
+      intcliparray.push_back(a);
+    }
+    boxes.push_back(intcliparray);
+
+  } // end for
+  return boxes;
+}
+
+std::vector<std::vector<std::vector<int>>>
+PostProcessor::FilterTagDetRes(std::vector<std::vector<std::vector<int>>> boxes,
+                               float ratio_h, float ratio_w, cv::Mat srcimg) {
+  int oriimg_h = srcimg.rows;
+  int oriimg_w = srcimg.cols;
+
+  std::vector<std::vector<std::vector<int>>> root_points;
+  for (int n = 0; n < boxes.size(); n++) {
+    boxes[n] = OrderPointsClockwise(boxes[n]);
+    for (int m = 0; m < boxes[0].size(); m++) {
+      boxes[n][m][0] /= ratio_w;
+      boxes[n][m][1] /= ratio_h;
+
+      boxes[n][m][0] = int(_min(_max(boxes[n][m][0], 0), oriimg_w - 1));
+      boxes[n][m][1] = int(_min(_max(boxes[n][m][1], 0), oriimg_h - 1));
+    }
+  }
+
+  for (int n = 0; n < boxes.size(); n++) {
+    int rect_width, rect_height;
+    rect_width = int(sqrt(pow(boxes[n][0][0] - boxes[n][1][0], 2) +
+                          pow(boxes[n][0][1] - boxes[n][1][1], 2)));
+    rect_height = int(sqrt(pow(boxes[n][0][0] - boxes[n][3][0], 2) +
+                           pow(boxes[n][0][1] - boxes[n][3][1], 2)));
+    if (rect_width <= 4 || rect_height <= 4)
+      continue;
+    root_points.push_back(boxes[n]);
+  }
+  return root_points;
+}
+
+} // namespace PaddleOCR

deploy/cpp_infer/src_system/preprocess_op.cpp

+// Copyright (c) 2020 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 "opencv2/core.hpp"
+#include "opencv2/imgcodecs.hpp"
+#include "opencv2/imgproc.hpp"
+#include "paddle_api.h"
+#include "paddle_inference_api.h"
+#include <chrono>
+#include <iomanip>
+#include <iostream>
+#include <ostream>
+#include <vector>
+
+#include <cstring>
+#include <fstream>
+#include <numeric>
+
+#include <include/preprocess_op.h>
+
+namespace PaddleOCR {
+
+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,
+                    const std::vector<float> &scale, const bool is_scale) {
+  double e = 1.0;
+  if (is_scale) {
+    e /= 255.0;
+  }
+  (*im).convertTo(*im, CV_32FC3, e);
+  std::vector<cv::Mat> bgr_channels(3);
+  cv::split(*im, bgr_channels);
+  for (auto i = 0; i < bgr_channels.size(); i++) {
+    bgr_channels[i].convertTo(bgr_channels[i], CV_32FC1, 1.0 * scale[i],
+                              (0.0 - mean[i]) * scale[i]);
+  }
+  cv::merge(bgr_channels, *im);
+}
+
+void ResizeImgType0::Run(const cv::Mat &img, cv::Mat &resize_img,
+                         int max_size_len, float &ratio_h, float &ratio_w,
+                         bool use_tensorrt) {
+  int w = img.cols;
+  int h = img.rows;
+
+  float ratio = 1.f;
+  int max_wh = w >= h ? w : h;
+  if (max_wh > max_size_len) {
+    if (h > w) {
+      ratio = float(max_size_len) / float(h);
+    } else {
+      ratio = float(max_size_len) / float(w);
+    }
+  }
+
+  int resize_h = int(float(h) * ratio);
+  int resize_w = int(float(w) * ratio);
+
+  resize_h = max(int(round(float(resize_h) / 32) * 32), 32);
+  resize_w = max(int(round(float(resize_w) / 32) * 32), 32);
+
+  cv::resize(img, resize_img, cv::Size(resize_w, resize_h));
+  ratio_h = float(resize_h) / float(h);
+  ratio_w = float(resize_w) / float(w);
+}
+
+void CrnnResizeImg::Run(const cv::Mat &img, cv::Mat &resize_img, float wh_ratio,
+                        bool use_tensorrt,
+                        const std::vector<int> &rec_image_shape) {
+  int imgC, imgH, imgW;
+  imgC = rec_image_shape[0];
+  imgH = rec_image_shape[1];
+  imgW = rec_image_shape[2];
+
+  imgW = int(32 * wh_ratio);
+
+  float ratio = float(img.cols) / float(img.rows);
+  int resize_w, resize_h;
+  if (ceilf(imgH * ratio) > imgW)
+    resize_w = imgW;
+  else
+    resize_w = int(ceilf(imgH * ratio));
+
+  cv::resize(img, resize_img, cv::Size(resize_w, imgH), 0.f, 0.f,
+             cv::INTER_LINEAR);
+  cv::copyMakeBorder(resize_img, resize_img, 0, 0, 0,
+                     int(imgW - resize_img.cols), cv::BORDER_CONSTANT,
+                     {127, 127, 127});
+}
+
+void ClsResizeImg::Run(const cv::Mat &img, cv::Mat &resize_img,
+                       bool use_tensorrt,
+                       const std::vector<int> &rec_image_shape) {
+  int imgC, imgH, imgW;
+  imgC = rec_image_shape[0];
+  imgH = rec_image_shape[1];
+  imgW = rec_image_shape[2];
+
+  float ratio = float(img.cols) / float(img.rows);
+  int resize_w, resize_h;
+  if (ceilf(imgH * ratio) > imgW)
+    resize_w = imgW;
+  else
+    resize_w = int(ceilf(imgH * ratio));
+
+  cv::resize(img, resize_img, cv::Size(resize_w, imgH), 0.f, 0.f,
+             cv::INTER_LINEAR);
+  if (resize_w < imgW) {
+    cv::copyMakeBorder(resize_img, resize_img, 0, 0, 0, imgW - resize_w,
+                       cv::BORDER_CONSTANT, cv::Scalar(0, 0, 0));
+  }
+}
+
+} // namespace PaddleOCR

