识别文本框时，对文本框按宽高比进行排序。

tools/infer/predict_rec.py

@@ -13,9 +13,9 @@
 # limitations under the License.
 import os
 import sys
-__dir__ = os.path.dirname(__file__)
+__dir__ = os.path.dirname(os.path.abspath(__file__))
 sys.path.append(__dir__)
-sys.path.append(os.path.join(__dir__, '../..'))
+sys.path.append(os.path.abspath(os.path.join(__dir__, '../..')))
 
 import tools.infer.utility as utility
 from ppocr.utils.utility import initial_logger
@@ -33,14 +33,12 @@ class TextRecognizer(object):
     def __init__(self, args):
         self.predictor, self.input_tensor, self.output_tensors =\
             utility.create_predictor(args, mode="rec")
-        image_shape = [int(v) for v in args.rec_image_shape.split(",")]
-        self.rec_image_shape = image_shape
+        self.rec_image_shape = [int(v) for v in args.rec_image_shape.split(",")]
         self.character_type = args.rec_char_type
         self.rec_batch_num = args.rec_batch_num
         self.rec_algorithm = args.rec_algorithm
-        char_ops_params = {}
-        char_ops_params["character_type"] = args.rec_char_type
-        char_ops_params["character_dict_path"] = args.rec_char_dict_path
+        char_ops_params = {"character_type": args.rec_char_type,
+                           "character_dict_path": args.rec_char_dict_path}
         if self.rec_algorithm != "RARE":
             char_ops_params['loss_type'] = 'ctc'
             self.loss_type = 'ctc'
@@ -51,16 +49,11 @@ class TextRecognizer(object):
 
     def resize_norm_img(self, img, max_wh_ratio):
         imgC, imgH, imgW = self.rec_image_shape
-        if self.character_type == "ch":
-            imgW = int(32 * max_wh_ratio)
-        h = img.shape[0]
-        w = img.shape[1]
-        ratio = w / float(h)
-        if math.ceil(imgH * ratio) > imgW:
-            resized_w = imgW
-        else:
-            resized_w = int(math.ceil(imgH * ratio))
-        resized_image = cv2.resize(img, (resized_w, imgH))
+        assert imgC == img.shape[2]
+        imgW = int(math.ceil(32 * max_wh_ratio))
+        h, w = img.shape[:2]
+        resized_w = int(math.ceil(imgH * w / float(h)))
+        resized_image = cv2.resize(img, (resized_w, imgH), interpolation=cv2.INTER_CUBIC)
         resized_image = resized_image.astype('float32')
         resized_image = resized_image.transpose((2, 0, 1)) / 255
         resized_image -= 0.5
@@ -71,7 +64,15 @@ class TextRecognizer(object):
 
     def __call__(self, img_list):
         img_num = len(img_list)
-        rec_res = []
+        # 统计所有文本条的宽高比
+        width_list = []
+        for img in img_list:
+            width_list.append(img.shape[1] / float(img.shape[0]))
+        # 对于文本框比较多且长短差异较大的情况下，通过排序再组合batch可以明显加速识别
+        indices = np.argsort(np.array(width_list))
+
+        # rec_res = []
+        rec_res = [['', 0.0]] * img_num
         batch_num = self.rec_batch_num
         predict_time = 0
         for beg_img_no in range(0, img_num, batch_num):
@@ -80,10 +81,12 @@ class TextRecognizer(object):
             max_wh_ratio = 0
             for ino in range(beg_img_no, end_img_no):
                 h, w = img_list[ino].shape[0:2]
+                # h, w = img_list[indices[ino]].shape[0:2]
                 wh_ratio = w * 1.0 / h
                 max_wh_ratio = max(max_wh_ratio, wh_ratio)
             for ino in range(beg_img_no, end_img_no):
                 norm_img = self.resize_norm_img(img_list[ino], max_wh_ratio)
+                # norm_img = self.resize_norm_img(img_list[indices[ino]], max_wh_ratio)
                 norm_img = norm_img[np.newaxis, :]
                 norm_img_batch.append(norm_img)
             norm_img_batch = np.concatenate(norm_img_batch)
@@ -111,7 +114,8 @@ class TextRecognizer(object):
                     blank = probs.shape[1]
                     valid_ind = np.where(ind != (blank - 1))[0]
                     score = np.mean(probs[valid_ind, ind[valid_ind]])
-                    rec_res.append([preds_text, score])
+                    # rec_res.append([preds_text, score])
+                    rec_res[indices[beg_img_no + rno]] = [preds_text, score]
             else:
                 rec_idx_batch = self.output_tensors[0].copy_to_cpu()
                 predict_batch = self.output_tensors[1].copy_to_cpu()
@@ -126,19 +130,19 @@ class TextRecognizer(object):
                         preds = rec_idx_batch[rno, 1:end_pos[1]]
                         score = np.mean(predict_batch[rno, 1:end_pos[1]])
                     preds_text = self.char_ops.decode(preds)
-                    rec_res.append([preds_text, score])
+                    # rec_res.append([preds_text, score])
+                    rec_res[indices[beg_img_no + rno]] = [preds_text, score]
 
         return rec_res, predict_time
 
 
-if __name__ == "__main__":
-    args = utility.parse_args()
+def main(args):
     image_file_list = get_image_file_list(args.image_dir)
     text_recognizer = TextRecognizer(args)
     valid_image_file_list = []
     img_list = []
     for image_file in image_file_list:
-        img = cv2.imread(image_file)
+        img = cv2.imread(image_file, cv2.IMREAD_COLOR)
         if img is None:
             logger.info("error in loading image:{}".format(image_file))
             continue
@@ -159,3 +163,7 @@ if __name__ == "__main__":
         print("Predicts of %s:%s" % (valid_image_file_list[ino], rec_res[ino]))
     print("Total predict time for %d images:%.3f" %
           (len(img_list), predict_time))
+
+
+if __name__ == "__main__":
+    main(utility.parse_args())

tools/infer/predict_system.py

@@ -75,6 +75,7 @@ class TextSystem(object):
     def __call__(self, img):
         ori_im = img.copy()
         dt_boxes, elapse = self.text_detector(img)
+        print("dt_boxes num : {}, elapse : {}".format(len(dt_boxes), elapse))
         if dt_boxes is None:
             return None, None
         img_crop_list = []
@@ -86,6 +87,7 @@ class TextSystem(object):
             img_crop = self.get_rotate_crop_image(ori_im, tmp_box)
             img_crop_list.append(img_crop)
         rec_res, elapse = self.text_recognizer(img_crop_list)
+        print("rec_res num  : {}, elapse : {}".format(len(rec_res), elapse))
         # self.print_draw_crop_rec_res(img_crop_list, rec_res)
         return dt_boxes, rec_res