【Hackathon 8th No.40】在 PaddleOCR 中复现 MixTeX 模型

[robinbg]

No description provided.

[paddle-bot]

Thanks for your contribution!

[Copilot]

Pull Request Overview

This PR implements the MixTeX model in PaddleOCR by adding new inference, training, and vocabulary generation scripts as well as the corresponding model components, data processing logic, loss function, and documentation.

• Added inference script (predict_mixtex.py) for evaluating MixTeX models.
• Integrated new model components including the backbone, head, and loss for MixTeX.
• Provided data processing tools, configuration files, and updated documentation in both English and Chinese.

Reviewed Changes

Copilot reviewed 13 out of 13 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
tools/infer/predict_mixtex.py	Added inference script and initialization for MixTeX recognition.
tools/generate_mixtex_vocab.py	Created script to generate frequency-based vocabulary for MixTeX.
ppocr/postprocess/mixtex_postprocess.py	Added post-processing utility for decoding model outputs into LaTeX strings.
ppocr/modeling/heads/mixtex_head.py	Implemented a transformer decoder head for MixTeX with both training and inference modes.
ppocr/modeling/backbones/mixtex_backbone.py	Introduced an efficient convolutional backbone for feature extraction.
ppocr/modeling/architectures/mixtex.py	Assembled the MixTeX architecture by combining the backbone and head.
ppocr/losses/mixtex_loss.py	Developed a custom loss function integrating cross-entropy with label smoothing.
ppocr/data/imaug/mixtex_process.py	Added data processing and tokenization routines for MixTeX training/inference.
docs/algorithm/formula_recognition/mixtex_en.md	Updated English documentation for the MixTeX model.
docs/algorithm/formula_recognition/mixtex.md	Updated Chinese documentation for the MixTeX model.
doc/mixtex/mixtex_architecture.txt	Provided textual architecture diagram for MixTeX.
doc/mixtex/mixtex_architecture.png	Added placeholder image for the MixTeX architecture diagram.
configs/rec/mixtex/mixtex_base.yml	Provided configuration file for training and evaluation of MixTeX.

Comments suppressed due to low confidence (1)

tools/infer/predict_mixtex.py:51

• The attribute name 'rec_algorith' appears to contain a typo; consider renaming it to 'rec_algorithm' for clarity.

self.rec_algorith = "MixTeX"

[Copilot]

[nitpick] The list of special characters used for tokenizing LaTeX in the 'encode' method is less comprehensive than the one used in the vocabulary generation script; consider aligning the tokenization rules to avoid potential inconsistencies.

Suggested change

	special_chars = ['\\', '{', '}', '_', '^', '&']
	special_chars = ['\\', '{', '}', '_', '^', '&', '%', '$', '#', '~', '[', ']', '(', ')', '|', '<', '>', '=', '+', '-', '*', '/', ':', ';', ',', '.', '!']

[luotao1]

请先提交RFC设计文档