LaTeX to Word converter

Convert a LaTeX paper to Microsoft Word (.docx) with pandoc while keeping the mathematics, the figures and an IEEE-style bibliography, and applying a Word style template.

The script handles the cases pandoc does not cover on its own:

• IEEE bibliography. Citations are numbered [1], [2], ... in order of appearance. The reference text is taken from the .bbl produced by IEEEtran.bst, and a DOI is appended to each entry (read from the doi field in the .bib).
• Cross references. \ref and \eqref are replaced by the numbers LaTeX already computed, read from the .aux file (so Algorithm~\ref{...} becomes Algorithm 3, Eq.~\eqref{...} becomes Eq. (2), and so on).
• Equations. Displayed equations are centered and numbered with the number at the right margin, using a right-aligned tab (Word's native equation-numbering layout, not a table). Pseudocode (algorithm / algorithmic) gets a left-aligned header (heading, Input, Output) framed by a top and bottom rule, followed by the numbered steps in a two-column table (one row per line) so every line stays separate, while the inline math is preserved as real Word equations.
• Figures. Figures are embedded whatever their format. They are looked up in the folder given with --fig-dir (the folder of the .tex when it is left empty), then in the \graphicspath folders and figures/, by bare name or with any extension. PDF, EPS, PS and SVG, which Word cannot embed, are rendered to PNG.
• Word styles. When a template is supplied, pandoc uses it, the tables are centered, and the auto-generated paragraph styles (figure/table captions, author, date) are mapped onto the template styles.

• Python 3.8 or newer (standard library only).
• pandoc 2.11 or newer (3.x recommended).
• A compiled LaTeX project: the converter needs the .aux and .bbl files, so run LaTeX and BibTeX once before converting. The .bbl must come from a numeric, appearance-ordered style such as IEEEtran.
• To embed PDF, EPS or SVG figures, one of poppler (pdftocairo, pdftoppm), PyMuPDF, ImageMagick or Ghostscript — whichever is installed is used, best first (rsvg-convert, Inkscape or cairosvg for SVG). PNG and JPG need nothing extra.

# Convert the bundled example:
python latex_to_word.py main.tex -d ./example -t ./reference.docx

Options:

Figures live next to the .tex unless you say otherwise; point --fig-dir at another folder when they are kept apart from the sources:

python latex_to_word.py main.tex -d ./my-paper -g ~/Pictures/my-paper-figures

By default the references keep the IEEE layout, with the year at the end of each entry. Add --year-after-authors to get the author-year layout:

[1] J. Yin, Z. Zheng, and L. Cao (2012), "USpan: ...," in Proceedings ... pp. 660-668. DOI: ...

The same settings can be passed through the L2W_MAIN, L2W_DIR, L2W_TEMPLATE, L2W_FIG_DIR, L2W_FIG_DPI and L2W_YEAR_AFTER_AUTHORS environment variables when you prefer to set them once:

L2W_DIR=example L2W_MAIN=main.tex L2W_TEMPLATE=reference.docx python latex_to_word.py

When it finishes, the converter prints a short review checklist: items found in this run (references without a DOI, cross references it could not resolve, figures that did not embed) followed by general reminders about the parts pandoc cannot reproduce exactly (theorem numbering, equation-number alignment, pseudocode layout, styles). Open the .docx in Word and go through the list.

Latex2WordHJS.ipynb runs the same conversion on Google Colab. The first cell installs a recent pandoc; the second cell has input boxes for the LaTeX folder, the Word template and the main file; the third cell downloads the result.

latex_to_word.py      the converter
Latex2WordHJS.ipynb   Google Colab notebook
reference.docx        Word style template (Title, Subtitle, Figure, Table, ...)
example/              minimal runnable sample paper

The example/ folder is a small, self-contained paper (title, abstract, numbered equations, two pseudocode algorithms with pruning, a lemma, a figure, a table, and cross references to all of them plus two citations). It already ships with the .aux and .bbl files and the figure, so you can run the converter on it right away. Point -d at your own LaTeX project to convert your paper instead.

pandoc maps document elements onto named paragraph styles. The supplied reference.docx defines Title, Subtitle, Abstract, Figure, Table and the heading styles. To change the look of the output, edit those styles in reference.docx (or supply your own template with -t) and run the converter again. The script copies the chosen template into the paper folder as reference.docx before calling pandoc.

Comments, bug reports and suggestions are welcome. Please open an issue on GitHub or send an email to thaitm@huflit.edu.vn.

Released under the MIT License. You may use, modify and redistribute the code freely, including in your own projects, as long as you keep the copyright and license notice.