[Model] change ocr pre and post (#568)

* change ocr pre and post

* add pybind

* change ocr

* fix bug

* fix bug

* fix bug

* fix bug

* fix bug

* fix bug

* fix copy bug

* fix code style

* fix bug

* add new function

* fix windows ci bug

fastdeploy/utils/utils.h

@@ -22,6 +22,7 @@
 #include <sstream>
 #include <string>
 #include <vector>
+#include <numeric>
 
 #if defined(_WIN32)
 #ifdef FASTDEPLOY_LIB

fastdeploy/vision.h

@@ -48,6 +48,7 @@
 #include "fastdeploy/vision/matting/ppmatting/ppmatting.h"
 #include "fastdeploy/vision/ocr/ppocr/classifier.h"
 #include "fastdeploy/vision/ocr/ppocr/dbdetector.h"
+#include "fastdeploy/vision/ocr/ppocr/utils/ocr_utils.h"
 #include "fastdeploy/vision/ocr/ppocr/ppocr_v2.h"
 #include "fastdeploy/vision/ocr/ppocr/ppocr_v3.h"
 #include "fastdeploy/vision/ocr/ppocr/recognizer.h"

fastdeploy/vision/ocr/ppocr/classifier.cc

@@ -41,16 +41,7 @@ Classifier::Classifier(const std::string& model_file,
   initialized = Initialize();
 }
 
-// Init
 bool Classifier::Initialize() {
-  // pre&post process parameters
-  cls_thresh = 0.9;
-  cls_image_shape = {3, 48, 192};
-  cls_batch_num = 1;
-  mean = {0.5f, 0.5f, 0.5f};
-  scale = {0.5f, 0.5f, 0.5f};
-  is_scale = true;
-
   if (!InitRuntime()) {
     FDERROR << "Failed to initialize fastdeploy backend." << std::endl;
     return false;
@@ -59,85 +50,23 @@ bool Classifier::Initialize() {
   return true;
 }
 
-void OcrClassifierResizeImage(Mat* mat,
-                              const std::vector<int>& rec_image_shape) {
-  int imgC = rec_image_shape[0];
-  int imgH = rec_image_shape[1];
-  int imgW = rec_image_shape[2];
-
-  float ratio = float(mat->Width()) / float(mat->Height());
-
-  int resize_w;
-  if (ceilf(imgH * ratio) > imgW)
-    resize_w = imgW;
-  else
-    resize_w = int(ceilf(imgH * ratio));
-
-  Resize::Run(mat, resize_w, imgH);
-
-  std::vector<float> value = {0, 0, 0};
-  if (resize_w < imgW) {
-    Pad::Run(mat, 0, 0, 0, imgW - resize_w, value);
+bool Classifier::BatchPredict(const std::vector<cv::Mat>& images,
+                              std::vector<int32_t>* cls_labels, std::vector<float>* cls_scores) {
+  std::vector<FDMat> fd_images = WrapMat(images);
+  if (!preprocessor_.Run(&fd_images, &reused_input_tensors_)) {
+    FDERROR << "Failed to preprocess the input image." << std::endl;
+    return false;
   }
-}
-
-bool Classifier::Preprocess(Mat* mat, FDTensor* output) {
-  // 1. cls resizes
-  // 2. normalize
-  // 3. batch_permute
-  OcrClassifierResizeImage(mat, cls_image_shape);
-
-  Normalize::Run(mat, mean, scale, true);
-
-  HWC2CHW::Run(mat);
-  Cast::Run(mat, "float");
-
-  mat->ShareWithTensor(output);
-  output->shape.insert(output->shape.begin(), 1);
-
-  return true;
-}
-
-bool Classifier::Postprocess(FDTensor& infer_result,
-                             std::tuple<int, float>* cls_result) {
-  std::vector<int64_t> output_shape = infer_result.shape;
-  FDASSERT(output_shape[0] == 1, "Only support batch =1 now.");
-
-  float* out_data = static_cast<float*>(infer_result.Data());
-
-  int label = std::distance(
-      &out_data[0], std::max_element(&out_data[0], &out_data[output_shape[1]]));
-
-  float score =
-      float(*std::max_element(&out_data[0], &out_data[output_shape[1]]));
-
-  std::get<0>(*cls_result) = label;
-  std::get<1>(*cls_result) = score;
-
-  return true;
-}
-
-bool Classifier::Predict(cv::Mat* img, std::tuple<int, float>* cls_result) {
-  Mat mat(*img);
-  std::vector<FDTensor> input_tensors(1);
-
-  if (!Preprocess(&mat, &input_tensors[0])) {
-    FDERROR << "Failed to preprocess input image." << std::endl;
+  reused_input_tensors_[0].name = InputInfoOfRuntime(0).name;
+  if (!Infer(reused_input_tensors_, &reused_output_tensors_)) {
+    FDERROR << "Failed to inference by runtime." << std::endl;
     return false;
   }
 
-  input_tensors[0].name = InputInfoOfRuntime(0).name;
-  std::vector<FDTensor> output_tensors;
-  if (!Infer(input_tensors, &output_tensors)) {
-    FDERROR << "Failed to inference." << std::endl;
+  if (!postprocessor_.Run(reused_output_tensors_, cls_labels, cls_scores)) {
+    FDERROR << "Failed to postprocess the inference cls_results by runtime." << std::endl;
     return false;
   }
-
-  if (!Postprocess(output_tensors[0], cls_result)) {
-    FDERROR << "Failed to post process." << std::endl;
-    return false;
-  }
-
   return true;
 }
 

fastdeploy/vision/ocr/ppocr/classifier.h

@@ -17,6 +17,8 @@
 #include "fastdeploy/vision/common/processors/transform.h"
 #include "fastdeploy/vision/common/result.h"
 #include "fastdeploy/vision/ocr/ppocr/utils/ocr_postprocess_op.h"
+#include "fastdeploy/vision/ocr/ppocr/cls_postprocessor.h"
+#include "fastdeploy/vision/ocr/ppocr/cls_preprocessor.h"
 
 namespace fastdeploy {
 namespace vision {
@@ -41,29 +43,22 @@ class FASTDEPLOY_DECL Classifier : public FastDeployModel {
              const ModelFormat& model_format = ModelFormat::PADDLE);
   /// Get model's name
   std::string ModelName() const { return "ppocr/ocr_cls"; }
-  /** \brief Predict the input image and get OCR classification model result.
+
+  /** \brief BatchPredict the input image and get OCR classification model cls_result.
    *
-   * \param[in] im The input image data, comes from cv::imread(), is a 3-D array with layout HWC, BGR format.
-   * \param[in] result The output of OCR classification model result will be writen to this structure.
+   * \param[in] images The list of input image data, comes from cv::imread(), is a 3-D array with layout HWC, BGR format.
+   * \param[in] cls_results The output of OCR classification model cls_result will be writen to this structure.
    * \return true if the prediction is successed, otherwise false.
    */
-  virtual bool Predict(cv::Mat* img, std::tuple<int, float>* result);
+  virtual bool BatchPredict(const std::vector<cv::Mat>& images,
+                            std::vector<int32_t>* cls_labels,
+                            std::vector<float>* cls_scores);
 
-  // Pre & Post parameters
-  float cls_thresh;
-  std::vector<int> cls_image_shape;
-  int cls_batch_num;
-
-  std::vector<float> mean;
-  std::vector<float> scale;
-  bool is_scale;
+  ClassifierPreprocessor preprocessor_;
+  ClassifierPostprocessor postprocessor_;
 
  private:
   bool Initialize();
-  /// Preprocess the input data, and set the preprocessed results to `outputs`
-  bool Preprocess(Mat* img, FDTensor* output);
-  /// Postprocess the inferenced results, and set the final result to `result`
-  bool Postprocess(FDTensor& infer_result, std::tuple<int, float>* result);
 };
 
 }  // namespace ocr

fastdeploy/vision/ocr/ppocr/cls_postprocessor.cc

@@ -0,0 +1,65 @@
+// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "fastdeploy/vision/ocr/ppocr/cls_postprocessor.h"
+#include "fastdeploy/utils/perf.h"
+#include "fastdeploy/vision/ocr/ppocr/utils/ocr_utils.h"
+
+namespace fastdeploy {
+namespace vision {
+namespace ocr {
+
+ClassifierPostprocessor::ClassifierPostprocessor() {
+  initialized_ = true;
+}
+
+bool SingleBatchPostprocessor(const float* out_data, const size_t& length, int* cls_label, float* cls_score) {
+
+  *cls_label = std::distance(
+      &out_data[0], std::max_element(&out_data[0], &out_data[length]));
+
+  *cls_score =
+      float(*std::max_element(&out_data[0], &out_data[length]));
+  return true;
+}
+
+bool ClassifierPostprocessor::Run(const std::vector<FDTensor>& tensors,
+                                  std::vector<int32_t>* cls_labels,
+                                  std::vector<float>* cls_scores) {
+  if (!initialized_) {
+    FDERROR << "Postprocessor is not initialized." << std::endl;
+    return false;
+  }
+  // Classifier have only 1 output tensor.
+  const FDTensor& tensor = tensors[0];
+
+  // For Classifier, the output tensor shape = [batch,2]
+  size_t batch = tensor.shape[0];
+  size_t length = accumulate(tensor.shape.begin()+1, tensor.shape.end(), 1, std::multiplies<int>());
+
+  cls_labels->resize(batch);
+  cls_scores->resize(batch);
+  const float* tensor_data = reinterpret_cast<const float*>(tensor.Data());
+ 
+  for (int i_batch = 0; i_batch < batch; ++i_batch) {
+    if(!SingleBatchPostprocessor(tensor_data, length, &cls_labels->at(i_batch),&cls_scores->at(i_batch))) return false;
+    tensor_data = tensor_data + length;
+  }
+
+  return true;
+}
+
+}  // namespace classification
+}  // namespace vision
+}  // namespace fastdeploy

fastdeploy/vision/ocr/ppocr/cls_postprocessor.h

@@ -0,0 +1,51 @@
+// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+#include "fastdeploy/vision/common/processors/transform.h"
+#include "fastdeploy/vision/common/result.h"
+#include "fastdeploy/vision/ocr/ppocr/utils/ocr_postprocess_op.h"
+
+namespace fastdeploy {
+namespace vision {
+
+namespace ocr {
+/*! @brief Postprocessor object for Classifier serials model.
+ */
+class FASTDEPLOY_DECL ClassifierPostprocessor {
+ public:
+  /** \brief Create a postprocessor instance for Classifier serials model
+   *
+   */
+  ClassifierPostprocessor();
+
+  /** \brief Process the result of runtime and fill to ClassifyResult structure
+   *
+   * \param[in] tensors The inference result from runtime
+   * \param[in] cls_labels The output result of classification
+   * \param[in] cls_scores The output result of classification
+   * \return true if the postprocess successed, otherwise false
+   */
+  bool Run(const std::vector<FDTensor>& tensors,
+          std::vector<int32_t>* cls_labels, std::vector<float>* cls_scores);
+
+  float cls_thresh_ = 0.9;
+
+ private:
+  bool initialized_ = false;
+};
+
+}  // namespace ocr
+}  // namespace vision
+}  // namespace fastdeploy

