Merge pull request #2606 from littletomatodonkey/dyg/add_db_post

add new db post score calculation method

deploy/cpp_infer/include/postprocess_op.h

@@ -51,6 +51,7 @@ public:
                                                float &ssid);
 
   float BoxScoreFast(std::vector<std::vector<float>> box_array, cv::Mat pred);
+  float PolygonScoreAcc(std::vector<cv::Point> contour, cv::Mat pred);
 
   std::vector<std::vector<std::vector<int>>>
   BoxesFromBitmap(const cv::Mat pred, const cv::Mat bitmap,

deploy/cpp_infer/src/postprocess_op.cpp

@@ -159,6 +159,39 @@ std::vector<std::vector<float>> PostProcessor::GetMiniBoxes(cv::RotatedRect box,
   return array;
 }
 
+float PostProcessor::PolygonScoreAcc(std::vector<cv::Point> contour,
+		                  cv::Mat pred){
+  int width = pred.cols;
+  int height = pred.rows;
+  std::vector<float> box_x;
+  std::vector<float> box_y;
+  for(int i=0; i<contour.size(); ++i){
+    box_x.push_back(contour[i].x);
+    box_y.push_back(contour[i].y);
+  }
+
+  int xmin = clamp(int(std::floor(*(std::min_element(box_x.begin(), box_x.end())))), 0, width - 1);
+  int xmax = clamp(int(std::ceil(*(std::max_element(box_x.begin(), box_x.end())))), 0, width - 1);
+  int ymin = clamp(int(std::floor(*(std::min_element(box_y.begin(), box_y.end())))), 0, height - 1);
+  int ymax = clamp(int(std::ceil(*(std::max_element(box_y.begin(), box_y.end())))), 0, height - 1);
+
+  cv::Mat mask;
+  mask = cv::Mat::zeros(ymax - ymin + 1, xmax - xmin + 1, CV_8UC1);
+
+  cv::Point rook_point[contour.size()];
+  for(int i=0; i<contour.size(); ++i){
+    rook_point[i] = cv::Point(int(box_x[i]) - xmin, int(box_y[i]) - ymin);
+  }
+  const cv::Point *ppt[1] = {rook_point};
+  int npt[] = {int(contour.size())};
+  cv::fillPoly(mask, ppt, npt, 1, cv::Scalar(1));
+
+  cv::Mat croppedImg;
+  pred(cv::Rect(xmin, ymin, xmax - xmin + 1, ymax - ymin + 1)).copyTo(croppedImg);
+  float score = cv::mean(croppedImg, mask)[0];
+  return score;
+}
+
 float PostProcessor::BoxScoreFast(std::vector<std::vector<float>> box_array,
                                   cv::Mat pred) {
   auto array = box_array;
@@ -235,6 +268,8 @@ PostProcessor::BoxesFromBitmap(const cv::Mat pred, const cv::Mat bitmap,
 
     float score;
     score = BoxScoreFast(array, pred);
+    /* compute using polygon*/ 
+    // score = PolygonScoreAcc(contours[_i], pred);
     if (score < box_thresh)
       continue;
 

ppocr/postprocess/db_postprocess.py

@@ -34,12 +34,18 @@ class DBPostProcess(object):
                  max_candidates=1000,
                  unclip_ratio=2.0,
                  use_dilation=False,
+                 score_mode="fast",
                  **kwargs):
         self.thresh = thresh
         self.box_thresh = box_thresh
         self.max_candidates = max_candidates
         self.unclip_ratio = unclip_ratio
         self.min_size = 3
+        self.score_mode = score_mode
+        assert score_mode in [
+            "slow", "fast"
+        ], "Score mode must be in [slow, fast] but got: {}".format(score_mode)
+
         self.dilation_kernel = None if not use_dilation else np.array(
             [[1, 1], [1, 1]])
 
@@ -69,7 +75,10 @@ class DBPostProcess(object):
             if sside < self.min_size:
                 continue
             points = np.array(points)
+            if self.score_mode == "fast":
                 score = self.box_score_fast(pred, points.reshape(-1, 2))
+            else:
+                score = self.box_score_slow(pred, contour)
             if self.box_thresh > score:
                 continue
 
@@ -120,6 +129,9 @@ class DBPostProcess(object):
         return box, min(bounding_box[1])
 
     def box_score_fast(self, bitmap, _box):
+        '''
+        box_score_fast: use bbox mean score as the mean score
+        '''
         h, w = bitmap.shape[:2]
         box = _box.copy()
         xmin = np.clip(np.floor(box[:, 0].min()).astype(np.int), 0, w - 1)
@@ -133,6 +145,27 @@ class DBPostProcess(object):
         cv2.fillPoly(mask, box.reshape(1, -1, 2).astype(np.int32), 1)
         return cv2.mean(bitmap[ymin:ymax + 1, xmin:xmax + 1], mask)[0]
 
+    def box_score_slow(self, bitmap, contour):
+        '''
+        box_score_slow: use polyon mean score as the mean score
+        '''
+        h, w = bitmap.shape[:2]
+        contour = contour.copy()
+        contour = np.reshape(contour, (-1, 2))
+
+        xmin = np.clip(np.min(contour[:, 0]), 0, w - 1)
+        xmax = np.clip(np.max(contour[:, 0]), 0, w - 1)
+        ymin = np.clip(np.min(contour[:, 1]), 0, h - 1)
+        ymax = np.clip(np.max(contour[:, 1]), 0, h - 1)
+
+        mask = np.zeros((ymax - ymin + 1, xmax - xmin + 1), dtype=np.uint8)
+
+        contour[:, 0] = contour[:, 0] - xmin
+        contour[:, 1] = contour[:, 1] - ymin
+
+        cv2.fillPoly(mask, contour.reshape(1, -1, 2).astype(np.int32), 1)
+        return cv2.mean(bitmap[ymin:ymax + 1, xmin:xmax + 1], mask)[0]
+
     def __call__(self, outs_dict, shape_list):
         pred = outs_dict['maps']
         if isinstance(pred, paddle.Tensor):

tools/infer/predict_det.py

@@ -62,6 +62,7 @@ class TextDetector(object):
             postprocess_params["max_candidates"] = 1000
             postprocess_params["unclip_ratio"] = args.det_db_unclip_ratio
             postprocess_params["use_dilation"] = args.use_dilation
+            postprocess_params["score_mode"] = args.det_db_score_mode
         elif self.det_algorithm == "EAST":
             postprocess_params['name'] = 'EASTPostProcess'
             postprocess_params["score_thresh"] = args.det_east_score_thresh

tools/infer/utility.py

@@ -48,6 +48,7 @@ def parse_args():
     parser.add_argument("--det_db_unclip_ratio", type=float, default=1.6)
     parser.add_argument("--max_batch_size", type=int, default=10)
     parser.add_argument("--use_dilation", type=bool, default=False)
+    parser.add_argument("--det_db_score_mode", type=str, default="fast")
     # EAST parmas
     parser.add_argument("--det_east_score_thresh", type=float, default=0.8)
     parser.add_argument("--det_east_cover_thresh", type=float, default=0.1)