Merge pull request #4538 from MissPenguin/dygraph

add batch infer for cpp rec

deploy/cpp_infer/include/ocr_rec.h

@@ -44,7 +44,8 @@ public:
                           const int &gpu_id, const int &gpu_mem,
                           const int &cpu_math_library_num_threads,
                           const bool &use_mkldnn, const string &label_path,
-                          const bool &use_tensorrt, const std::string &precision) {
+                          const bool &use_tensorrt, const std::string &precision,
+                          const int &rec_batch_num) {
     this->use_gpu_ = use_gpu;
     this->gpu_id_ = gpu_id;
     this->gpu_mem_ = gpu_mem;
@@ -52,6 +53,7 @@ public:
     this->use_mkldnn_ = use_mkldnn;
     this->use_tensorrt_ = use_tensorrt;
     this->precision_ = precision;
+    this->rec_batch_num_ = rec_batch_num;
 
     this->label_list_ = Utility::ReadDict(label_path);
     this->label_list_.insert(this->label_list_.begin(),
@@ -64,7 +66,7 @@ public:
   // Load Paddle inference model
   void LoadModel(const std::string &model_dir);
 
-  void Run(cv::Mat &img, std::vector<double> *times);
+  void Run(std::vector<cv::Mat> img_list, std::vector<double> *times);
 
 private:
   std::shared_ptr<Predictor> predictor_;
@@ -82,10 +84,12 @@ private:
   bool is_scale_ = true;
   bool use_tensorrt_ = false;
   std::string precision_ = "fp32";
+  int rec_batch_num_ = 6;
+    
   // pre-process
   CrnnResizeImg resize_op_;
   Normalize normalize_op_;
-  Permute permute_op_;
+  PermuteBatch permute_op_;
 
   // post-process
   PostProcessor post_processor_;

deploy/cpp_infer/include/preprocess_op.h

@@ -44,6 +44,11 @@ public:
   virtual void Run(const cv::Mat *im, float *data);
 };
 
+class PermuteBatch {
+public:
+  virtual void Run(const std::vector<cv::Mat> imgs, float *data);
+};
+    
 class ResizeImgType0 {
 public:
   virtual void Run(const cv::Mat &img, cv::Mat &resize_img, int max_size_len,

deploy/cpp_infer/include/utility.h

@@ -50,6 +50,9 @@ public:
     
   static cv::Mat GetRotateCropImage(const cv::Mat &srcimage,
                           std::vector<std::vector<int>> box);
+    
+  static std::vector<int> argsort(const std::vector<float>& array);
+
 };
 
 } // namespace PaddleOCR
\ No newline at end of file

deploy/cpp_infer/src/main.cpp

@@ -61,7 +61,7 @@ DEFINE_string(cls_model_dir, "", "Path of cls inference model.");
 DEFINE_double(cls_thresh, 0.9, "Threshold of cls_thresh.");
 // recognition related
 DEFINE_string(rec_model_dir, "", "Path of rec inference model.");
-DEFINE_int32(rec_batch_num, 1, "rec_batch_num.");
+DEFINE_int32(rec_batch_num, 6, "rec_batch_num.");
 DEFINE_string(char_list_file, "../../ppocr/utils/ppocr_keys_v1.txt", "Path of dictionary.");
 
 
@@ -146,8 +146,9 @@ int main_rec(std::vector<cv::String> cv_all_img_names) {
     CRNNRecognizer rec(FLAGS_rec_model_dir, FLAGS_use_gpu, FLAGS_gpu_id,
                        FLAGS_gpu_mem, FLAGS_cpu_threads,
                        FLAGS_enable_mkldnn, char_list_file,
-                       FLAGS_use_tensorrt, FLAGS_precision);
+                       FLAGS_use_tensorrt, FLAGS_precision, FLAGS_rec_batch_num);
 
+    std::vector<cv::Mat> img_list;
     for (int i = 0; i < cv_all_img_names.size(); ++i) {
       LOG(INFO) << "The predict img: " << cv_all_img_names[i];
 
@@ -156,22 +157,21 @@ int main_rec(std::vector<cv::String> cv_all_img_names) {
         std::cerr << "[ERROR] image read failed! image path: " << cv_all_img_names[i] << endl;
         exit(1);
       }
-
-      std::vector<double> rec_times;
-      rec.Run(srcimg, &rec_times);
-        
-      time_info[0] += rec_times[0];
-      time_info[1] += rec_times[1];
-      time_info[2] += rec_times[2];
+      img_list.push_back(srcimg);
     }
-        
+    std::vector<double> rec_times;
+    rec.Run(img_list, &rec_times);
+    time_info[0] += rec_times[0];
+    time_info[1] += rec_times[1];
+    time_info[2] += rec_times[2];
+    
     if (FLAGS_benchmark) {
         AutoLogger autolog("ocr_rec", 
                            FLAGS_use_gpu,
                            FLAGS_use_tensorrt,
                            FLAGS_enable_mkldnn,
                            FLAGS_cpu_threads,
-                           1, 
+                           FLAGS_rec_batch_num, 
                            "dynamic", 
                            FLAGS_precision, 
                            time_info, 
@@ -209,7 +209,7 @@ int main_system(std::vector<cv::String> cv_all_img_names) {
     CRNNRecognizer rec(FLAGS_rec_model_dir, FLAGS_use_gpu, FLAGS_gpu_id,
                        FLAGS_gpu_mem, FLAGS_cpu_threads,
                        FLAGS_enable_mkldnn, char_list_file,
-                       FLAGS_use_tensorrt, FLAGS_precision);
+                       FLAGS_use_tensorrt, FLAGS_precision, FLAGS_rec_batch_num);
 
