Contributions that improve portability, evaluation integrity, documentation, or RGB-T modeling are welcome.
-
Create a focused branch.
-
Keep datasets, checkpoints, predictions, and credentials outside Git.
-
Add or update a result card for every metric-changing contribution.
-
State whether a result is a single model, protocol-optimized model, routed system, or post-processing result.
-
Run:
python scripts/validate_release.py python -m unittest discover -s tests -v
-
Describe the exact protocol, seed, checkpoint hash, and evidence source in the pull request.
Please preserve the AGPL-3.0 license and upstream attribution.