In the crnn+ctc recognition model, the lmdb script only converts part of the data set

 (For example: there are a total of 68114 images in the dataset, only 64564 images in lmdb format can be converted)

Problem point: The lmdb script can only convert part of the image text correctly, so that the dictionary mapping error occurs during crnn model training.

Reason: It is related to the format of the self-made script, for example: LLDCL/7×7_1.ttf and fonts/LLDCL/12px.ttf have different effects on the character symbol (℃).

Guess you like

Origin blog.csdn.net/weixin_45866058/article/details/126442452