🦎 CamoGED

Platform pillars: Generation · Evaluation · Detection

A cross-domain open platform for the science, art, and engineering of concealment — spanning Nature · War · Machines and the Physical · Digital · Intelligent domains.

---

Working thesis. Camouflage can be studied as an observer-, channel-, and task-dependent interaction between a hider and a seeker, with evaluation defining what counts as success. Biology, military engineering, visual culture, and AI share parts of this analytical structure, while retaining different mechanisms, histories, constraints, and utility functions.

CamoGED brings camouflage generation, detection, and evaluation into one open research environment, with natural-science, military, visual-culture, and machine-perception perspectives represented across physical, digital, and learning-based settings. The project combines a living monograph, curated metadata, reproducible evaluation tools, and research demos. Claims of novelty or comprehensive coverage are treated as versioned literature-search conclusions rather than permanent promotional facts; see docs/BOOK_SEARCH_METHODOLOGY.md.

Component status

Component	Current status	Trust boundary
Monograph	publication candidate	not yet rights- and format-cleared
camo-eval COD core	validated research preview	report package version and protocol
camo-eval extensions	experimental	use explicit *_lite or descriptive names
Paper/dataset registry	metadata preview	coverage is selective; licenses may be unknown
Results registry	minimal source-checked records	not a comprehensive ranking
Browser demo	synthetic teaching tool	no model inference or standard difficulty score
coder / flowcamo / dualvcod	research agendas	no integrated methods or verified results

Security reports use SECURITY.md. Third-party asset policy is recorded in THIRD_PARTY_ASSETS.yml. Release state is recorded in RELEASE_MANIFEST.yml.

Analytical map

                  Hide–reveal interaction                  ← working framework
   ┌──────────────────────────┼──────────────────────────┐
 Nature                 Military systems           Visual culture
   └──────────────────────────┼──────────────────────────┘
             Physical ── Digital ── Learning-based       ← settings
   ┌──────────────────────────┼──────────────────────────┐
 Generation                 Detection                Evaluation
  (hider)                    (seeker)              (protocol/judge)

The map is an organizing framework, not a claim that all four domains share identical mechanisms or a single solved game-theoretic model.

Quick start

# Clone
git clone https://github.com/MichaelCSHN/CamoGED.git && cd CamoGED

# Evaluation toolkit
cd camo-eval && pip install -e ".[dev]" && pytest -q

# Book (requires Quarto)
cd book && quarto preview

# Website (requires Node 18+)
cd web && npm install && npm run docs:dev

Roadmap

Build the slow-changing pieces first, the fast-iterating ones last: ① repository foundation → ② monograph and evaluation contracts → ③ curated metadata → ④ website and demos.

• Foundation: governance, licenses, CI, metadata schemas, and camo-eval core.
• Monograph: editorially controlled chapters, fact-check register, executable examples, and publication proofing.
• Metadata: verified papers, datasets, and protocol-aware result records.
• Demos: stable evaluation demos first; heavier generation or model demos only after reproducibility and risk review.
• Publication: archive a reviewed release, then add DOI and publisher/companion-paper metadata.

Full plan: docs/PROJECT_PLAN.md · Book compilation plan: docs/BOOK_COMPILATION_PLAN.md.

Contributing

We welcome papers, datasets, methods, demos, and corrections. See CONTRIBUTING.md and use the structured issue templates. Please also read our Code of Conduct.

Responsible use

CamoGED covers military camouflage and adversarial attacks. All such content is provided for understanding, defense, and robustness evaluation. We do not publish directly weaponizable recipes. See ETHICS.md.

Citing

If CamoGED helps your work, please cite it via CITATION.cff. A DOI will be added only after a reviewed release is archived in a DOI-minting repository.

Acknowledgements

CamoGED builds on and links to prior community work, including the Awesome lists by visionxiang, ChunmingHe, clelouch, and the 2024 survey A Survey of Camouflaged Object Detection and Beyond.

License

Repository code is Apache-2.0 unless a subdirectory states otherwise; web/ code is MIT. Prose content (book, Awesome list, and documentation) is CC BY 4.0. Third-party material retains its upstream terms. Datasets are linked, not redistributed unless explicitly registered in THIRD_PARTY_ASSETS.yml.