Skip to content

Contributing Guidelines

Welcome! We curate human activity and motion datasets with high editorial standards. Follow the steps below to propose new datasets, improve existing cards, or add tooling.

Table of Contents

  1. Ground Rules
  2. Adding a Dataset
  3. Updating an Existing Card
  4. Submitting Tooling or Docs
  5. Style Notes
  6. Review Checklist

Ground Rules

  • Be precise. Provide official links, paper citations, license terms, and version numbers.
  • Respect licenses. If data requires application or special approval, note it clearly.
  • Transparency only: do not mirror datasets without permission; always point to the original source.
  • Keep tone neutral and factual. Avoid marketing language.
  • Run validation helpers before opening a pull request (link checker, schema checks).

Adding a Dataset

  1. Create or copy a card template
  2. Use datasets/_template.md as a starting point.
  3. Place the new file inside the appropriate modality folder (vision/, wearable/, skeleton/, multimodal/, or emerging/).
  4. Fill in required sections
  5. Overview (modality, sensors, tasks, scale).
  6. Access & licensing (include application instructions if any).
  7. Reference paper(s) with BibTeX-ready citation.
  8. Baselines / benchmarks (minimum of one published baseline).
  9. Tooling & ecosystem (code repos, loaders, evaluation kits).
  10. Known challenges or caveats (class imbalance, annotation noise, etc.).
  11. Add to the taxonomy table
  12. Update README.md to include the dataset with modality and quick facts.
  13. Validate
  14. Run tools/validate_links.ps1 (in development) or the GitHub Action output on your fork.
  15. Run python tools/normalize_ascii.py --dry-run to ensure no unintended Unicode slips in.
  16. Ensure Markdown passes markdownlint (recommended).
  17. Open a pull request
  18. Use the "Dataset addition" PR template.
  19. Fill in the checklist and provide context on why the dataset matters.

Updating an Existing Card

  • Keep change scope focused; if you are fixing multiple datasets, split into separate commits.
  • Include references when updating baseline numbers or SOTA claims (conference, leaderboard).
  • If a link is dead, provide a working alternative and explain the change in your PR.
  • Flag major licensing changes in the PR description so reviewers can highlight it in release notes.

Submitting Tooling or Docs

  • Place scripts in tools/ with clear CLI usage and docstrings.
  • Provide minimal reproduction steps or tests when adding automation.
  • Documentation additions belong in docs/ (surveys, comparisons, design notes) unless tightly coupled with a dataset card.
  • Localization files live in i18n/. When starting a translation, copy the language-switch banner, keep relative links working, and add yourself as a contributor in the PR description.

Style Notes

  • Prefer American English spelling for consistency (e.g., "modeling").
  • Use present tense and active voice.
  • Tables should be GitHub-compatible Markdown.
  • Keep paragraphs under four sentences; use bullet lists for dense facts.
  • Link to DOI or arXiv when possible (https://doi.org/... or https://arxiv.org/abs/...).

Review Checklist

Reviewers confirm the following before merging: - [ ] Dataset card uses the template and required headings. - [ ] All external links resolve (no HTTP errors). - [ ] Licensing terms are documented and non-conflicting. - [ ] Baseline metrics cite a published source or official leaderboard. - [ ] README taxonomy table updated (if applicable). - [ ] Tests / validation scripts (if touched) run successfully. - [ ] PR title follows <type>: <summary> convention (e.g., dataset: add BABEL v1.0).

Thank you for making this catalog better for the community! Reach out via GitHub Discussions for questions or to propose larger initiatives.