Install LaTeX-OCR recognition tool on M1

date

Feb 25, 2023

slug

install-latex-ocr-on-apple-silicon-m1

status

Published

summary

LaTeX-OCR is a well-known, open-source, and free OCR recognition tool that is widely used as an alternative to Mathpix. However, due to compatibility issues with cross-platform modules such as pytorch and qt, installing this tool on an arm64 host requires certain operations.

Solution

Install compiled qt library through brew

Install pytorch through conda

Steps in details

pip install "pix2tex[gui]" the module

brew install pyqt@5 the GUI dependency

sudo cp -r /opt/homebrew/Cellar/pyqt@5/5.15.7_2/lib/python3.9/site-packages/* /Users/rey/miniconda3/lib/python3.9/site-packages/ * since you cannot find a compatible version on the official forge of pip or conda

pip install pynput screeninfo required by the package

conda install pytorch torchvision conda will try to install a better version of pytorch on arm 64

Usage

Start python -m pix2tex latexocr

Screenshot the LaTeX image to be recognized to the clipboard

Wait for the terminal to display the recognized text, and the LaTeX text has been automatically copied to the clipboard.

Recognition effect

For the chained probability formula in the Stanford CS324 Large Language Model lecture notes:

Recognized as LaTeX code:

{\\mathbf{p}}(\\mathbf{x}*{1,\\mathrm{L}})=\\mathbf{p}(\\mathbf{x}*{1})\\mathbf{p}(\\mathbf{x}*{2}\\mid\\mathbf{x}*{1})\\mathbf{p}(\\mathbf{x}*{3}\\mid\\mathbf{x}*{1},\\mathbf{x}*{2})\\cdots\\mathbf{p}(\\mathbf{x}*{\\mathrm{L}}\\mid\\mathbf{x}*{1,\\mathrm{L}-1})=\\prod*{\\mathrm{i=1}}^{\\mathrm{L}}\\mathbf{p}(\\mathbf{x}*{\\mathrm{i}}\\mid\\mathbf{x}*{1,\\mathrm{i,t-1}).

Rendered equivalent to:

You can see that there are some small errors, but it can save a lot of time.

Reference

[2] Official installation tutorial: https://github.com/lukas-blecher/LaTeX-OCR

用中文阅读