fastdeploy/vision/ocr/ppocr/cls_preprocessor.cc

@@ -0,0 +1,88 @@
+// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "fastdeploy/vision/ocr/ppocr/cls_preprocessor.h"
+#include "fastdeploy/utils/perf.h"
+#include "fastdeploy/vision/ocr/ppocr/utils/ocr_utils.h"
+#include "fastdeploy/function/concat.h"
+
+namespace fastdeploy {
+namespace vision {
+namespace ocr {
+
+ClassifierPreprocessor::ClassifierPreprocessor() {
+  initialized_ = true;
+}
+
+void OcrClassifierResizeImage(FDMat* mat,
+                              const std::vector<int>& cls_image_shape) {
+  int imgC = cls_image_shape[0];
+  int imgH = cls_image_shape[1];
+  int imgW = cls_image_shape[2];
+
+  float ratio = float(mat->Width()) / float(mat->Height());
+
+  int resize_w;
+  if (ceilf(imgH * ratio) > imgW)
+    resize_w = imgW;
+  else
+    resize_w = int(ceilf(imgH * ratio));
+
+  Resize::Run(mat, resize_w, imgH);
+
+  std::vector<float> value = {0, 0, 0};
+  if (resize_w < imgW) {
+    Pad::Run(mat, 0, 0, 0, imgW - resize_w, value);
+  }
+}
+
+bool ClassifierPreprocessor::Run(std::vector<FDMat>* images, std::vector<FDTensor>* outputs) {
+  if (!initialized_) {
+    FDERROR << "The preprocessor is not initialized." << std::endl;
+    return false;
+  }
+  if (images->size() == 0) {
+    FDERROR << "The size of input images should be greater than 0." << std::endl;
+    return false;
+  }
+
+  for (size_t i = 0; i < images->size(); ++i) {
+    FDMat* mat = &(images->at(i));
+    OcrClassifierResizeImage(mat, cls_image_shape_);
+    NormalizeAndPermute::Run(mat, mean_, scale_, is_scale_);
+    /*
+    Normalize::Run(mat, mean_, scale_, is_scale_);
+    HWC2CHW::Run(mat);
+    Cast::Run(mat, "float");
+    */
+  }
+  // Only have 1 output Tensor.
+  outputs->resize(1);
+  // Concat all the preprocessed data to a batch tensor
+  std::vector<FDTensor> tensors(images->size()); 
+  for (size_t i = 0; i < images->size(); ++i) {
+    (*images)[i].ShareWithTensor(&(tensors[i]));
+    tensors[i].ExpandDim(0);
+  }
+  if (tensors.size() == 1) {
+    (*outputs)[0] = std::move(tensors[0]);
+  } else {
+    function::Concat(tensors, &((*outputs)[0]), 0);
+  }
+  return true;
+}
+
+}  // namespace ocr
+}  // namespace vision
+}  // namespace fastdeploy

fastdeploy/vision/ocr/ppocr/cls_preprocessor.h

@@ -0,0 +1,51 @@
+// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+#include "fastdeploy/vision/common/processors/transform.h"
+#include "fastdeploy/vision/common/result.h"
+
+namespace fastdeploy {
+namespace vision {
+
+namespace ocr {
+/*! @brief Preprocessor object for Classifier serials model.
+ */
+class FASTDEPLOY_DECL ClassifierPreprocessor {
+ public:
+  /** \brief Create a preprocessor instance for Classifier serials model
+   *
+   */
+  ClassifierPreprocessor();
+
+  /** \brief Process the input image and prepare input tensors for runtime
+   *
+   * \param[in] images The input image data list, all the elements are returned by cv::imread()
+   * \param[in] outputs The output tensors which will feed in runtime
+   * \return true if the preprocess successed, otherwise false
+   */
+  bool Run(std::vector<FDMat>* images, std::vector<FDTensor>* outputs);
+
+  std::vector<float> mean_ = {0.5f, 0.5f, 0.5f};
+  std::vector<float> scale_ = {0.5f, 0.5f, 0.5f};
+  bool is_scale_ = true;
+  std::vector<int> cls_image_shape_ = {3, 48, 192};
+
+ private:
+  bool initialized_ = false;
+};
+
+}  // namespace ocr
+}  // namespace vision
+}  // namespace fastdeploy

fastdeploy/vision/ocr/ppocr/dbdetector.cc

@@ -43,158 +43,50 @@ DBDetector::DBDetector(const std::string& model_file,
 
 // Init
 bool DBDetector::Initialize() {
-  // pre&post process parameters
-  max_side_len = 960;
-
-  det_db_thresh = 0.3;
-  det_db_box_thresh = 0.6;
-  det_db_unclip_ratio = 1.5;
-  det_db_score_mode = "slow";
-  use_dilation = false;
-
-  mean = {0.485f, 0.456f, 0.406f};
-  scale = {0.229f, 0.224f, 0.225f};
-  is_scale = true;
-
   if (!InitRuntime()) {
     FDERROR << "Failed to initialize fastdeploy backend." << std::endl;
     return false;
   }
-
-  return true;
-}
-
-void OcrDetectorResizeImage(Mat* img, int max_size_len, float* ratio_h,
-                            float* ratio_w) {
-  int w = img->Width();
-  int h = img->Height();
-
-  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 = std::max(int(std::round(float(resize_h) / 32) * 32), 32);
-  resize_w = std::max(int(std::round(float(resize_w) / 32) * 32), 32);
-
-  Resize::Run(img, resize_w, resize_h);
-
-  *ratio_h = float(resize_h) / float(h);
-  *ratio_w = float(resize_w) / float(w);
-}
-
-bool DBDetector::Preprocess(
-    Mat* mat, FDTensor* output,
-    std::map<std::string, std::array<float, 2>>* im_info) {
-  // Resize
-  OcrDetectorResizeImage(mat, max_side_len, &ratio_h, &ratio_w);
-  // Normalize
-  Normalize::Run(mat, mean, scale, true);
-
-  (*im_info)["output_shape"] = {static_cast<float>(mat->Height()),
-                                static_cast<float>(mat->Width())};
-  //-CHW
-  HWC2CHW::Run(mat);
-  Cast::Run(mat, "float");
-
-  mat->ShareWithTensor(output);
-  output->shape.insert(output->shape.begin(), 1);
-  return true;
-}
-
-bool DBDetector::Postprocess(
-    FDTensor& infer_result, std::vector<std::array<int, 8>>* boxes_result,
-    const std::map<std::string, std::array<float, 2>>& im_info) {
-  std::vector<int64_t> output_shape = infer_result.shape;
-  FDASSERT(output_shape[0] == 1, "Only support batch =1 now.");
-  int n2 = output_shape[2];
-  int n3 = output_shape[3];
-  int n = n2 * n3;
-
-  float* out_data = static_cast<float*>(infer_result.Data());
-  // prepare bitmap
-  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 = det_db_thresh * 255;
-  const double maxvalue = 255;
-  cv::Mat bit_map;
-  cv::threshold(cbuf_map, bit_map, threshold, maxvalue, cv::THRESH_BINARY);
-  if (use_dilation) {
-    cv::Mat dila_ele =
-        cv::getStructuringElement(cv::MORPH_RECT, cv::Size(2, 2));
-    cv::dilate(bit_map, bit_map, dila_ele);
-  }
-
-  std::vector<std::vector<std::vector<int>>> boxes;
-
-  boxes =
-      post_processor_.BoxesFromBitmap(pred_map, bit_map, det_db_box_thresh,
-                                      det_db_unclip_ratio, det_db_score_mode);
-
-  boxes = post_processor_.FilterTagDetRes(boxes, ratio_h, ratio_w, im_info);
-
-  // boxes to boxes_result
-  for (int i = 0; i < boxes.size(); i++) {
-    std::array<int, 8> new_box;
-    int k = 0;
-    for (auto& vec : boxes[i]) {
-      for (auto& e : vec) {
-        new_box[k++] = e;
-      }
-    }
-    boxes_result->push_back(new_box);
-  }
-
   return true;
 }
 
 bool DBDetector::Predict(cv::Mat* img,
                          std::vector<std::array<int, 8>>* boxes_result) {
-  Mat mat(*img);
+  if (!Predict(*img, boxes_result)) {
+    return false;
+  }
+  return true;
+}
 
-  std::vector<FDTensor> input_tensors(1);
+bool DBDetector::Predict(const cv::Mat& img,
+                         std::vector<std::array<int, 8>>* boxes_result) {
+  std::vector<std::vector<std::array<int, 8>>> det_results;
+  if (!BatchPredict({img}, &det_results)) {
+    return false;
+  }
+  *boxes_result = std::move(det_results[0]);
+  return true;
+}
 
-  std::map<std::string, std::array<float, 2>> im_info;
-
-  // Record the shape of image and the shape of preprocessed image
-  im_info["input_shape"] = {static_cast<float>(mat.Height()),
-                            static_cast<float>(mat.Width())};
-  im_info["output_shape"] = {static_cast<float>(mat.Height()),
-                             static_cast<float>(mat.Width())};
-
-  if (!Preprocess(&mat, &input_tensors[0], &im_info)) {
+bool DBDetector::BatchPredict(const std::vector<cv::Mat>& images,
+                              std::vector<std::vector<std::array<int, 8>>>* det_results) {
+  std::vector<FDMat> fd_images = WrapMat(images);
+  std::vector<std::array<int, 4>> batch_det_img_info;
+  if (!preprocessor_.Run(&fd_images, &reused_input_tensors_, &batch_det_img_info)) {
     FDERROR << "Failed to preprocess input image." << std::endl;
     return false;
   }
 
-  input_tensors[0].name = InputInfoOfRuntime(0).name;
-  std::vector<FDTensor> output_tensors;
-  if (!Infer(input_tensors, &output_tensors)) {
-    FDERROR << "Failed to inference." << std::endl;
+  reused_input_tensors_[0].name = InputInfoOfRuntime(0).name;
+  if (!Infer(reused_input_tensors_, &reused_output_tensors_)) {
+    FDERROR << "Failed to inference by runtime." << std::endl;
     return false;
   }
 
