A curated, dataset-first map of diffusion research — peer-reviewed first, with rare canonical preprint exceptions.
Diffusion Lighthouse helps researchers, students, and practitioners navigate the fast‑moving diffusion literature by surfacing work with lasting conceptual signal.
It is not an exhaustive index and it does not rank papers by benchmark performance. Instead, it curates papers that shaped how diffusion models are understood, trained, and applied — grounded in datasets, with explicit editorial judgment.
Like a lighthouse, this project does not chart every wave —
it helps you navigate toward the most important signals.
👉 https://GKalliatakis.github.io/diffusion-lighthouse/
The website is the canonical index.
- The repository supports the website.
- The website defines inclusion.
- If a paper is not on the website, it is not included.
- Not a leaderboard
- Not an automatic crawler
- Not a complete bibliography
- Not a benchmark comparison site
Exclusion is intentional.
This repository exists to power the website.
- Paper metadata is curated in YAML
- The site consumes a built
papers.json - The README intentionally does not duplicate the paper list
See the website for the actual index.
Diffusion Lighthouse is curated through an explicit editorial process.
Additions, revisions, and reclassifications take time by design. Papers are evaluated for lasting conceptual, methodological, or dataset-level signal — not for recency, popularity, or benchmark performance.
As a result:
- Not every notable paper is included
- Some papers may appear only after their historical role becomes clear
- Absence does not imply omission or oversight
The index evolves deliberately. Speed is not the goal; clarity is.
Well-defined contributions are welcome.
Because Diffusion Lighthouse is editorially curated, contributions are expected to be specific, motivated, and grounded in the project’s criteria. Examples of acceptable contributions include:
- Proposing a paper with a clear editorial rationale (why it delivers lasting conceptual, methodological, or dataset-level signal)
- Suggesting a correction or clarification to an existing entry
- Pointing out missing or incorrect canonical publication links
- Identifying relation errors or misclassifications
General paper lists, popularity-based suggestions, or benchmark-driven additions will not be considered.
All contributions are reviewed editorially, and inclusion is discretionary.
For the exact, repeatable steps to:
- update citations
- rebuild
papers.json - refresh the site
- troubleshoot common failures (Scholar blocks, stale artifacts, build issues)
see MAINTENANCE.md.
This project is released under the MIT License.
See LICENSE for details.