使用img2table+EasyOCR识别泰米尔语文本时触发RuntimeError

问题场景 #

我正在用img2table提取含泰米尔语的表格文本，因EasyOCR支持多语言识别，故将二者结合使用，但运行以下代码时出现错误：

img = cv2.imread("./3.jpg")
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
languages = ['en', 'ta']
kw = {"gpu": False}
easyocr = EasyOCR(lang = languages,kw=kw)

错误信息 #

RuntimeError                              Traceback (most recent call last)
Cell In[52], line 1
----> 1 easyocr = EasyOCR(lang = languages,kw=kw)

File c:\Users\XXXXX\AppData\Local\Programs\Python\Python311\Lib\site-packages\img2table\ocr\easyocr.py:38, in EasyOCR.__init__(self, lang, kw)
     35 kw["lang_list"] = self.lang
     36 kw["verbose"] = kw.get("verbose") or False
---> 38 self.reader = Reader(**kw)

File c:\Users\XXXXX\AppData\Local\Programs\Python\Python311\Lib\site-packages\easyocr\easyocr.py:231, in Reader.__init__(self, lang_list, gpu, model_storage_directory, user_network_directory, detect_network, recog_network, download_enabled, detector, recognizer, verbose, quantize, cudnn_benchmark)
    229 else:
    230     network_params = recog_config['network_params']
---> 231 self.recognizer, self.converter = get_recognizer(recog_network, network_params,\
    232                                              self.character, separator_list,\
    233                                              dict_list, model_path, device = self.device, quantize=quantize)

File c:\Users\XXXXX\AppData\Local\Programs\Python\Python311\Lib\site-packages\easyocr\recognition.py:174, in get_recognizer(recog_network, network_params, character, separator_list, dict_list, model_path, device, quantize)
    172     new_key = key[7:]
    173     new_state_dict[new_key] = value
---> 174 model.load_state_dict(new_state_dict)
    175 if quantize:
    176     try:

File c:\Users\XXXXX\AppData\Local\Programs\Python\Python311\Lib\site-packages\torch\nn\modules\module.py:2152, in Module.load_state_dict(self, state_dict, strict, assign)
   2147         error_msgs.insert(
   2148             0, 'Missing key(s) in state_dict: {}. '.format(
   2149                 ', '.join(f'"{k}"' for k in missing_keys)))
   2151 if len(error_msgs) > 0:
-> 2152     raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
   2153                        self.__class__.__name__, "\n\t".join(error_msgs)))
   2154 return _IncompatibleKeys(missing_keys, unexpected_keys)

RuntimeError: Error(s) in loading state_dict for Model:
    size mismatch for Prediction.weight: copying a param with shape torch.Size([143, 512]) from checkpoint, the shape in current model is torch.Size([127, 512]).
    size mismatch for Prediction.bias: copying a param with shape torch.Size([143]) from checkpoint, the shape in current model is torch.Size([127]).

环境版本 #

• Python 3.11.8
• easyocr 1.7.1
• img2table 1.2.8

解决方案 #

这个错误源于EasyOCR预训练模型与当前字符集维度不匹配，泰米尔语模型的字符数和加载的模型参数维度不一致，以下是几种可行的解决方法：

1. 清理缓存并重新下载模型 #

EasyOCR会将预训练模型缓存到本地，缓存的旧版本模型可能与当前easyocr版本不兼容：

• 找到模型缓存目录：C:\Users\你的用户名\.EasyOCR\model
• 删除该目录下所有文件和文件夹
• 重新运行代码，EasyOCR会自动下载对应版本的多语言模型

2. 显式指定兼容的识别网络 #

初始化时指定recog_network参数，选择支持泰米尔语的网络版本：

img = cv2.imread("./3.jpg")
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
languages = ['en', 'ta']
kw = {"gpu": False, "recog_network": 'english_g2'}
easyocr = EasyOCR(lang = languages, kw=kw)

也可尝试使用多语言专用网络'multi_lang'（需确认版本支持）。

3. 调整依赖版本 #

当前img2table 1.2.8与easyocr 1.7.1存在兼容性问题，可降低easyocr版本：

pip install easyocr==1.6.2

4. 绕过img2table封装，直接初始化EasyOCR Reader #

直接创建EasyOCR的Reader实例，再传给img2table使用：

import cv2
from easyocr import Reader
from img2table.ocr import EasyOCR as Img2TableEasyOCR

img = cv2.imread("./3.jpg")
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)

# 直接初始化EasyOCR Reader
reader = Reader(lang_list=['en', 'ta'], gpu=False)
# 传给img2table的EasyOCR封装
easyocr = Img2TableEasyOCR(reader=reader)

# 后续执行img2table表格提取逻辑

内容的提问来源于stack exchange，提问作者NISHITH