-  if (!Postprocess(output_tensors[0], boxes_result, im_info)) {
-    FDERROR << "Failed to post process." << std::endl;
+  if (!postprocessor_.Run(reused_output_tensors_, det_results, batch_det_img_info)) {
+    FDERROR << "Failed to postprocess the inference cls_results by runtime." << std::endl;
     return false;
   }
-
   return true;
 }
 

fastdeploy/vision/ocr/ppocr/dbdetector.h

@@ -17,6 +17,8 @@
 #include "fastdeploy/vision/common/processors/transform.h"
 #include "fastdeploy/vision/common/result.h"
 #include "fastdeploy/vision/ocr/ppocr/utils/ocr_postprocess_op.h"
+#include "fastdeploy/vision/ocr/ppocr/det_postprocessor.h"
+#include "fastdeploy/vision/ocr/ppocr/det_preprocessor.h"
 
 namespace fastdeploy {
 namespace vision {
@@ -44,40 +46,34 @@ class FASTDEPLOY_DECL DBDetector : public FastDeployModel {
   std::string ModelName() const { return "ppocr/ocr_det"; }
   /** \brief Predict the input image and get OCR detection model result.
    *
-   * \param[in] im The input image data, comes from cv::imread(), is a 3-D array with layout HWC, BGR format.
+   * \param[in] img The input image data, comes from cv::imread(), is a 3-D array with layout HWC, BGR format.
    * \param[in] boxes_result The output of OCR detection model result will be writen to this structure.
    * \return true if the prediction is successed, otherwise false.
    */
-  virtual bool Predict(cv::Mat* im,
+  virtual bool Predict(cv::Mat* img,
                        std::vector<std::array<int, 8>>* boxes_result);
+   /** \brief Predict the input image and get OCR detection model result.
+   *
+   * \param[in] img The input image data, comes from cv::imread(), is a 3-D array with layout HWC, BGR format.
+   * \param[in] boxes_result The output of OCR detection model result will be writen to this structure.
+   * \return true if the prediction is successed, otherwise false.
+   */
+  virtual bool Predict(const cv::Mat& img,
+                       std::vector<std::array<int, 8>>* boxes_result);
+   /** \brief BatchPredict the input image and get OCR detection model result.
+   *
+   * \param[in] images The list input of image data, comes from cv::imread(), is a 3-D array with layout HWC, BGR format.
+   * \param[in] det_results The output of OCR detection model result will be writen to this structure.
+   * \return true if the prediction is successed, otherwise false.
+   */
+  virtual bool BatchPredict(const std::vector<cv::Mat>& images,
+          std::vector<std::vector<std::array<int, 8>>>* det_results);
 
-  // Pre & Post process parameters
-  int max_side_len;
-
-  float ratio_h{};
-  float ratio_w{};
-
-  double det_db_thresh;
-  double det_db_box_thresh;
-  double det_db_unclip_ratio;
-  std::string det_db_score_mode;
-  bool use_dilation;
-
-  std::vector<float> mean;
-  std::vector<float> scale;
-  bool is_scale;
+  DBDetectorPreprocessor preprocessor_;
+  DBDetectorPostprocessor postprocessor_;
 
  private:
   bool Initialize();
-  /// Preprocess the input data, and set the preprocessed results to `outputs`
-  bool Preprocess(Mat* mat, FDTensor* outputs,
-                  std::map<std::string, std::array<float, 2>>* im_info);
-  /*! @brief Postprocess the inferenced results, and set the final result to `boxes_result`
-   */
-  bool Postprocess(FDTensor& infer_result,
-                   std::vector<std::array<int, 8>>* boxes_result,
-                   const std::map<std::string, std::array<float, 2>>& im_info);
-  PostProcessor post_processor_;
 };
 
 }  // namespace ocr

fastdeploy/vision/ocr/ppocr/det_postprocessor.cc

@@ -0,0 +1,110 @@
+// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "fastdeploy/vision/ocr/ppocr/det_postprocessor.h"
+#include "fastdeploy/utils/perf.h"
+#include "fastdeploy/vision/ocr/ppocr/utils/ocr_utils.h"
+
+namespace fastdeploy {
+namespace vision {
+namespace ocr {
+
+DBDetectorPostprocessor::DBDetectorPostprocessor() {
+  initialized_ = true;
+}
+
+bool DBDetectorPostprocessor::SingleBatchPostprocessor(
+      const float* out_data,
+      int n2,
+      int n3,
+      const std::array<int,4>& det_img_info,
+      std::vector<std::array<int, 8>>* boxes_result
+    ) {
+  int n = n2 * n3;
+  
+  // prepare bitmap
+  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 = det_db_thresh_ * 255;
+  const double maxvalue = 255;
+  cv::Mat bit_map;
+  cv::threshold(cbuf_map, bit_map, threshold, maxvalue, cv::THRESH_BINARY);
+  if (use_dilation_) {
+    cv::Mat dila_ele =
+        cv::getStructuringElement(cv::MORPH_RECT, cv::Size(2, 2));
+    cv::dilate(bit_map, bit_map, dila_ele);
+  }
+
+  std::vector<std::vector<std::vector<int>>> boxes;
+
+  boxes =
+      post_processor_.BoxesFromBitmap(pred_map, bit_map, det_db_box_thresh_,
+                                      det_db_unclip_ratio_, det_db_score_mode_);
+
+  boxes = post_processor_.FilterTagDetRes(boxes, det_img_info);
+
+  // boxes to boxes_result
+  for (int i = 0; i < boxes.size(); i++) {
+    std::array<int, 8> new_box;
+    int k = 0;
+    for (auto& vec : boxes[i]) {
+      for (auto& e : vec) {
+        new_box[k++] = e;
+      }
+    }
+    boxes_result->push_back(new_box);
+  }
+
+  return true;
+}
+
+bool DBDetectorPostprocessor::Run(const std::vector<FDTensor>& tensors,
+                                  std::vector<std::vector<std::array<int, 8>>>* results,
+                                  const std::vector<std::array<int,4>>& batch_det_img_info) {
+  if (!initialized_) {
+    FDERROR << "Postprocessor is not initialized." << std::endl;
+    return false;
+  }
+  // DBDetector have only 1 output tensor.
+  const FDTensor& tensor = tensors[0];
+
+  // For DBDetector, the output tensor shape = [batch, 1, ?, ?]
+  size_t batch = tensor.shape[0];
+  size_t length = accumulate(tensor.shape.begin()+1, tensor.shape.end(), 1, std::multiplies<int>());
+  const float* tensor_data = reinterpret_cast<const float*>(tensor.Data());
+ 
+  results->resize(batch);
+  for (int i_batch = 0; i_batch < batch; ++i_batch) {
+    if(!SingleBatchPostprocessor(tensor_data,
+                                 tensor.shape[2],
+                                 tensor.shape[3],
+                                 batch_det_img_info[i_batch],
+                                 &results->at(i_batch)
+                                ))return false;
+    tensor_data = tensor_data + length;
+  }
+  return true;
+}
+
+}  // namespace ocr
+}  // namespace vision
+}  // namespace fastdeploy

fastdeploy/vision/ocr/ppocr/det_postprocessor.h

@@ -0,0 +1,62 @@
+// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+#include "fastdeploy/vision/common/processors/transform.h"
+#include "fastdeploy/vision/common/result.h"
+#include "fastdeploy/vision/ocr/ppocr/utils/ocr_postprocess_op.h"
+
+namespace fastdeploy {
+namespace vision {
+
+namespace ocr {
+/*! @brief Postprocessor object for DBDetector serials model.
+ */
+class FASTDEPLOY_DECL DBDetectorPostprocessor {
+ public:
+  /** \brief Create a postprocessor instance for DBDetector serials model
+   *
+   */
+  DBDetectorPostprocessor();
+
+  /** \brief Process the result of runtime and fill to results structure
+   *
+   * \param[in] tensors The inference result from runtime
+   * \param[in] results The output result of detector
+   * \param[in] batch_det_img_info The detector_preprocess result
+   * \return true if the postprocess successed, otherwise false
+   */
+  bool Run(const std::vector<FDTensor>& tensors,
+           std::vector<std::vector<std::array<int, 8>>>* results,
+           const std::vector<std::array<int, 4>>& batch_det_img_info);
+
+  double det_db_thresh_ = 0.3;
+  double det_db_box_thresh_ = 0.6;
+  double det_db_unclip_ratio_ = 1.5;
+  std::string det_db_score_mode_ = "slow";
+  bool use_dilation_ = false;
+
+ private:
+  bool initialized_ = false;
+  PostProcessor post_processor_;
+  bool SingleBatchPostprocessor(const float* out_data,
+                                int n2,
+                                int n3,
+                                const std::array<int, 4>& det_img_info,
+                                std::vector<std::array<int, 8>>* boxes_result);
+};
+
+}  // namespace ocr
+}  // namespace vision
+}  // namespace fastdeploy

fastdeploy/vision/ocr/ppocr/det_preprocessor.cc

@@ -0,0 +1,113 @@
+// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "fastdeploy/vision/ocr/ppocr/det_preprocessor.h"
+#include "fastdeploy/utils/perf.h"
+#include "fastdeploy/vision/ocr/ppocr/utils/ocr_utils.h"
+#include "fastdeploy/function/concat.h"
+
+namespace fastdeploy {
+namespace vision {
+namespace ocr {
+
+DBDetectorPreprocessor::DBDetectorPreprocessor() {
+  initialized_ = true;
+}
+
+std::array<int, 4> OcrDetectorGetInfo(FDMat* img, int max_size_len) {
+  int w = img->Width();
+  int h = img->Height();
+
+  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 = std::max(int(std::round(float(resize_h) / 32) * 32), 32);
+  resize_w = std::max(int(std::round(float(resize_w) / 32) * 32), 32);
+
+  return {w,h,resize_w,resize_h};
+  /*
+  *ratio_h = float(resize_h) / float(h);
+  *ratio_w = float(resize_w) / float(w);
+  */
+}
+bool OcrDetectorResizeImage(FDMat* img,
+                            int resize_w,
+                            int resize_h,
+                            int max_resize_w,
+                            int max_resize_h) {
+  Resize::Run(img, resize_w, resize_h);
+  std::vector<float> value = {0, 0, 0};
+  Pad::Run(img, 0, max_resize_h-resize_h, 0, max_resize_w - resize_w, value);
+  return true;
+}
+
+bool DBDetectorPreprocessor::Run(std::vector<FDMat>* images,
+                                 std::vector<FDTensor>* outputs,
+                                 std::vector<std::array<int, 4>>* batch_det_img_info_ptr) {
+  if (!initialized_) {
+    FDERROR << "The preprocessor is not initialized." << std::endl;
+    return false;
+  }
+  if (images->size() == 0) {
+    FDERROR << "The size of input images should be greater than 0." << std::endl;
+    return false;
+  }
+  int max_resize_w = 0;
+  int max_resize_h = 0;
+  std::vector<std::array<int, 4>>& batch_det_img_info = *batch_det_img_info_ptr;
+  batch_det_img_info.clear();
+  batch_det_img_info.resize(images->size());
+  for (size_t i = 0; i < images->size(); ++i) {
+    FDMat* mat = &(images->at(i));
+    batch_det_img_info[i] = OcrDetectorGetInfo(mat,max_side_len_);
+    max_resize_w = std::max(max_resize_w,batch_det_img_info[i][2]);
+    max_resize_h = std::max(max_resize_h,batch_det_img_info[i][3]);
+  }
+  for (size_t i = 0; i < images->size(); ++i) {
+    FDMat* mat = &(images->at(i));
+    OcrDetectorResizeImage(mat, batch_det_img_info[i][2],batch_det_img_info[i][3],max_resize_w,max_resize_h);
+    NormalizeAndPermute::Run(mat, mean_, scale_, is_scale_);
+    /*
+    Normalize::Run(mat, mean_, scale_, is_scale_);
+    HWC2CHW::Run(mat);
+    Cast::Run(mat, "float");
+    */
+  }
+  // Only have 1 output Tensor.
+  outputs->resize(1);
+  // Concat all the preprocessed data to a batch tensor
+  std::vector<FDTensor> tensors(images->size()); 
+  for (size_t i = 0; i < images->size(); ++i) {
+    (*images)[i].ShareWithTensor(&(tensors[i]));
+    tensors[i].ExpandDim(0);
+  }
+  if (tensors.size() == 1) {
+    (*outputs)[0] = std::move(tensors[0]);
+  } else {
+    function::Concat(tensors, &((*outputs)[0]), 0);
+  }
+  return true;
+}
+
+}  // namespace ocr
+}  // namespace vision
+}  // namespace fastdeploy