deploy/cpp_infer/src_system/utility.cpp

+// Copyright (c) 2020 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 <dirent.h>
+#include <include/utility.h>
+#include <iostream>
+#include <ostream>
+#include <sys/stat.h>
+#include <sys/types.h>
+#include <vector>
+
+namespace PaddleOCR {
+
+std::vector<std::string> Utility::ReadDict(const std::string &path) {
+  std::ifstream in(path);
+  std::string line;
+  std::vector<std::string> m_vec;
+  if (in) {
+    while (getline(in, line)) {
+      m_vec.push_back(line);
+    }
+  } else {
+    std::cout << "no such label file: " << path << ", exit the program..."
+              << std::endl;
+    exit(1);
+  }
+  return m_vec;
+}
+
+void Utility::VisualizeBboxes(
+    const cv::Mat &srcimg,
+    const std::vector<std::vector<std::vector<int>>> &boxes) {
+  cv::Mat img_vis;
+  srcimg.copyTo(img_vis);
+  for (int n = 0; n < boxes.size(); n++) {
+    cv::Point rook_points[4];
+    for (int m = 0; m < boxes[n].size(); m++) {
+      rook_points[m] = cv::Point(int(boxes[n][m][0]), int(boxes[n][m][1]));
+    }
+
+    const cv::Point *ppt[1] = {rook_points};
+    int npt[] = {4};
+    cv::polylines(img_vis, ppt, npt, 1, 1, CV_RGB(0, 255, 0), 2, 8, 0);
+  }
+
+  cv::imwrite("./ocr_vis.png", img_vis);
+  std::cout << "The detection visualized image saved in ./ocr_vis.png"
+            << std::endl;
+}
+
+// list all files under a directory
+void Utility::GetAllFiles(const char *dir_name,
+                          std::vector<std::string> &all_inputs) {
+  if (NULL == dir_name) {
+    std::cout << " dir_name is null ! " << std::endl;
+    return;
+  }
+  struct stat s;
+  lstat(dir_name, &s);
+  if (!S_ISDIR(s.st_mode)) {
+    std::cout << "dir_name is not a valid directory !" << std::endl;
+    all_inputs.push_back(dir_name);
+    return;
+  } else {
+    struct dirent *filename; // return value for readdir()
+    DIR *dir;                // return value for opendir()
+    dir = opendir(dir_name);
+    if (NULL == dir) {
+      std::cout << "Can not open dir " << dir_name << std::endl;
+      return;
+    }
+    std::cout << "Successfully opened the dir !" << std::endl;
+    while ((filename = readdir(dir)) != NULL) {
+      if (strcmp(filename->d_name, ".") == 0 ||
+          strcmp(filename->d_name, "..") == 0)
+        continue;
+      // img_dir + std::string("/") + all_inputs[0];
+      all_inputs.push_back(dir_name + std::string("/") +
+                           std::string(filename->d_name));
+    }
+  }
+}
+
+} // namespace PaddleOCR
\ No newline at end of file

deploy/cpp_infer/tools/build.sh

-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
+set -o errexit
+
+if [ $# != 1 ] ; then
+echo "USAGE: $0 MODE (one of ['det', 'rec', 'system'])"
+echo " e.g.: $0 system"
+exit 1;
+fi
+
+# MODE be one of ['det', 'rec', 'system']
+MODE=$1
+cp CMakeLists_$MODE.txt CMakeLists.txt
+
+
+OPENCV_DIR=/paddle/git/new/PaddleOCR/deploy/cpp_infer/opencv-3.4.7/opencv3/
+LIB_DIR=/paddle/git/new/PaddleOCR/deploy/cpp_infer/paddle_inference/
+CUDA_LIB_DIR=/usr/local/cuda/lib64/
+CUDNN_LIB_DIR=/usr/lib/x86_64-linux-gnu/
+
 
 BUILD_DIR=build
 rm -rf ${BUILD_DIR}