new MFR model with SOTA performance

Merge pull request #63 from breezedeus/dev

add demo urls

Support Multilingual Recognition

Major Changes:

• Support for recognizing 80+ languages; for a detailed list of supported languages, see List of Supported Languages;
• Added domestic sites for automatic model downloads;
• Optimized the logic for merging detection boxes.

bugfix

bugfixed from @hiroi-sora , thanks much 👍 .

fix: remove useless CATEGORY_MAPPINGS from `consts.py`

bugfix: 去掉 consts.py 无用的 CATEGORY_MAPPINGS。

fix: results from `self.recognize_by_clf` have no `line_number`

Merge pull request #32 from breezedeus/dev

fix: results from `self.recognize_by_clf` have no `line_number`

optimize the whole processing flow

Merge pull request #27 from breezedeus/dev

optimize the processing method for mixed images

bugfix

Merge pull request #14 from breezedeus/dev

bugfix

Fix the cropping error caused by the detected rotation box

主要变更：

• 修复旋转框导致的识别结果错误；
• 去掉代码中不小心包含的 breakpoint()。

optimize the order of boxes (new version of CnSTD), and the latex result

主要变更：

• 增加后处理机制优化Latex-OCR的识别结果；
• 使用最新的 CnSTD 和 CnOCR，它们修复了一些bug。

支持识别既有文字又有公式的混合图片

利用 CnSTD 新版的数学公式检测（Mathematical Formula Detection，简称 MFD）能力，P2T V0.2 支持识别既包含文字又包含公式的混合图片