fastdeploy/vision/ocr/ppocr/det_preprocessor.h

@@ -0,0 +1,54 @@
+// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+#include "fastdeploy/vision/common/processors/transform.h"
+#include "fastdeploy/vision/common/result.h"
+
+namespace fastdeploy {
+namespace vision {
+
+namespace ocr {
+/*! @brief Preprocessor object for DBDetector serials model.
+ */
+class FASTDEPLOY_DECL DBDetectorPreprocessor {
+ public:
+  /** \brief Create a preprocessor instance for DBDetector serials model
+   *
+   */
+  DBDetectorPreprocessor();
+
+  /** \brief Process the input image and prepare input tensors for runtime
+   *
+   * \param[in] images The input image data list, all the elements are returned by cv::imread()
+   * \param[in] outputs The output tensors which will feed in runtime
+   * \param[in] batch_det_img_info_ptr The output of preprocess
+   * \return true if the preprocess successed, otherwise false
+   */
+  bool Run(std::vector<FDMat>* images,
+           std::vector<FDTensor>* outputs,
+           std::vector<std::array<int, 4>>* batch_det_img_info_ptr);
+
+  int max_side_len_ = 960;
+  std::vector<float> mean_ = {0.485f, 0.456f, 0.406f};
+  std::vector<float> scale_ = {0.229f, 0.224f, 0.225f};
+  bool is_scale_ = true;
+
+ private:
+  bool initialized_ = false;
+};
+
+}  // namespace ocr
+}  // namespace vision
+}  // namespace fastdeploy

fastdeploy/vision/ocr/ppocr/ocrmodel_pybind.cc

@@ -16,45 +16,171 @@
 
 namespace fastdeploy {
 void BindPPOCRModel(pybind11::module& m) {
+  m.def("sort_boxes", [](std::vector<std::array<int, 8>>& boxes) {
+       vision::ocr::SortBoxes(&boxes);
+       return boxes;
+  });
   // DBDetector
   pybind11::class_<vision::ocr::DBDetector, FastDeployModel>(m, "DBDetector")
       .def(pybind11::init<std::string, std::string, RuntimeOption,
                           ModelFormat>())
       .def(pybind11::init<>())
+      .def_readwrite("preprocessor", &vision::ocr::DBDetector::preprocessor_)
+      .def_readwrite("postprocessor", &vision::ocr::DBDetector::postprocessor_);
 
-      .def_readwrite("max_side_len", &vision::ocr::DBDetector::max_side_len)
-      .def_readwrite("det_db_thresh", &vision::ocr::DBDetector::det_db_thresh)
-      .def_readwrite("det_db_box_thresh",
-                     &vision::ocr::DBDetector::det_db_box_thresh)
-      .def_readwrite("det_db_unclip_ratio",
-                     &vision::ocr::DBDetector::det_db_unclip_ratio)
-      .def_readwrite("det_db_score_mode",
-                     &vision::ocr::DBDetector::det_db_score_mode)
-      .def_readwrite("use_dilation", &vision::ocr::DBDetector::use_dilation)
-      .def_readwrite("mean", &vision::ocr::DBDetector::mean)
-      .def_readwrite("scale", &vision::ocr::DBDetector::scale)
-      .def_readwrite("is_scale", &vision::ocr::DBDetector::is_scale);
+  pybind11::class_<vision::ocr::DBDetectorPreprocessor>(m, "DBDetectorPreprocessor")
+      .def(pybind11::init<>())
+      .def_readwrite("max_side_len", &vision::ocr::DBDetectorPreprocessor::max_side_len_)
+      .def_readwrite("mean", &vision::ocr::DBDetectorPreprocessor::mean_)
+      .def_readwrite("scale", &vision::ocr::DBDetectorPreprocessor::scale_)
+      .def_readwrite("is_scale", &vision::ocr::DBDetectorPreprocessor::is_scale_)
+      .def("run", [](vision::ocr::DBDetectorPreprocessor& self, std::vector<pybind11::array>& im_list) {
+        std::vector<vision::FDMat> images;
+        for (size_t i = 0; i < im_list.size(); ++i) {
+          images.push_back(vision::WrapMat(PyArrayToCvMat(im_list[i])));
+        }
+        std::vector<FDTensor> outputs;
+        std::vector<std::array<int, 4>> batch_det_img_info;
+        self.Run(&images, &outputs, &batch_det_img_info);
+        for(size_t i = 0; i< outputs.size(); ++i){
+          outputs[i].StopSharing();
+        }
+        return make_pair(outputs, batch_det_img_info);
+      });
+
+  pybind11::class_<vision::ocr::DBDetectorPostprocessor>(m, "DBDetectorPostprocessor")
+      .def(pybind11::init<>())
+      .def_readwrite("det_db_thresh", &vision::ocr::DBDetectorPostprocessor::det_db_thresh_)
+      .def_readwrite("det_db_box_thresh", &vision::ocr::DBDetectorPostprocessor::det_db_box_thresh_)
+      .def_readwrite("det_db_unclip_ratio", &vision::ocr::DBDetectorPostprocessor::det_db_unclip_ratio_)
+      .def_readwrite("det_db_score_mode", &vision::ocr::DBDetectorPostprocessor::det_db_score_mode_)
+      .def_readwrite("use_dilation", &vision::ocr::DBDetectorPostprocessor::use_dilation_)
+      .def("run", [](vision::ocr::DBDetectorPostprocessor& self,
+                     std::vector<FDTensor>& inputs,
+                     const std::vector<std::array<int, 4>>& batch_det_img_info) {
+        std::vector<std::vector<std::array<int, 8>>> results;
+
+        if (!self.Run(inputs, &results, batch_det_img_info)) {
+          pybind11::eval("raise Exception('Failed to preprocess the input data in DBDetectorPostprocessor.')");
+        }
+        return results;
+      })
+      .def("run", [](vision::ocr::DBDetectorPostprocessor& self,
+                     std::vector<pybind11::array>& input_array,
+                     const std::vector<std::array<int, 4>>& batch_det_img_info) {
+        std::vector<std::vector<std::array<int, 8>>> results;
+        std::vector<FDTensor> inputs;
+        PyArrayToTensorList(input_array, &inputs, /*share_buffer=*/true);
+        if (!self.Run(inputs, &results, batch_det_img_info)) {
+          pybind11::eval("raise Exception('Failed to preprocess the input data in DBDetectorPostprocessor.')");
+        }
+        return results;
+      });
 
   // Classifier
   pybind11::class_<vision::ocr::Classifier, FastDeployModel>(m, "Classifier")
       .def(pybind11::init<std::string, std::string, RuntimeOption,
                           ModelFormat>())
       .def(pybind11::init<>())
+      .def_readwrite("preprocessor", &vision::ocr::Classifier::preprocessor_)
+      .def_readwrite("postprocessor", &vision::ocr::Classifier::postprocessor_);
+
+    pybind11::class_<vision::ocr::ClassifierPreprocessor>(m, "ClassifierPreprocessor")
+      .def(pybind11::init<>())
+      .def_readwrite("cls_image_shape", &vision::ocr::ClassifierPreprocessor::cls_image_shape_)
+      .def_readwrite("mean", &vision::ocr::ClassifierPreprocessor::mean_)
+      .def_readwrite("scale", &vision::ocr::ClassifierPreprocessor::scale_)
+      .def_readwrite("is_scale", &vision::ocr::ClassifierPreprocessor::is_scale_)
+      .def("run", [](vision::ocr::ClassifierPreprocessor& self, std::vector<pybind11::array>& im_list) {
+        std::vector<vision::FDMat> images;
+        for (size_t i = 0; i < im_list.size(); ++i) {
+          images.push_back(vision::WrapMat(PyArrayToCvMat(im_list[i])));
+        }
+        std::vector<FDTensor> outputs;
+        if (!self.Run(&images, &outputs)) {
+          pybind11::eval("raise Exception('Failed to preprocess the input data in ClassifierPreprocessor.')");
+        }
+        for(size_t i = 0; i< outputs.size(); ++i){
+          outputs[i].StopSharing();
+        }
+        return outputs;
+      });
+
+  pybind11::class_<vision::ocr::ClassifierPostprocessor>(m, "ClassifierPostprocessor")
+      .def(pybind11::init<>())
+      .def_readwrite("cls_thresh", &vision::ocr::ClassifierPostprocessor::cls_thresh_)
+      .def("run", [](vision::ocr::ClassifierPostprocessor& self,
+                     std::vector<FDTensor>& inputs) {
+        std::vector<int> cls_labels;
+        std::vector<float> cls_scores;
+        if (!self.Run(inputs, &cls_labels, &cls_scores)) {
+          pybind11::eval("raise Exception('Failed to preprocess the input data in ClassifierPostprocessor.')");
+        }
+        return make_pair(cls_labels,cls_scores);
+      })
+      .def("run", [](vision::ocr::ClassifierPostprocessor& self,
+                     std::vector<pybind11::array>& input_array) {
+        std::vector<FDTensor> inputs;
+        PyArrayToTensorList(input_array, &inputs, /*share_buffer=*/true);
+        std::vector<int> cls_labels;
+        std::vector<float> cls_scores;
+        if (!self.Run(inputs, &cls_labels, &cls_scores)) {
+          pybind11::eval("raise Exception('Failed to preprocess the input data in ClassifierPostprocessor.')");
+        }
+        return make_pair(cls_labels,cls_scores);
+      });
 
-      .def_readwrite("cls_thresh", &vision::ocr::Classifier::cls_thresh)
-      .def_readwrite("cls_image_shape",
-                     &vision::ocr::Classifier::cls_image_shape)
-      .def_readwrite("cls_batch_num", &vision::ocr::Classifier::cls_batch_num);
 
   // Recognizer
   pybind11::class_<vision::ocr::Recognizer, FastDeployModel>(m, "Recognizer")
-
       .def(pybind11::init<std::string, std::string, std::string, RuntimeOption,
                           ModelFormat>())
       .def(pybind11::init<>())
+      .def_readwrite("preprocessor", &vision::ocr::Recognizer::preprocessor_)
+      .def_readwrite("postprocessor", &vision::ocr::Recognizer::postprocessor_);
 
