Skip to content

Contributing

This is an opinionated playbook, not an awesome-list. Contributions are welcome, but the rules are different.

The rule for the Dump (easy contributions)

Anyone can add to the Dump of any section. The bar is:

  • The entry is useful — not a toy project or an abandoned fork.
  • It has a one-line verdict — what it's good at, what it's bad at. Not a marketing blurb.
  • It's in the right section — if it's a face-recognition library, it goes under face/, not under the VLM section because it happens to use a transformer.

Open a PR against the Dump portion of the relevant section. Include a link and a one-line honest take.

Changing the top-3 picks for any subtask needs more than "I think X is better now." It needs:

  • Why the current pick is wrong — not "outdated" in the abstract, but a specific failure mode.
  • Why the new pick is better — with a measurement or a clear qualitative argument.
  • A "when not to use this" section — every pick has one. A pick without a downside is a pick you didn't think about.

Open a PR. Expect discussion. If the argument is weak, it won't merge. This is by design.

The rule for a Graveyard entry

If a Recommended pick is being retired, it moves to the Graveyard section of that page. Don't delete it. Somebody somewhere still runs it, and the Graveyard is how they find out why the community moved on.

Graveyard entry format:

  • Name + year-retired.
  • Why it was the pick.
  • What replaced it and why.

Adding a whole new section

If you want to add a new top-level section (say, point-cloud or action-recognition), open an issue first. The section needs a clear scope and at least the three Recommended picks to justify its existence. An empty placeholder is worse than no section.

What won't merge

  • Self-promotion of your own project without a clear case.
  • "Awesome" formatting — long lists of GitHub links without verdicts.
  • Picks without a when-not-to-use clause.
  • Anything with emoji icons 🎯✨🚀. This is a reference, not a LinkedIn post.

The LLM-refresh caveat

If/when the scripts/refresh.md approach goes live, the Recommended picks on each page may be regenerated by running the Dump through an LLM with a curation prompt. In that case your Dump contributions still stick — the picks may rotate based on the dump, which is the point. If a pick rotates out that you thought was correct, open an issue with the argument for keeping it and we'll refine the prompt.