你好，这两个文件的生成能开源吗？或者有没有什么可以指点的地方，比如您参考的是哪个开源算法？

水印的Laplace模板用于精确定位

wm_filter = [np.load('data/watermark1_laplace.npy'), np.load('data/watermark2_laplace.npy')]

二值图模板用于生成蒙版以进行后续去水印操作

wm_img = [np.load('data/watermark1_solid.npy'), np.load('data/watermark2_solid.npy')]

(chineseocr) quh@quh-Z370P-D3:~/pythonwork/tr/CCF-OCR$ python main.py
read Chinese alphabet from file
Traceback (most recent call last):
File "main.py", line 4, in
import recognition
File "/home/quh/pythonwork/tr/CCF-OCR/recognition.py", line 14, in
alphabet = f.read().replace('\n', '')
File "/home/quh/.conda/envs/chineseocr/lib/python3.6/codecs.py", line 321, in decode
(result, consumed) = self._buffer_decode(data, self.errors, final)
UnicodeDecodeError: 'utf-8' codec can't decode byte 0xb6 in position 23: invalid start byte

RuntimeError: Error(s) in loading state_dict for chsNet:
size mismatch for cnn.conv6.weight: copying a param with shape torch.Size([6857, 512, 1, 1]) from checkpoint, the shape in current model is torch.Size([6937, 512, 1, 1]).
size mismatch for cnn.conv6.bias: copying a param with shape torch.Size([6857]) from checkpoint, the shape in current model is torch.Size([6937]).

最近在学习OCR识别模型，想复现下实验结果，可否详细讲下使用的训练集及训练过程？谢谢您～