-      .def_readwrite("rec_img_h", &vision::ocr::Recognizer::rec_img_h)
-      .def_readwrite("rec_img_w", &vision::ocr::Recognizer::rec_img_w)
-      .def_readwrite("rec_batch_num", &vision::ocr::Recognizer::rec_batch_num);
+    pybind11::class_<vision::ocr::RecognizerPreprocessor>(m, "RecognizerPreprocessor")
+      .def(pybind11::init<>())
+      .def_readwrite("rec_image_shape", &vision::ocr::RecognizerPreprocessor::rec_image_shape_)
+      .def_readwrite("mean", &vision::ocr::RecognizerPreprocessor::mean_)
+      .def_readwrite("scale", &vision::ocr::RecognizerPreprocessor::scale_)
+      .def_readwrite("is_scale", &vision::ocr::RecognizerPreprocessor::is_scale_)
+      .def("run", [](vision::ocr::RecognizerPreprocessor& self, std::vector<pybind11::array>& im_list) {
+        std::vector<vision::FDMat> images;
+        for (size_t i = 0; i < im_list.size(); ++i) {
+          images.push_back(vision::WrapMat(PyArrayToCvMat(im_list[i])));
+        }
+        std::vector<FDTensor> outputs;
+        if (!self.Run(&images, &outputs)) {
+          pybind11::eval("raise Exception('Failed to preprocess the input data in RecognizerPreprocessor.')");
+        }
+        for(size_t i = 0; i< outputs.size(); ++i){
+          outputs[i].StopSharing();
+        }
+        return outputs;
+      });
+
+  pybind11::class_<vision::ocr::RecognizerPostprocessor>(m, "RecognizerPostprocessor")
+      .def(pybind11::init<std::string>())
+      .def("run", [](vision::ocr::RecognizerPostprocessor& self,
+                     std::vector<FDTensor>& inputs) {
+        std::vector<std::string> texts;
+        std::vector<float> rec_scores;
+        if (!self.Run(inputs, &texts, &rec_scores)) {
+          pybind11::eval("raise Exception('Failed to preprocess the input data in RecognizerPostprocessor.')");
+        }
+        return make_pair(texts, rec_scores);
+      })
+      .def("run", [](vision::ocr::RecognizerPostprocessor& self,
+                     std::vector<pybind11::array>& input_array) {
+        std::vector<FDTensor> inputs;
+        PyArrayToTensorList(input_array, &inputs, /*share_buffer=*/true);
+        std::vector<std::string> texts;
+        std::vector<float> rec_scores;
+        if (!self.Run(inputs, &texts, &rec_scores)) {
+          pybind11::eval("raise Exception('Failed to preprocess the input data in RecognizerPostprocessor.')");
+        }
+        return make_pair(texts, rec_scores);
+      });
 }
 }  // namespace fastdeploy

fastdeploy/vision/ocr/ppocr/ppocr_pybind.cc

@@ -31,6 +31,15 @@ void BindPPOCRv3(pybind11::module& m) {
         vision::OCRResult res;
         self.Predict(&mat, &res);
         return res;
+      })
+      .def("batch_predict", [](pipeline::PPOCRv3& self, std::vector<pybind11::array>& data) {
+        std::vector<cv::Mat> images;
+        for (size_t i = 0; i < data.size(); ++i) {
+          images.push_back(PyArrayToCvMat(data[i]));
+        }
+        std::vector<vision::OCRResult> results;
+        self.BatchPredict(images, &results);
+        return results;
       });
 }
 
@@ -49,6 +58,15 @@ void BindPPOCRv2(pybind11::module& m) {
         vision::OCRResult res;
         self.Predict(&mat, &res);
         return res;
+      })
+      .def("batch_predict", [](pipeline::PPOCRv2& self, std::vector<pybind11::array>& data) {
+        std::vector<cv::Mat> images;
+        for (size_t i = 0; i < data.size(); ++i) {
+          images.push_back(PyArrayToCvMat(data[i]));
+        }
+        std::vector<vision::OCRResult> results;
+        self.BatchPredict(images, &results);
+        return results;
       });
 }
 

fastdeploy/vision/ocr/ppocr/ppocr_v2.cc

@@ -22,13 +22,15 @@ PPOCRv2::PPOCRv2(fastdeploy::vision::ocr::DBDetector* det_model,
                              fastdeploy::vision::ocr::Classifier* cls_model,
                              fastdeploy::vision::ocr::Recognizer* rec_model)
     : detector_(det_model), classifier_(cls_model), recognizer_(rec_model) {
-  recognizer_->rec_image_shape[1] = 32;
+  Initialized();
+  recognizer_->preprocessor_.rec_image_shape_[1] = 32;
 }
 
 PPOCRv2::PPOCRv2(fastdeploy::vision::ocr::DBDetector* det_model,
                              fastdeploy::vision::ocr::Recognizer* rec_model)
     : detector_(det_model), recognizer_(rec_model) {
-  recognizer_->rec_image_shape[1] = 32;
+  Initialized();
+  recognizer_->preprocessor_.rec_image_shape_[1] = 32;
 }
 
 bool PPOCRv2::Initialized() const {
@@ -47,76 +49,68 @@ bool PPOCRv2::Initialized() const {
   return true; 
 }
 
-bool PPOCRv2::Detect(cv::Mat* img,
-                           fastdeploy::vision::OCRResult* result) {
-  if (!detector_->Predict(img, &(result->boxes))) {
-    FDERROR << "There's error while detecting image in PPOCR." << std::endl;
-    return false;
-  }
-  vision::ocr::SortBoxes(result);
-  return true;
-}
-
-bool PPOCRv2::Recognize(cv::Mat* img,
-                              fastdeploy::vision::OCRResult* result) {
-  std::tuple<std::string, float> rec_result;
-  if (!recognizer_->Predict(img, &rec_result)) {
-    FDERROR << "There's error while recognizing image in PPOCR." << std::endl;
-    return false;
-  }
-
-  result->text.push_back(std::get<0>(rec_result));
-  result->rec_scores.push_back(std::get<1>(rec_result));
-  return true;
-}
-
-bool PPOCRv2::Classify(cv::Mat* img,
-                             fastdeploy::vision::OCRResult* result) {
-  std::tuple<int, float> cls_result;
-
-  if (!classifier_->Predict(img, &cls_result)) {
-    FDERROR << "There's error while classifying image in PPOCR." << std::endl;
-    return false;
-  }
-
-  result->cls_labels.push_back(std::get<0>(cls_result));
-  result->cls_scores.push_back(std::get<1>(cls_result));
-  return true;
-}
-
 bool PPOCRv2::Predict(cv::Mat* img,
                             fastdeploy::vision::OCRResult* result) {
-  result->Clear();
-  if (nullptr != detector_ && !Detect(img, result)) {
-    FDERROR << "Failed to detect image." << std::endl;
-    return false;
-  }
-
-  // Get croped images by detection result
-  std::vector<cv::Mat> image_list;
-  for (size_t i = 0; i < result->boxes.size(); ++i) {
-    auto crop_im = vision::ocr::GetRotateCropImage(*img, (result->boxes)[i]);
-    image_list.push_back(crop_im);
-  }
-  if (result->boxes.size() == 0) {
-    image_list.push_back(*img);
-  }
-
-  for (size_t i = 0; i < image_list.size(); ++i) {
-    if (nullptr != classifier_ && !Classify(&(image_list[i]), result)) {
-      FDERROR << "Failed to classify croped image of index " << i << "." << std::endl;
-      return false;
-    }
-    if (nullptr != classifier_ && result->cls_labels[i] % 2 == 1 && result->cls_scores[i] > classifier_->cls_thresh) {
-      cv::rotate(image_list[i], image_list[i], 1);
-    }
-    if (nullptr != recognizer_ && !Recognize(&(image_list[i]), result)) {
-      FDERROR << "Failed to recgnize croped image of index " << i << "." << std::endl;
-      return false;
-    }
-  }
+  std::vector<fastdeploy::vision::OCRResult> batch_result(1);
+  BatchPredict({*img},&batch_result);
+  *result = std::move(batch_result[0]);
   return true;
 };
 
+bool PPOCRv2::BatchPredict(const std::vector<cv::Mat>& images,
+                           std::vector<fastdeploy::vision::OCRResult>* batch_result) {
+  batch_result->clear();
+  batch_result->resize(images.size());
+  std::vector<std::vector<std::array<int, 8>>> batch_boxes(images.size());
+
+  if (!detector_->BatchPredict(images, &batch_boxes)) {
+    FDERROR << "There's error while detecting image in PPOCR." << std::endl;
+    return false;
+  }
+  for(int i_batch = 0; i_batch < batch_boxes.size(); ++i_batch) {
+    vision::ocr::SortBoxes(&(batch_boxes[i_batch]));
+    (*batch_result)[i_batch].boxes = batch_boxes[i_batch];
+  }
+  
+  
+  for(int i_batch = 0; i_batch < images.size(); ++i_batch) {
+    fastdeploy::vision::OCRResult& ocr_result = (*batch_result)[i_batch];
+    // Get croped images by detection result
+    const std::vector<std::array<int, 8>>& boxes = ocr_result.boxes;
+    const cv::Mat& img = images[i_batch];
+    std::vector<cv::Mat> image_list;
+    if (boxes.size() == 0) {
+      image_list.emplace_back(img);
+    }else{
+      image_list.resize(boxes.size());
+      for (size_t i_box = 0; i_box < boxes.size(); ++i_box) {
+        image_list[i_box] = vision::ocr::GetRotateCropImage(img, boxes[i_box]);
+      }
+    }
+    std::vector<int32_t>* cls_labels_ptr = &ocr_result.cls_labels;
+    std::vector<float>* cls_scores_ptr = &ocr_result.cls_scores;
+
+    std::vector<std::string>* text_ptr = &ocr_result.text;
+    std::vector<float>* rec_scores_ptr = &ocr_result.rec_scores;
+
+    if (!classifier_->BatchPredict(image_list, cls_labels_ptr, cls_scores_ptr)) {
+      FDERROR << "There's error while recognizing image in PPOCR." << std::endl;
+      return false;
+    }else{
+      for (size_t i_img = 0; i_img < image_list.size(); ++i_img) {
+        if(cls_labels_ptr->at(i_img) % 2 == 1 && cls_scores_ptr->at(i_img) > classifier_->postprocessor_.cls_thresh_) {
+          cv::rotate(image_list[i_img], image_list[i_img], 1);
+        }
+      }
+    }
+
+    if (!recognizer_->BatchPredict(image_list, text_ptr, rec_scores_ptr)) {
+      FDERROR << "There's error while recognizing image in PPOCR." << std::endl;
+      return false;
+    }
+  }
+  return true;
+}
+
 }  // namesapce pipeline
 }  // namespace fastdeploy

fastdeploy/vision/ocr/ppocr/ppocr_v2.h

@@ -59,6 +59,14 @@ class FASTDEPLOY_DECL PPOCRv2 : public FastDeployModel {
    * \return true if the prediction successed, otherwise false.
    */
   virtual bool Predict(cv::Mat* img, fastdeploy::vision::OCRResult* result);
+  /** \brief BatchPredict the input image and get OCR result.
+   *
+   * \param[in] images The list of input image data, comes from cv::imread(), is a 3-D array with layout HWC, BGR format.
+   * \param[in] batch_result The output list of OCR result will be writen to this structure.
+   * \return true if the prediction successed, otherwise false.
+   */
+  virtual bool BatchPredict(const std::vector<cv::Mat>& images,
+               std::vector<fastdeploy::vision::OCRResult>* batch_result);
   bool Initialized() const override;
 
