Merge pull request #50 from kotaro-kinoshita/fix/split-voiced-point-code

add prediction text normalize

src/yomitoku/text_recognizer.py

@@ -2,6 +2,7 @@
 
 import numpy as np
 import torch
+import unicodedata
 from pydantic import conlist
 
 from .base import BaseModelCatalog, BaseModule, BaseSchema
@@ -72,6 +73,8 @@ def preprocess(self, img, polygons):
 
     def postprocess(self, p, points):
         pred, score = self.tokenizer.decode(p)
+        pred = [unicodedata.normalize("NFKC", x) for x in pred]
+
         directions = []
         for point in points:
             point = np.array(point)