     for (int i = 0; i < cv_all_img_names.size(); ++i) {
       LOG(INFO) << "The predict img: " << cv_all_img_names[i];
@@ -228,19 +228,22 @@ int main_system(std::vector<cv::String> cv_all_img_names) {
       time_info_det[1] += det_times[1];
       time_info_det[2] += det_times[2];
         
-      cv::Mat crop_img;
+      std::vector<cv::Mat> img_list;
       for (int j = 0; j < boxes.size(); j++) {
-        crop_img = Utility::GetRotateCropImage(srcimg, boxes[j]);
-
-        if (cls != nullptr) {
-          crop_img = cls->Run(crop_img);
-        }
-        rec.Run(crop_img, &rec_times);
-        time_info_rec[0] += rec_times[0];
-        time_info_rec[1] += rec_times[1];
-        time_info_rec[2] += rec_times[2];
+          cv::Mat crop_img;
+          crop_img = Utility::GetRotateCropImage(srcimg, boxes[j]);
+          if (cls != nullptr) {
+              crop_img = cls->Run(crop_img);
+          }
+          img_list.push_back(crop_img);
       }
+
+      rec.Run(img_list, &rec_times);
+      time_info_rec[0] += rec_times[0];
+      time_info_rec[1] += rec_times[1];
+      time_info_rec[2] += rec_times[2];
     }
+    
     if (FLAGS_benchmark) {
         AutoLogger autolog_det("ocr_det", 
                             FLAGS_use_gpu,
@@ -257,7 +260,7 @@ int main_system(std::vector<cv::String> cv_all_img_names) {
                             FLAGS_use_tensorrt,
                             FLAGS_enable_mkldnn,
                             FLAGS_cpu_threads,
-                            1, 
+                            FLAGS_rec_batch_num, 
                             "dynamic", 
                             FLAGS_precision, 
                             time_info_rec, 

