Fix Diff & Improve OCR readme (#184)

* Add PaddleOCR Support * Add PaddleOCR Support * Add PaddleOCRv3 Support * Add PaddleOCRv3 Support * Update README.md * Update README.md * Update README.md * Update README.md * Add PaddleOCRv3 Support * Add PaddleOCRv3 Supports * Add PaddleOCRv3 Suport * Fix Rec diff * Remove useless functions * Remove useless comments * Add PaddleOCRv2 Support * Add PaddleOCRv3 & PaddleOCRv2 Support * remove useless parameters * Add utils of sorting det boxes * Fix code naming convention * Fix code naming convention * Fix code naming convention * Fix bug in the Classify process * Imporve OCR Readme * Fix diff in Cls model * Update Model Download Link in Readme
2025-12-24 13:28:13 +08:00 · 2022-09-05 09:36:06 +08:00
parent 5f880dcdd0
commit fad82b4e5d
9 changed files with 46 additions and 53 deletions
--- a/csrc/fastdeploy/vision/ocr/ppocr/classifier.cc
+++ b/csrc/fastdeploy/vision/ocr/ppocr/classifier.cc
@@ -46,7 +46,7 @@ bool Classifier::Initialize() {
  cls_thresh = 0.9;
  cls_image_shape = {3, 48, 192};
  cls_batch_num = 1;
-  mean = {0.485f, 0.456f, 0.406f};
+  mean = {0.5f, 0.5f, 0.5f};
  scale = {0.5f, 0.5f, 0.5f};
  is_scale = true;

--- a/csrc/fastdeploy/vision/ocr/ppocr/dbdetector.cc
+++ b/csrc/fastdeploy/vision/ocr/ppocr/dbdetector.cc
@@ -144,10 +144,11 @@ bool DBDetector::Postprocess(

  std::vector<std::vector<std::vector<int>>> boxes;

-  post_processor_.BoxesFromBitmap(pred_map, &boxes, bit_map, det_db_box_thresh,
-                                  det_db_unclip_ratio, det_db_score_mode);
+  boxes =
+      post_processor_.BoxesFromBitmap(pred_map, bit_map, det_db_box_thresh,
+                                      det_db_unclip_ratio, det_db_score_mode);

-  post_processor_.FilterTagDetRes(&boxes, ratio_h, ratio_w, im_info);
+  boxes = post_processor_.FilterTagDetRes(boxes, ratio_h, ratio_w, im_info);

  // boxes to boxes_result
  for (int i = 0; i < boxes.size(); i++) {
--- a/csrc/fastdeploy/vision/ocr/ppocr/ppocr_system_v2.cc
+++ b/csrc/fastdeploy/vision/ocr/ppocr/ppocr_system_v2.cc
@@ -36,7 +36,7 @@ void PPOCRSystemv2::Detect(cv::Mat* img,
 void PPOCRSystemv2::Recognize(cv::Mat* img,
                              fastdeploy::vision::OCRResult* result) {
  std::tuple<std::string, float> rec_result;
-
+  this->recognizer->rec_image_shape[1] = 32;
  this->recognizer->Predict(img, &rec_result);

  result->text.push_back(std::get<0>(rec_result));
--- a/csrc/fastdeploy/vision/ocr/ppocr/ppocr_system_v3.cc
+++ b/csrc/fastdeploy/vision/ocr/ppocr/ppocr_system_v3.cc
@@ -94,7 +94,6 @@ bool PPOCRSystemv3::Predict(cv::Mat* img,
      for (int i = 0; i < img_list.size(); i++) {
        if ((result->cls_labels)[i] % 2 == 1 &&
            (result->cls_scores)[i] > this->classifier->cls_thresh) {
-          std::cout << "Rotate this image " << std::endl;
          cv::rotate(img_list[i], img_list[i], 1);
        }
      }
--- a/csrc/fastdeploy/vision/ocr/ppocr/utils/ocr_postprocess_op.cc
+++ b/csrc/fastdeploy/vision/ocr/ppocr/utils/ocr_postprocess_op.cc
@@ -250,9 +250,8 @@ float PostProcessor::BoxScoreFast(std::vector<std::vector<float>> box_array,

 //这个应该是DB（差分二值化）相关的内容，方法从 Bitmap 图中获取检测框
 //涉及到box_thresh（低于这个阈值的boxs不予显示）和det_db_unclip_ratio（文本框扩张的系数，关系到文本框的大小）
-void PostProcessor::BoxesFromBitmap(
-    const cv::Mat pred, std::vector<std::vector<std::vector<int>>> *boxes,
-    const cv::Mat bitmap, const float &box_thresh,
+std::vector<std::vector<std::vector<int>>> PostProcessor::BoxesFromBitmap(
+    const cv::Mat pred, const cv::Mat bitmap, const float &box_thresh,
    const float &det_db_unclip_ratio, const std::string &det_db_score_mode) {
  const int min_size = 3;
  const int max_candidates = 1000;
@@ -269,6 +268,8 @@ void PostProcessor::BoxesFromBitmap(
  int num_contours =
      contours.size() >= max_candidates ? max_candidates : contours.size();