  protected:
@@ -66,11 +74,6 @@ class FASTDEPLOY_DECL PPOCRv2 : public FastDeployModel {
   fastdeploy::vision::ocr::Classifier* classifier_ = nullptr;
   fastdeploy::vision::ocr::Recognizer* recognizer_ = nullptr;
   /// Launch the detection process in OCR.
-  virtual bool Detect(cv::Mat* img, fastdeploy::vision::OCRResult* result);
-  /// Launch the recognition process in OCR.
-  virtual bool Recognize(cv::Mat* img, fastdeploy::vision::OCRResult* result);
-  /// Launch the classification process in OCR.
-  virtual bool Classify(cv::Mat* img, fastdeploy::vision::OCRResult* result);
 };
 
 namespace application {

fastdeploy/vision/ocr/ppocr/ppocr_v3.h

@@ -36,7 +36,7 @@ class FASTDEPLOY_DECL PPOCRv3 : public PPOCRv2 {
                 fastdeploy::vision::ocr::Recognizer* rec_model)
                 : PPOCRv2(det_model, cls_model, rec_model) {
     // The only difference between v2 and v3
-    recognizer_->rec_image_shape[1] = 48;
+    recognizer_->preprocessor_.rec_image_shape_[1] = 48;
   }
   /** \brief Classification model is optional, so this function is set up the detection model path and recognition model path respectively.
    *
@@ -47,7 +47,7 @@ class FASTDEPLOY_DECL PPOCRv3 : public PPOCRv2 {
                 fastdeploy::vision::ocr::Recognizer* rec_model)
                 : PPOCRv2(det_model, rec_model) {
     // The only difference between v2 and v3
-    recognizer_->rec_image_shape[1] = 48;
+    recognizer_->preprocessor_.rec_image_shape_[1] = 48;
   }
 };
 

fastdeploy/vision/ocr/ppocr/rec_postprocessor.cc

@@ -0,0 +1,112 @@
+// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "fastdeploy/vision/ocr/ppocr/rec_postprocessor.h"
+#include "fastdeploy/utils/perf.h"
+#include "fastdeploy/vision/ocr/ppocr/utils/ocr_utils.h"
+
+namespace fastdeploy {
+namespace vision {
+namespace ocr {
+
+std::vector<std::string> ReadDict(const std::string& path) {
+  std::ifstream in(path);
+  FDASSERT(in, "Cannot open file %s to read.", path.c_str());
+  std::string line;
+  std::vector<std::string> m_vec;
+  while (getline(in, line)) {
+    m_vec.push_back(line);
+  }
+  m_vec.insert(m_vec.begin(), "#");  // blank char for ctc
+  m_vec.push_back(" ");
+  return m_vec;
+}
+
+RecognizerPostprocessor::RecognizerPostprocessor(){
+  initialized_ = true;
+}
+
+RecognizerPostprocessor::RecognizerPostprocessor(const std::string& label_path) {
+  // init label_lsit
+  label_list_ = ReadDict(label_path);
+  initialized_ = true;
+}
+
+bool RecognizerPostprocessor::SingleBatchPostprocessor(const float* out_data,
+                              const std::vector<int64_t>& output_shape,
+                              std::string* text, float* rec_score) {
+  std::string& str_res = *text;
+  float& score = *rec_score;
+  score = 0.f;
+  int argmax_idx;
+  int last_index = 0;
+  int count = 0;
+  float max_value = 0.0f;
+
+  for (int n = 0; n < output_shape[1]; n++) {
+    argmax_idx = int(
+        std::distance(&out_data[n * output_shape[2]],
+                      std::max_element(&out_data[n * output_shape[2]],
+                                       &out_data[(n + 1) * output_shape[2]])));
+
+    max_value = float(*std::max_element(&out_data[n * output_shape[2]],
+                                        &out_data[(n + 1) * output_shape[2]]));
+
+    if (argmax_idx > 0 && (!(n > 0 && argmax_idx == last_index))) {
+      score += max_value;
+      count += 1;
+      if(argmax_idx > label_list_.size()) {
+        FDERROR << "The output index: " << argmax_idx << " is larger than the size of label_list: "
+        << label_list_.size() << ". Please check the label file!" << std::endl;
+        return false; 
+      }
+      str_res += label_list_[argmax_idx];
+    }
+    last_index = argmax_idx;
+  }
+  score /= (count + 1e-6);
+  if (count == 0 || std::isnan(score)) {
+    score = 0.f;
+  }
+  return true;
+}
+
+bool RecognizerPostprocessor::Run(const std::vector<FDTensor>& tensors,
+                                  std::vector<std::string>* texts, std::vector<float>* rec_scores) {
+  if (!initialized_) {
+    FDERROR << "Postprocessor is not initialized." << std::endl;
+    return false;
+  }
+  // Recognizer have only 1 output tensor.
+  const FDTensor& tensor = tensors[0];
+  // For Recognizer, the output tensor shape = [batch, ?, 6625]
+  size_t batch = tensor.shape[0];
+  size_t length = accumulate(tensor.shape.begin()+1, tensor.shape.end(), 1, std::multiplies<int>());
+
+  texts->resize(batch);
+  rec_scores->resize(batch);
+  const float* tensor_data = reinterpret_cast<const float*>(tensor.Data());
+    for (int i_batch = 0; i_batch < batch; ++i_batch) {
+    if(!SingleBatchPostprocessor(tensor_data, tensor.shape, &texts->at(i_batch), &rec_scores->at(i_batch))) {
+      return false;
+    }
+    tensor_data = tensor_data + length;
+  }
+
+  return true;
+}
+
+}  // namespace ocr
+}  // namespace vision
+}  // namespace fastdeploy

fastdeploy/vision/ocr/ppocr/rec_postprocessor.h

@@ -0,0 +1,55 @@
+// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+#include "fastdeploy/vision/common/processors/transform.h"
+#include "fastdeploy/vision/common/result.h"
+#include "fastdeploy/vision/ocr/ppocr/utils/ocr_postprocess_op.h"
+
+namespace fastdeploy {
+namespace vision {
+
+namespace ocr {
+/*! @brief Postprocessor object for Recognizer serials model.
+ */
+class FASTDEPLOY_DECL RecognizerPostprocessor {
+ public:
+  RecognizerPostprocessor();
+  /** \brief Create a postprocessor instance for Recognizer serials model
+   *
+   * \param[in] label_path The path of label_dict
+   */
+  explicit RecognizerPostprocessor(const std::string& label_path);
+
+  /** \brief Process the result of runtime and fill to ClassifyResult structure
+   *
+   * \param[in] tensors The inference result from runtime
+   * \param[in] texts The output result of recognizer
+   * \param[in] rec_scores The output result of recognizer
+   * \return true if the postprocess successed, otherwise false
+   */
+  bool Run(const std::vector<FDTensor>& tensors,
+           std::vector<std::string>* texts, std::vector<float>* rec_scores);
+
+ private:
+  bool SingleBatchPostprocessor(const float* out_data,
+                              const std::vector<int64_t>& output_shape,
+                              std::string* text, float* rec_score);
+  bool initialized_ = false;
+  std::vector<std::string> label_list_;
+};
+
+}  // namespace ocr
+}  // namespace vision
+}  // namespace fastdeploy

fastdeploy/vision/ocr/ppocr/rec_preprocessor.cc

@@ -0,0 +1,99 @@
+// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "fastdeploy/vision/ocr/ppocr/rec_preprocessor.h"
+#include "fastdeploy/utils/perf.h"
+#include "fastdeploy/vision/ocr/ppocr/utils/ocr_utils.h"
+#include "fastdeploy/function/concat.h"
+
+namespace fastdeploy {
+namespace vision {
+namespace ocr {
+
+RecognizerPreprocessor::RecognizerPreprocessor() {
+  initialized_ = true;
+}
+
+void OcrRecognizerResizeImage(FDMat* mat, float max_wh_ratio,
+                              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(imgH * max_wh_ratio);
+
+  float ratio = float(mat->Width()) / float(mat->Height());
+  int resize_w;
+  if (ceilf(imgH * ratio) > imgW) {
+    resize_w = imgW;
+  }else{
+    resize_w = int(ceilf(imgH * ratio));
+  }
+  Resize::Run(mat, resize_w, imgH);
+
+  std::vector<float> value = {0, 0, 0};
+  Pad::Run(mat, 0, 0, 0, int(imgW - mat->Width()), value);
+}
+
+bool RecognizerPreprocessor::Run(std::vector<FDMat>* images, std::vector<FDTensor>* outputs) {
+  if (!initialized_) {
+    FDERROR << "The preprocessor is not initialized." << std::endl;
+    return false;
+  }
+  if (images->size() == 0) {
+    FDERROR << "The size of input images should be greater than 0." << std::endl;
+    return false;
+  }
+
+  int imgH = rec_image_shape_[1];
+  int imgW = rec_image_shape_[2];
+  float max_wh_ratio = imgW * 1.0 / imgH;
+  float ori_wh_ratio;
+  
+  for (size_t i = 0; i < images->size(); ++i) {
+    FDMat* mat = &(images->at(i));
+    ori_wh_ratio = mat->Width() * 1.0 / mat->Height();
+    max_wh_ratio = std::max(max_wh_ratio, ori_wh_ratio);
+  }
+
+  for (size_t i = 0; i < images->size(); ++i) {
+    FDMat* mat = &(images->at(i));
+    OcrRecognizerResizeImage(mat, max_wh_ratio, rec_image_shape_);
+    NormalizeAndPermute::Run(mat, mean_, scale_, is_scale_);
+    /*
+    Normalize::Run(mat, mean_, scale_, is_scale_);
+    HWC2CHW::Run(mat);
+    Cast::Run(mat, "float");
+    */
+  }
+  // Only have 1 output Tensor.
+  outputs->resize(1);
+  // Concat all the preprocessed data to a batch tensor
+  std::vector<FDTensor> tensors(images->size()); 
+  for (size_t i = 0; i < images->size(); ++i) {
+    (*images)[i].ShareWithTensor(&(tensors[i]));
+    tensors[i].ExpandDim(0);
+  }
+  if (tensors.size() == 1) {
+    (*outputs)[0] = std::move(tensors[0]);
+  } else {
+    function::Concat(tensors, &((*outputs)[0]), 0);
+  }
+  return true;
+}
+
+}  // namespace ocr
+}  // namespace vision
+}  // namespace fastdeploy

fastdeploy/vision/ocr/ppocr/rec_preprocessor.h

@@ -0,0 +1,52 @@
+// Copyright (c) 2022 PaddlePaddle Authors. All Rights Reserved.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+#include "fastdeploy/vision/common/processors/transform.h"
+#include "fastdeploy/vision/common/result.h"
+
+namespace fastdeploy {
+namespace vision {
+
+namespace ocr {
+/*! @brief Preprocessor object for PaddleClas serials model.
+ */
+class FASTDEPLOY_DECL RecognizerPreprocessor {
+ public:
+  /** \brief Create a preprocessor instance for PaddleClas serials model
+   *
+   * \param[in] config_file Path of configuration file for deployment, e.g resnet/infer_cfg.yml
+   */
+  RecognizerPreprocessor();
+
+  /** \brief Process the input image and prepare input tensors for runtime
+   *
+   * \param[in] images The input image data list, all the elements are returned by cv::imread()
+   * \param[in] outputs The output tensors which will feed in runtime
+   * \return true if the preprocess successed, otherwise false
+   */
+  bool Run(std::vector<FDMat>* images, std::vector<FDTensor>* outputs);
+
+  std::vector<int> rec_image_shape_ = {3, 48, 320};
+  std::vector<float> mean_ = {0.5f, 0.5f, 0.5f};
+  std::vector<float> scale_ = {0.5f, 0.5f, 0.5f};
+  bool is_scale_ = true;
+
+ private:
+  bool initialized_ = false;
+};
+
+}  // namespace ocr
+}  // namespace vision
+}  // namespace fastdeploy

fastdeploy/vision/ocr/ppocr/recognizer.cc

@@ -20,29 +20,13 @@ namespace fastdeploy {
 namespace vision {
 namespace ocr {
 
-std::vector<std::string> 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;
-}
-
 Recognizer::Recognizer() {}
 