deploy/cpp_infer/src/ocr_rec.cpp

@@ -15,83 +15,108 @@
 #include <include/ocr_rec.h>
 
 namespace PaddleOCR {
-
-void CRNNRecognizer::Run(cv::Mat &img, std::vector<double> *times) {
-  cv::Mat srcimg;
-  img.copyTo(srcimg);
-  cv::Mat resize_img;
-
-  float wh_ratio = float(srcimg.cols) / float(srcimg.rows);
-  auto preprocess_start = std::chrono::steady_clock::now();
-  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());
-  auto preprocess_end = std::chrono::steady_clock::now();
-
-  // 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});
-  auto inference_start = std::chrono::steady_clock::now();
-  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,
+    
+void CRNNRecognizer::Run(std::vector<cv::Mat> img_list, std::vector<double> *times) {
+    std::chrono::duration<float> preprocess_diff = std::chrono::steady_clock::now() - std::chrono::steady_clock::now();
+    std::chrono::duration<float> inference_diff = std::chrono::steady_clock::now() - std::chrono::steady_clock::now();
+    std::chrono::duration<float> postprocess_diff = std::chrono::steady_clock::now() - std::chrono::steady_clock::now();
+
+    int img_num = img_list.size();
+    std::vector<float> width_list;
+    for (int i = 0; i < img_num; i++) {
+        width_list.push_back(float(img_list[i].cols) / img_list[i].rows);
+    }
+    std::vector<int> indices = Utility::argsort(width_list);
+
+    for (int beg_img_no = 0; beg_img_no < img_num; beg_img_no += this->rec_batch_num_) {
+        auto preprocess_start = std::chrono::steady_clock::now();
+        int end_img_no = min(img_num, beg_img_no + this->rec_batch_num_);
+        float max_wh_ratio = 0;
+        for (int ino = beg_img_no; ino < end_img_no; ino ++) {
+            int h = img_list[indices[ino]].rows;
+            int w = img_list[indices[ino]].cols;
+            float wh_ratio = w * 1.0 / h;
+            max_wh_ratio = max(max_wh_ratio, wh_ratio);
+        }
+        std::vector<cv::Mat> norm_img_batch;
+        for (int ino = beg_img_no; ino < end_img_no; ino ++) {
+            cv::Mat srcimg;
+            img_list[indices[ino]].copyTo(srcimg);
+            cv::Mat resize_img;
+            this->resize_op_.Run(srcimg, resize_img, max_wh_ratio, this->use_tensorrt_);
+            this->normalize_op_.Run(&resize_img, this->mean_, this->scale_, this->is_scale_);
+            norm_img_batch.push_back(resize_img);
+        }
+        
+        int batch_width = int(ceilf(32 * max_wh_ratio)) - 1;
+        std::vector<float> input(this->rec_batch_num_ * 3 * 32 * batch_width, 0.0f);
+        this->permute_op_.Run(norm_img_batch, input.data());
+        auto preprocess_end = std::chrono::steady_clock::now();
+        preprocess_diff += preprocess_end - preprocess_start;
+
+        // Inference.
+        auto input_names = this->predictor_->GetInputNames();
+        auto input_t = this->predictor_->GetInputHandle(input_names[0]);
+        input_t->Reshape({this->rec_batch_num_, 3, 32, batch_width});
+        auto inference_start = std::chrono::steady_clock::now();
+        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());
-  auto inference_end = std::chrono::steady_clock::now();
-
-  // ctc decode
-  auto postprocess_start = std::chrono::steady_clock::now();
-  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]);
+        predict_batch.resize(out_num);
+
+        output_t->CopyToCpu(predict_batch.data());
+        auto inference_end = std::chrono::steady_clock::now();
+        inference_diff += inference_end - inference_start;
+        
+        // ctc decode
+        auto postprocess_start = std::chrono::steady_clock::now();
+        for (int m = 0; m < predict_shape[0]; m++) {
+            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[(m * predict_shape[1] + n) * predict_shape[2]],
+                                        &predict_batch[(m * predict_shape[1] + n + 1) * predict_shape[2]]));
+                max_value =
+                    float(*std::max_element(&predict_batch[(m * predict_shape[1] + n) * predict_shape[2]],
+                                            &predict_batch[(m * predict_shape[1] + 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;
+            if (isnan(score))
+                continue;
+            for (int i = 0; i < str_res.size(); i++) {
+                std::cout << str_res[i];
+            }
+            std::cout << "\tscore: " << score << std::endl;
+        }
+        auto postprocess_end = std::chrono::steady_clock::now();
+        postprocess_diff += postprocess_end - postprocess_start;
     }
-    last_index = argmax_idx;
-  }
-  auto postprocess_end = std::chrono::steady_clock::now();
-  score /= count;
-  for (int i = 0; i < str_res.size(); i++) {
-    std::cout << str_res[i];
-  }
-  std::cout << "\tscore: " << score << std::endl;
-
-  std::chrono::duration<float> preprocess_diff = preprocess_end - preprocess_start;
-  times->push_back(double(preprocess_diff.count() * 1000));
-  std::chrono::duration<float> inference_diff = inference_end - inference_start;
-  times->push_back(double(inference_diff.count() * 1000));
-  std::chrono::duration<float> postprocess_diff = postprocess_end - postprocess_start;
-  times->push_back(double(postprocess_diff.count() * 1000));
+    times->push_back(double(preprocess_diff.count() * 1000));
+    times->push_back(double(inference_diff.count() * 1000));
+    times->push_back(double(postprocess_diff.count() * 1000));
 }
 
+    
 void CRNNRecognizer::LoadModel(const std::string &model_dir) {
   //   AnalysisConfig config;
   paddle_infer::Config config;

deploy/cpp_infer/src/preprocess_op.cpp

@@ -40,6 +40,17 @@ void Permute::Run(const cv::Mat *im, float *data) {
   }
 }
 
+void PermuteBatch::Run(const std::vector<cv::Mat> imgs, float *data) {
+    for (int j = 0; j < imgs.size(); j ++){
+        int rh = imgs[j].rows;
+        int rw = imgs[j].cols;
+        int rc = imgs[j].channels();
+        for (int i = 0; i < rc; ++i) {
+            cv::extractChannel(imgs[j], cv::Mat(rh, rw, CV_32FC1, data + (j * rc + 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;
@@ -90,16 +101,17 @@ void CrnnResizeImg::Run(const cv::Mat &img, cv::Mat &resize_img, float wh_ratio,
   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,

deploy/cpp_infer/src/utility.cpp

@@ -147,4 +147,17 @@ cv::Mat Utility::GetRotateCropImage(const cv::Mat &srcimage,
   }
 }
 
+std::vector<int> Utility::argsort(const std::vector<float>& array)
+{
+    const int array_len(array.size());
+    std::vector<int> array_index(array_len, 0);
+    for (int i = 0; i < array_len; ++i)
+        array_index[i] = i;
+
+    std::sort(array_index.begin(), array_index.end(),
+        [&array](int pos1, int pos2) {return (array[pos1] < array[pos2]); });
+
+    return array_index;
+}
+
 } // namespace PaddleOCR
\ No newline at end of file

tools/infer/predict_rec.py

@@ -106,7 +106,6 @@ class TextRecognizer(object):
             return norm_img.astype(np.float32) / 128. - 1.
 
         assert imgC == img.shape[2]
-        max_wh_ratio = max(max_wh_ratio, imgW / imgH)
         imgW = int((32 * max_wh_ratio))
         h, w = img.shape[:2]
         ratio = w / float(h)