← rmpp add-ons

An open handwriting dataset, written by the reMarkable community

Write a short practice notebook on your reMarkable Paper Pro, in your own language, and share it as pen strokes. Everything is public (CC BY 4.0), so anyone can use it: handwriting recognition, note apps, other devices, research.

Take part

You write the language chosen at the top of the page; each practice notebook mixes it with English and digits.

–writers
–pages
–lines
–character samples
–languages

Repository

What is collected

Handwriting on printed practice pages, stored as pen strokes (coordinates, pressure, width). There are four kinds of pages:

PageHow to write
CellsOne character per cell with a light cross guide (for scripts like Hangul, kana or Chinese characters). Cell sizes change per row, for different writing sizes.
Digits and symbolsNumbers, dates, prices, and look-alike characters (0 O o, 1 l I, ...) in cells.
English four linesEnglish sentences on four guide lines.
Free linesNatural writing on a baseline; some lines quickly, like everyday notes.

Every cell and line carries the text that was printed, so the data works for character-level and line-level training.

Languages

Each practice notebook mixes your language + English + digits (English writers: English + digits), and the data of each language lives in its own folder data/<code>/. Supported languages are in languages.json; choosing one at the top shows this page in that language.

Your language is missing? Ask for it with the "Add a language" form (or "My language isn't listed" at the top). A new language first collects 100 practice sentences (sent by anyone, checked automatically); then its join form opens and it can be chosen on this page.

Privacy

How to take part

  1. What you need A reMarkable Paper Pro with OS 3.29 and developer mode on, a computer (Windows, macOS, Linux), a GitHub account and the GitHub CLI (gh). Log in once with gh auth login. rmpp-hwr is made for and tested on the Paper Pro with OS 3.29 only; other reMarkable models are not supported and their compatibility is not guaranteed. Following newer OS versions happens when there is time, and may not happen.
  2. Install rmpp-hwr On the Paper Pro: run rmpp setup from rmpp-packages and tick rmpp-hwr. After a reMarkable OS update, open rmpp setup again and press Restore. On the computer, the hwr tool makes practice notebooks and sends your pages:
    git clone https://github.com/lorancecall/rmpp-hwr
    cd rmpp-hwr
    uv venv ../.venv && uv pip install -p ../.venv -r collect/requirements.txt
    cp .env.example .env      # the Paper Pro's address, your language (HWR_LANG)
    Details: rmpp-hwr.
  3. Join (consent) Open the join form for your language (pick it at the top of the page) and tick the five consent items. When the confirmation comment appears, note the issue number.
  4. Get a practice notebook
    hwr practice --v2 --lang en    # English + digits
    hwr practice --v2 --lang ko    # Korean + English + digits
    A notebook rmpp-practice-v2-… (8 pages) appears in the Paper Pro folder hwr-db/<language>.
  5. Write Follow the line at the top of each page: one character per cell, not far outside it. Your usual handwriting helps most. Erase and rewrite mistakes.
  6. Collect and check
    hwr pull
    hwr qc
    It reports empty cells or strokes that run into the next cell; fix them on the Paper Pro and pull again.
  7. Submit
    hwr contribute --consent <issue number>
    --consent is needed the first time only. A pull request with the anonymized pages opens automatically; after the checks pass, a maintainer merges it. Later, hwr contribute sends only new pages.

Consent in short

Withdraw

Use the withdraw form from the same GitHub account as your join issue. Your writer id is in hwr-data/writer.json on your computer. Your data is removed from the next version of the dataset and of the Hugging Face mirror.

Use the data

Questions

Does my handwriting have to be neat?

No. Your everyday handwriting helps most: the point is many different hands.

How much should I write?

One notebook (8 pages) is plenty; more is welcome. Every notebook has new sentences.

Does it improve my own model too?

Yes. Your pages stay on your computer, and hwr train uses them for a model tuned to your handwriting. The public model is trained regularly on everyone's data.