+  std::vector<std::vector<std::vector<int>>> boxes;
+
  for (int _i = 0; _i < num_contours; _i++) {
    if (contours[_i].size() <= 2) {
      continue;
@@ -318,45 +319,42 @@ void PostProcessor::BoxesFromBitmap(
              0, float(dest_height)))};
      intcliparray.push_back(a);
    }
-    boxes->push_back(intcliparray);
+    boxes.push_back(intcliparray);

  }  // end for
-  // return true;
+  return boxes;
 }

-//方法根据识别结果获取目标框位置
-void PostProcessor::FilterTagDetRes(
-    std::vector<std::vector<std::vector<int>>> *boxes, const float ratio_h,
-    const float ratio_w,
-    const std::map<std::string, std::array<float, 2>> &im_info) {
+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];

-  for (int n = 0; n < boxes->size(); n++) {
-    (*boxes)[n] = OrderPointsClockwise((*boxes)[n]);
-    for (int m = 0; m < (*boxes)[0].size(); m++) {
-      (*boxes)[n][m][0] /= ratio_w;
-      (*boxes)[n][m][1] /= ratio_h;
+  std::vector<std::vector<std::vector<int>>> root_points;
+  for (int n = 0; n < boxes.size(); n++) {
+    boxes[n] = OrderPointsClockwise(boxes[n]);
+    for (int m = 0; m < boxes[0].size(); m++) {
+      boxes[n][m][0] /= ratio_w;
+      boxes[n][m][1] /= ratio_h;

-      (*boxes)[n][m][0] = int(_min(_max((*boxes)[n][m][0], 0), oriimg_w - 1));
-      (*boxes)[n][m][1] = int(_min(_max((*boxes)[n][m][1], 0), oriimg_h - 1));
+      boxes[n][m][0] = int(_min(_max(boxes[n][m][0], 0), oriimg_w - 1));
+      boxes[n][m][1] = int(_min(_max(boxes[n][m][1], 0), oriimg_h - 1));
    }
  }

-  //此时已经拿到所有的点. 再进行下面的筛选
-  for (int n = (*boxes).size() - 1; n >= 0; n--) {
+  for (int n = 0; n < boxes.size(); n++) {
    int rect_width, rect_height;
-    rect_width = int(sqrt(pow((*boxes)[n][0][0] - (*boxes)[n][1][0], 2) +
-                          pow((*boxes)[n][0][1] - (*boxes)[n][1][1], 2)));
-    rect_height = int(sqrt(pow((*boxes)[n][0][0] - (*boxes)[n][3][0], 2) +
-                           pow((*boxes)[n][0][1] - (*boxes)[n][3][1], 2)));
-
-    //小于4的删除掉. erase配合逆序遍历.
-    if (rect_width <= 4 || rect_height <= 4) {
-      boxes->erase(boxes->begin() + n);
-    }
+    rect_width = int(sqrt(pow(boxes[n][0][0] - boxes[n][1][0], 2) +
+                          pow(boxes[n][0][1] - boxes[n][1][1], 2)));
+    rect_height = int(sqrt(pow(boxes[n][0][0] - boxes[n][3][0], 2) +
+                           pow(boxes[n][0][1] - boxes[n][3][1], 2)));
+    if (rect_width <= 4 || rect_height <= 4) continue;
+    root_points.push_back(boxes[n]);
  }
+  return root_points;
 }
+
 }  // namespace ocr
 }  // namespace vision
 }  // namespace fastdeploy
--- a/csrc/fastdeploy/vision/ocr/ppocr/utils/ocr_postprocess_op.h
+++ b/csrc/fastdeploy/vision/ocr/ppocr/utils/ocr_postprocess_op.h
@@ -52,15 +52,13 @@ class PostProcessor {
  float BoxScoreFast(std::vector<std::vector<float>> box_array, cv::Mat pred);
  float PolygonScoreAcc(std::vector<cv::Point> contour, cv::Mat pred);

-  void BoxesFromBitmap(const cv::Mat pred,
-                       std::vector<std::vector<std::vector<int>>> *boxes,
-                       const cv::Mat bitmap, const float &box_thresh,
-                       const float &det_db_unclip_ratio,
-                       const std::string &det_db_score_mode);
+  std::vector<std::vector<std::vector<int>>> BoxesFromBitmap(
+      const cv::Mat pred, const cv::Mat bitmap, const float &box_thresh,
+      const float &det_db_unclip_ratio, const std::string &det_db_score_mode);

-  void FilterTagDetRes(
-      std::vector<std::vector<std::vector<int>>> *boxes, const float ratio_h,
-      const float ratio_w,
+  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);

 private: