denoise.registry
denoise.registry — local model registry for Noise2Inverse trained models.
- Registry location (in order of precedence):
DENOISE_REGISTRY environment variable
~/.denoise/registry/
- Each entry is a sub-directory containing:
config.yaml — full denoise config (with metadata: block) best_val_model.pth best_lcl_model.pth best_edge_model.pth
- denoise.registry.list_registry()[source]
Return all registry entries as a list of (name, config) tuples.
- denoise.registry.register(config_path, model_dir, name=None)[source]
Copy model_dir/{best_val,best_lcl,best_edge}_model.pth and config_path into the registry under an auto-generated (or user-supplied) name.
Returns the Path of the registry entry directory.
- denoise.registry.search(config_path)[source]
Search the registry for models matching the noise fingerprint of config_path.
Returns a list of dicts, sorted by score descending:
[{'dir': Path, 'config': dict, 'matched': int, 'total': int, 'score': float, 'registered': str}, ...]
Returns an empty list when the registry is empty or the config has no
metadata:block.