 Recognizer::Recognizer(const std::string& model_file,
                        const std::string& params_file,
                        const std::string& label_path,
                        const RuntimeOption& custom_option,
-                       const ModelFormat& model_format) {
+                       const ModelFormat& model_format):postprocessor_(label_path) {
   if (model_format == ModelFormat::ONNX) {
     valid_cpu_backends = {Backend::ORT,
                           Backend::OPENVINO};  
@@ -56,27 +40,11 @@ Recognizer::Recognizer(const std::string& model_file,
   runtime_option.model_format = model_format;
   runtime_option.model_file = model_file;
   runtime_option.params_file = params_file;
-
   initialized = Initialize();
-
-  // init label_lsit
-  label_list = ReadDict(label_path);
-  label_list.insert(label_list.begin(), "#");  // blank char for ctc
-  label_list.push_back(" ");
 }
 
 // Init
 bool Recognizer::Initialize() {
-  // pre&post process parameters
-  rec_batch_num = 1;
-  rec_img_h = 48;
-  rec_img_w = 320;
-  rec_image_shape = {3, rec_img_h, rec_img_w};
-
-  mean = {0.5f, 0.5f, 0.5f};
-  scale = {0.5f, 0.5f, 0.5f};
-  is_scale = true;
-
   if (!InitRuntime()) {
     FDERROR << "Failed to initialize fastdeploy backend." << std::endl;
     return false;
@@ -85,119 +53,23 @@ bool Recognizer::Initialize() {
   return true;
 }
 
-void OcrRecognizerResizeImage(Mat* mat, const float& wh_ratio,
-                              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(imgH * wh_ratio);
-
-  float ratio = float(mat->Width()) / float(mat->Height());
-  int resize_w;
-  if (ceilf(imgH * ratio) > imgW)
-    resize_w = imgW;
-  else
-    resize_w = int(ceilf(imgH * ratio));
-
-  Resize::Run(mat, resize_w, imgH);
-
-  std::vector<float> value = {127, 127, 127};
-  Pad::Run(mat, 0, 0, 0, int(imgW - mat->Width()), value);
-}
-
-bool Recognizer::Preprocess(Mat* mat, FDTensor* output,
-                            const std::vector<int>& rec_image_shape) {
-  int imgH = rec_image_shape[1];
-  int imgW = rec_image_shape[2];
-  float wh_ratio = imgW * 1.0 / imgH;
-
-  float ori_wh_ratio = mat->Width() * 1.0 / mat->Height();
-  wh_ratio = std::max(wh_ratio, ori_wh_ratio);
-
-  OcrRecognizerResizeImage(mat, wh_ratio, rec_image_shape);
-
-  Normalize::Run(mat, mean, scale, true);
-
-  HWC2CHW::Run(mat);
-  Cast::Run(mat, "float");
-
-  mat->ShareWithTensor(output);
-  output->shape.insert(output->shape.begin(), 1);
-
-  return true;
-}
-
-bool Recognizer::Postprocess(FDTensor& infer_result,
-                             std::tuple<std::string, float>* rec_result) {
-  std::vector<int64_t> output_shape = infer_result.shape;
-  FDASSERT(output_shape[0] == 1, "Only support batch =1 now.");
-
-  float* out_data = static_cast<float*>(infer_result.Data());
-
-  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 < output_shape[1]; n++) {
-    argmax_idx = int(
-        std::distance(&out_data[n * output_shape[2]],
-                      std::max_element(&out_data[n * output_shape[2]],
-                                       &out_data[(n + 1) * output_shape[2]])));
-
-    max_value = float(*std::max_element(&out_data[n * output_shape[2]],
-                                        &out_data[(n + 1) * output_shape[2]]));
-
-    if (argmax_idx > 0 && (!(n > 0 && argmax_idx == last_index))) {
-      score += max_value;
-      count += 1;
-      if(argmax_idx > label_list.size()){
-        FDERROR << "The output index: " << argmax_idx << " is larger than the size of label_list: "
-        << label_list.size() << ". Please check the label file!" << std::endl;
+bool Recognizer::BatchPredict(const std::vector<cv::Mat>& images,
+                              std::vector<std::string>* texts, std::vector<float>* rec_scores) {
+  std::vector<FDMat> fd_images = WrapMat(images);
+  if (!preprocessor_.Run(&fd_images, &reused_input_tensors_)) {
+    FDERROR << "Failed to preprocess the input image." << std::endl;
     return false;
   }
-      str_res += label_list[argmax_idx];
-    }
-    last_index = argmax_idx;
-  }
-  score /= (count + 1e-6);
-  if (count == 0 || std::isnan(score)) {
-    score = 0.f;
-  }
-  std::get<0>(*rec_result) = str_res;
-  std::get<1>(*rec_result) = score;
-
-  return true;
-}
-
-bool Recognizer::Predict(cv::Mat* img,
-                         std::tuple<std::string, float>* rec_result) {
-  Mat mat(*img);
-
-  std::vector<FDTensor> input_tensors(1);
-
-  if (!Preprocess(&mat, &input_tensors[0], rec_image_shape)) {
-    FDERROR << "Failed to preprocess input image." << std::endl;
+  reused_input_tensors_[0].name = InputInfoOfRuntime(0).name;
+  if (!Infer(reused_input_tensors_, &reused_output_tensors_)) {
+    FDERROR << "Failed to inference by runtime." << std::endl;
     return false;
   }
 
-  input_tensors[0].name = InputInfoOfRuntime(0).name;
-  std::vector<FDTensor> output_tensors;
-
-  if (!Infer(input_tensors, &output_tensors)) {
-    FDERROR << "Failed to inference." << std::endl;
+  if (!postprocessor_.Run(reused_output_tensors_, texts, rec_scores)) {
+    FDERROR << "Failed to postprocess the inference cls_results by runtime." << std::endl;
     return false;
   }
-
-  if (!Postprocess(output_tensors[0], rec_result)) {
-    FDERROR << "Failed to post process." << std::endl;
-    return false;
-  }
-
   return true;
 }
 

fastdeploy/vision/ocr/ppocr/recognizer.h

@@ -17,6 +17,8 @@
 #include "fastdeploy/vision/common/processors/transform.h"
 #include "fastdeploy/vision/common/result.h"
 #include "fastdeploy/vision/ocr/ppocr/utils/ocr_postprocess_op.h"
+#include "fastdeploy/vision/ocr/ppocr/rec_preprocessor.h"
+#include "fastdeploy/vision/ocr/ppocr/rec_postprocessor.h"
 
 namespace fastdeploy {
 namespace vision {
@@ -43,35 +45,20 @@ class FASTDEPLOY_DECL Recognizer : public FastDeployModel {
              const ModelFormat& model_format = ModelFormat::PADDLE);
   /// Get model's name
   std::string ModelName() const { return "ppocr/ocr_rec"; }
-  /** \brief Predict the input image and get OCR recognition model result.
+  /** \brief BatchPredict the input image and get OCR recognition model result.
    *
-   * \param[in] im The input image data, comes from cv::imread(), is a 3-D array with layout HWC, BGR format.
-   * \param[in] rec_result The output of OCR recognition model result will be writen to this structure.
+   * \param[in] images The list of input image data, comes from cv::imread(), is a 3-D array with layout HWC, BGR format.
+   * \param[in] rec_results The output of OCR recognition model result will be writen to this structure.
    * \return true if the prediction is successed, otherwise false.
    */
-  virtual bool Predict(cv::Mat* img,
-                       std::tuple<std::string, float>* rec_result);
+  virtual bool BatchPredict(const std::vector<cv::Mat>& images,
+               std::vector<std::string>* texts, std::vector<float>* rec_scores);
 
-  // Pre & Post parameters
-  std::vector<std::string> label_list;
-  int rec_batch_num;
-  int rec_img_h;
-  int rec_img_w;
-  std::vector<int> rec_image_shape;
-
-  std::vector<float> mean;
-  std::vector<float> scale;
-  bool is_scale;
+  RecognizerPreprocessor preprocessor_;
+  RecognizerPostprocessor postprocessor_;
 
  private:
   bool Initialize();
-  /// Preprocess the input data, and set the preprocessed results to `outputs`
-  bool Preprocess(Mat* img, FDTensor* outputs,
-                  const std::vector<int>& rec_image_shape);
-  /*! @brief Postprocess the inferenced results, and set the final result to `rec_result`
-   */
-  bool Postprocess(FDTensor& infer_result,
-                   std::tuple<std::string, float>* rec_result);
 };
 
 }  // namespace ocr

fastdeploy/vision/ocr/ppocr/utils/ocr_postprocess_op.cc

@@ -318,10 +318,12 @@ std::vector<std::vector<std::vector<int>>> PostProcessor::BoxesFromBitmap(
 }
 
 std::vector<std::vector<std::vector<int>>> PostProcessor::FilterTagDetRes(
-    std::vector<std::vector<std::vector<int>>> boxes, float ratio_h,
-    float ratio_w, const std::map<std::string, std::array<float, 2>> &im_info) {
-  int oriimg_h = im_info.at("input_shape")[0];
-  int oriimg_w = im_info.at("input_shape")[1];
+    std::vector<std::vector<std::vector<int>>> boxes,
+    const std::array<int,4>& det_img_info) {
+  int oriimg_w = det_img_info[0];
+  int oriimg_h = det_img_info[1];
+  float ratio_w = float(det_img_info[2])/float(oriimg_w);
+  float ratio_h = float(det_img_info[3])/float(oriimg_h);
 
   std::vector<std::vector<std::vector<int>>> root_points;
   for (int n = 0; n < boxes.size(); n++) {

fastdeploy/vision/ocr/ppocr/utils/ocr_postprocess_op.h

@@ -57,9 +57,8 @@ class PostProcessor {
       const float &det_db_unclip_ratio, const std::string &det_db_score_mode);
 
   std::vector<std::vector<std::vector<int>>> FilterTagDetRes(
-      std::vector<std::vector<std::vector<int>>> boxes, float ratio_h,
-      float ratio_w,
-      const std::map<std::string, std::array<float, 2>> &im_info);
+      std::vector<std::vector<std::vector<int>>> boxes,
+      const std::array<int, 4>& det_img_info);
 
  private:
   static bool XsortInt(std::vector<int> a, std::vector<int> b);

fastdeploy/vision/ocr/ppocr/utils/ocr_utils.h

@@ -28,10 +28,10 @@ namespace fastdeploy {
 namespace vision {
 namespace ocr {
 
-cv::Mat GetRotateCropImage(const cv::Mat& srcimage,
+FASTDEPLOY_DECL cv::Mat GetRotateCropImage(const cv::Mat& srcimage,
                            const std::array<int, 8>& box);
 
