update ocr server and mobile to 1.1.1 to fix recognize space bug

modules/image/text_recognition/chinese_ocr_db_crnn_mobile/README.md

@@ -152,3 +152,7 @@ pyclipper
 * 1.1.0
 
   使用超轻量级的三阶段模型（文本框检测-角度分类-文字识别）识别图片文字。
+
+* 1.1.1
+
+   支持文本中空格识别。

modules/image/text_recognition/chinese_ocr_db_crnn_mobile/character.py

@@ -17,7 +17,10 @@ import string
 
 
 class CharacterOps(object):
-    """ Convert between text-label and text-index """
+    """ Convert between text-label and text-index
+    Args:
+        config: config from yaml file
+    """
 
     def __init__(self, config):
         self.character_type = config['character_type']
@@ -26,6 +29,7 @@ class CharacterOps(object):
         if self.character_type == "en":
             self.character_str = "0123456789abcdefghijklmnopqrstuvwxyz"
             dict_character = list(self.character_str)
+        # use the custom dictionary
         elif self.character_type == "ch":
             character_dict_path = config['character_dict_path']
             add_space = False
@@ -50,10 +54,12 @@ class CharacterOps(object):
             "Nonsupport type of the character: {}".format(self.character_str)
         self.beg_str = "sos"
         self.end_str = "eos"
+        # add start and end str for attention
         if self.loss_type == "attention":
             dict_character = [self.beg_str, self.end_str] + dict_character
         elif self.loss_type == "srn":
             dict_character = dict_character + [self.beg_str, self.end_str]
+        # create char dict
         self.dict = {}
         for i, char in enumerate(dict_character):
             self.dict[char] = i
@@ -122,6 +128,21 @@ class CharacterOps(object):
 
 
 def cal_predicts_accuracy(char_ops, preds, preds_lod, labels, labels_lod, is_remove_duplicate=False):
+    """
+    Calculate prediction accuracy
+    Args:
+        char_ops: CharacterOps
+        preds: preds result,text index
+        preds_lod: lod tensor of preds
+        labels: label of input image, text index
+        labels_lod:  lod tensor of label
+        is_remove_duplicate: Whether to remove duplicate characters,
+                                 The default is False
+    Return:
+        acc: The accuracy of test set
+        acc_num: The correct number of samples predicted
+        img_num: The total sample number of the test set
+    """
     acc_num = 0
     img_num = 0
     for ino in range(len(labels_lod) - 1):

modules/image/text_recognition/chinese_ocr_db_crnn_mobile/module.py

@@ -21,7 +21,7 @@ from chinese_ocr_db_crnn_mobile.utils import base64_to_cv2, draw_ocr, get_image_
 
 @moduleinfo(
     name="chinese_ocr_db_crnn_mobile",
-    version="1.1.0",
+    version="1.1.1",
     summary="The module can recognize the chinese texts in an image. Firstly, it will detect the text box positions \
         based on the differentiable_binarization_chn module. Then it classifies the text angle and recognizes the chinese texts. ",
     author="paddle-dev",
@@ -100,7 +100,7 @@ class ChineseOCRDBCRNN(hub.Module):
         """
         if not self._text_detector_module:
             self._text_detector_module = hub.Module(
-                name='chinese_text_detection_db_mobile', enable_mkldnn=self.enable_mkldnn, version='1.0.3')
+                name='chinese_text_detection_db_mobile', enable_mkldnn=self.enable_mkldnn, version='1.0.4')
         return self._text_detector_module
 
     def read_images(self, paths=[]):

modules/image/text_recognition/chinese_ocr_db_crnn_server/README.md

@@ -147,3 +147,7 @@ pyclipper
 * 1.1.0
 
    使用三阶段模型（文本框检测-角度分类-文字识别）识别图片文字。
+
+* 1.1.1
+
+   支持文本中空格识别。

modules/image/text_recognition/chinese_ocr_db_crnn_server/character.py

@@ -17,7 +17,10 @@ import string
 
 
 class CharacterOps(object):
-    """ Convert between text-label and text-index """
+    """ Convert between text-label and text-index
+    Args:
+        config: config from yaml file
+    """
 
     def __init__(self, config):
         self.character_type = config['character_type']
@@ -26,6 +29,7 @@ class CharacterOps(object):
         if self.character_type == "en":
             self.character_str = "0123456789abcdefghijklmnopqrstuvwxyz"
             dict_character = list(self.character_str)
+        # use the custom dictionary
         elif self.character_type == "ch":
             character_dict_path = config['character_dict_path']
             add_space = False
@@ -50,10 +54,12 @@ class CharacterOps(object):
             "Nonsupport type of the character: {}".format(self.character_str)
         self.beg_str = "sos"
         self.end_str = "eos"
+        # add start and end str for attention
         if self.loss_type == "attention":
             dict_character = [self.beg_str, self.end_str] + dict_character
         elif self.loss_type == "srn":
             dict_character = dict_character + [self.beg_str, self.end_str]
+        # create char dict
         self.dict = {}
         for i, char in enumerate(dict_character):
             self.dict[char] = i
@@ -122,6 +128,21 @@ class CharacterOps(object):
 
 
 def cal_predicts_accuracy(char_ops, preds, preds_lod, labels, labels_lod, is_remove_duplicate=False):
+    """
+    Calculate prediction accuracy
+    Args:
+        char_ops: CharacterOps
+        preds: preds result,text index
+        preds_lod: lod tensor of preds
+        labels: label of input image, text index
+        labels_lod:  lod tensor of label
+        is_remove_duplicate: Whether to remove duplicate characters,
+                                 The default is False
+    Return:
+        acc: The accuracy of test set
+        acc_num: The correct number of samples predicted
+        img_num: The total sample number of the test set
+    """
     acc_num = 0
     img_num = 0
     for ino in range(len(labels_lod) - 1):

modules/image/text_recognition/chinese_ocr_db_crnn_server/module.py

@@ -25,7 +25,7 @@ from chinese_ocr_db_crnn_server.utils import base64_to_cv2, draw_ocr, get_image_
 
 @moduleinfo(
     name="chinese_ocr_db_crnn_server",
-    version="1.1.0",
+    version="1.1.1",
     summary=
     "The module can recognize the chinese texts in an image. Firstly, it will detect the text box positions based on the differentiable_binarization_chn module. Then it recognizes the chinese texts. ",
     author="paddle-dev",