-void SortBoxes(OCRResult* result);
+FASTDEPLOY_DECL void SortBoxes(std::vector<std::array<int, 8>>* boxes);
 
 }  // namespace ocr
 }  // namespace vision

fastdeploy/vision/ocr/ppocr/utils/sorted_boxes.cc

@@ -29,17 +29,17 @@ bool CompareBox(const std::array<int, 8>& result1,
   }
 }
 
-void SortBoxes(OCRResult* result) {
-  std::sort(result->boxes.begin(), result->boxes.end(), CompareBox);
+void SortBoxes(std::vector<std::array<int, 8>>* boxes) {
+  std::sort(boxes->begin(), boxes->end(), CompareBox);
 
-  if (result->boxes.size() == 0) {
+  if (boxes->size() == 0) {
     return;
   }
 
-  for (int i = 0; i < result->boxes.size() - 1; i++) {
-    if (abs(result->boxes[i + 1][1] - result->boxes[i][1]) < 10 &&
-        (result->boxes[i + 1][0] < result->boxes[i][0])) {
-      std::swap(result->boxes[i], result->boxes[i + 1]);
+  for (int i = 0; i < boxes->size() - 1; i++) {
+    if (abs((*boxes)[i + 1][1] - (*boxes)[i][1]) < 10 &&
+        ((*boxes)[i + 1][0] < (*boxes)[i][0])) {
+      std::swap((*boxes)[i], (*boxes)[i + 1]);
     }
   }
 }

python/fastdeploy/vision/ocr/__init__.py

@@ -13,10 +13,4 @@
 # limitations under the License.
 from __future__ import absolute_import
 
-from .ppocr import PPOCRv3
-from .ppocr import PPOCRv2
-from .ppocr import PPOCRSystemv3
-from .ppocr import PPOCRSystemv2
-from .ppocr import DBDetector
-from .ppocr import Classifier
-from .ppocr import Recognizer
+from .ppocr import *

python/fastdeploy/vision/ocr/ppocr/__init__.py

@@ -41,40 +41,11 @@ class DBDetector(FastDeployModel):
             assert self.initialized, "DBDetector initialize failed."
 
     # 一些跟DBDetector模型有关的属性封装
-    @property
-    def max_side_len(self):
-        return self._model.max_side_len
-
+    '''
     @property
     def det_db_thresh(self):
         return self._model.det_db_thresh
 
-    @property
-    def det_db_box_thresh(self):
-        return self._model.det_db_box_thresh
-
-    @property
-    def det_db_unclip_ratio(self):
-        return self._model.det_db_unclip_ratio
-
-    @property
-    def det_db_score_mode(self):
-        return self._model.det_db_score_mode
-
-    @property
-    def use_dilation(self):
-        return self._model.use_dilation
-
-    @property
-    def is_scale(self):
-        return self._model.max_wh
-
-    @max_side_len.setter
-    def max_side_len(self, value):
-        assert isinstance(
-            value, int), "The value to set `max_side_len` must be type of int."
-        self._model.max_side_len = value
-
     @det_db_thresh.setter
     def det_db_thresh(self, value):
         assert isinstance(
@@ -82,6 +53,10 @@ class DBDetector(FastDeployModel):
             float), "The value to set `det_db_thresh` must be type of float."
         self._model.det_db_thresh = value
 
+    @property
+    def det_db_box_thresh(self):
+        return self._model.det_db_box_thresh
+
     @det_db_box_thresh.setter
     def det_db_box_thresh(self, value):
         assert isinstance(
@@ -89,6 +64,10 @@ class DBDetector(FastDeployModel):
         ), "The value to set `det_db_box_thresh` must be type of float."
         self._model.det_db_box_thresh = value
 
+    @property
+    def det_db_unclip_ratio(self):
+        return self._model.det_db_unclip_ratio
+
     @det_db_unclip_ratio.setter
     def det_db_unclip_ratio(self, value):
         assert isinstance(
@@ -96,6 +75,10 @@ class DBDetector(FastDeployModel):
         ), "The value to set `det_db_unclip_ratio` must be type of float."
         self._model.det_db_unclip_ratio = value
 
+    @property
+    def det_db_score_mode(self):
+        return self._model.det_db_score_mode
+
     @det_db_score_mode.setter
     def det_db_score_mode(self, value):
         assert isinstance(
@@ -103,6 +86,10 @@ class DBDetector(FastDeployModel):
             str), "The value to set `det_db_score_mode` must be type of str."
         self._model.det_db_score_mode = value
 
+    @property
+    def use_dilation(self):
+        return self._model.use_dilation
+
     @use_dilation.setter
     def use_dilation(self, value):
         assert isinstance(
@@ -110,11 +97,26 @@ class DBDetector(FastDeployModel):
             bool), "The value to set `use_dilation` must be type of bool."
         self._model.use_dilation = value
 
+    @property
+    def max_side_len(self):
+        return self._model.max_side_len
+
+    @max_side_len.setter
+    def max_side_len(self, value):
+        assert isinstance(
+            value, int), "The value to set `max_side_len` must be type of int."
+        self._model.max_side_len = value
+
+    @property
+    def is_scale(self):
+        return self._model.max_wh
+
     @is_scale.setter
     def is_scale(self, value):
         assert isinstance(
             value, bool), "The value to set `is_scale` must be type of bool."
         self._model.is_scale = value
+    '''
 
 
 class Classifier(FastDeployModel):
@@ -139,6 +141,7 @@ class Classifier(FastDeployModel):
                 model_file, params_file, self._runtime_option, model_format)
             assert self.initialized, "Classifier initialize failed."
 
+    '''
     @property
     def cls_thresh(self):
         return self._model.cls_thresh
@@ -170,6 +173,7 @@ class Classifier(FastDeployModel):
             value,
             int), "The value to set `cls_batch_num` must be type of int."
         self._model.cls_batch_num = value
+    '''
 
 
 class Recognizer(FastDeployModel):
@@ -197,6 +201,7 @@ class Recognizer(FastDeployModel):
                 model_format)
             assert self.initialized, "Recognizer initialize failed."
 
+    '''
     @property
     def rec_img_h(self):
         return self._model.rec_img_h
@@ -227,6 +232,7 @@ class Recognizer(FastDeployModel):
             value,
             int), "The value to set `rec_batch_num` must be type of int."
         self._model.rec_batch_num = value
+    '''
 
 
 class PPOCRv3(FastDeployModel):
@@ -253,6 +259,14 @@ class PPOCRv3(FastDeployModel):
         """
         return self.system.predict(input_image)
 
+    def batch_predict(self, images):
+        """Predict a batch of input image
+        :param images: (list of numpy.ndarray) The input image list, each element is a 3-D array with layout HWC, BGR format
+        :return: OCRBatchResult
+        """
+
+        return self.system.batch_predict(images)
+
 
 class PPOCRSystemv3(PPOCRv3):
     def __init__(self, det_model=None, cls_model=None, rec_model=None):
@@ -289,6 +303,14 @@ class PPOCRv2(FastDeployModel):
         """
         return self.system.predict(input_image)
 
+    def batch_predict(self, images):
+        """Predict a batch of input image
+        :param images: (list of numpy.ndarray) The input image list, each element is a 3-D array with layout HWC, BGR format
+        :return: OCRBatchResult
+        """
+
+        return self.system.batch_predict(images)
+
 
 class PPOCRSystemv2(PPOCRv2):
     def __init__(self, det_model=None, cls_model=None, rec_model=None):
@@ -299,3 +321,93 @@ class PPOCRSystemv2(PPOCRv2):
 
     def predict(self, input_image):
         return super(PPOCRSystemv2, self).predict(input_image)
+
+
+class DBDetectorPreprocessor:
+    def __init__(self):
+        """Create a preprocessor for DBDetectorModel
+        """
+        self._preprocessor = C.vision.ocr.DBDetectorPreprocessor()
+
+    def run(self, input_ims):
+        """Preprocess input images for DBDetectorModel
+        :param: input_ims: (list of numpy.ndarray) The input image
+        :return: pair(list of FDTensor, list of std::array<int, 4>)
+        """
+        return self._preprocessor.run(input_ims)
+
+
+class DBDetectorPostprocessor:
+    def __init__(self):
+        """Create a postprocessor for DBDetectorModel
+        """
+        self._postprocessor = C.vision.ocr.DBDetectorPostprocessor()
+
+    def run(self, runtime_results, batch_det_img_info):
+        """Postprocess the runtime results for DBDetectorModel
+        :param: runtime_results: (list of FDTensor or list of pyArray)The output FDTensor results from runtime
+        :param: batch_det_img_info: (list of std::array<int, 4>)The output of det_preprocessor
+        :return: list of Result(If the runtime_results is predict by batched samples, the length of this list equals to the batch size)
+        """
+        return self._postprocessor.run(runtime_results, batch_det_img_info)
+
+
+class RecognizerPreprocessor:
+    def __init__(self):
+        """Create a preprocessor for RecognizerModel
+        """
+        self._preprocessor = C.vision.ocr.RecognizerPreprocessor()
+
+    def run(self, input_ims):
+        """Preprocess input images for RecognizerModel
+        :param: input_ims: (list of numpy.ndarray)The input image
+        :return: list of FDTensor
+        """
+        return self._preprocessor.run(input_ims)
+
+
+class RecognizerPostprocessor:
+    def __init__(self, label_path):
+        """Create a postprocessor for RecognizerModel
+        :param label_path: (str)Path of label file
+        """
+        self._postprocessor = C.vision.ocr.RecognizerPostprocessor(label_path)
+
+    def run(self, runtime_results):
+        """Postprocess the runtime results for RecognizerModel
+        :param: runtime_results: (list of FDTensor or list of pyArray)The output FDTensor results from runtime
+        :return: list of Result(If the runtime_results is predict by batched samples, the length of this list equals to the batch size)
+        """
+        return self._postprocessor.run(runtime_results)
+
+
+class ClassifierPreprocessor:
+    def __init__(self):
+        """Create a preprocessor for ClassifierModel
+        """
+        self._preprocessor = C.vision.ocr.ClassifierPreprocessor()
+
+    def run(self, input_ims):
+        """Preprocess input images for ClassifierModel
+        :param: input_ims: (list of numpy.ndarray)The input image
+        :return: list of FDTensor
+        """
+        return self._preprocessor.run(input_ims)
+
+
+class ClassifierPostprocessor:
+    def __init__(self):
+        """Create a postprocessor for ClassifierModel
+        """
+        self._postprocessor = C.vision.ocr.ClassifierPostprocessor()
+
+    def run(self, runtime_results):
+        """Postprocess the runtime results for ClassifierModel
+        :param: runtime_results: (list of FDTensor or list of pyArray)The output FDTensor results from runtime
+        :return: list of Result(If the runtime_results is predict by batched samples, the length of this list equals to the batch size)
+        """
+        return self._postprocessor.run(runtime_results)
+
+
+def sort_boxes(boxes):
+    return C.vision.ocr.sort_boxes(boxes)