Version 1.6.1 (2025-12-22)

🔧 Plugin Schema Versioning and ML-Ready Architecture

Version 1.6.1 introduces schema versioning for detector plugin contracts and multi-framework image support, laying the groundwork for ML-based detectors in v2.x.

Highlights

• Schema versioning: DetectionContext and DetectionResult now include schema_version field
• Multi-framework images: New ImageLike type supports numpy, PyTorch, and PIL
• Enhanced diagnostics: DetectionResult.metrics for standardized performance metrics
• Serialization support: DetectionResult.to_dict() for JSON-compatible output

Why This Change?

This release prepares the plugin architecture for:

1. Future schema migrations: Version field enables gradual migration without breaking existing plugins
2. ML detector support: ImageLike type allows detectors to work with PyTorch tensors natively
3. Standardized metrics: Consistent diagnostics across different detector implementations

Schema Changes

DetectionContext (v1.6.1):

@dataclass
class DetectionContext:
    current_image: ImageLike      # NEW: Union[np.ndarray, torch.Tensor, PIL.Image]
    previous_image: ImageLike     # NEW: Union[np.ndarray, torch.Tensor, PIL.Image]
    roi_mask: Any
    runtime_params: Dict[str, Any]
    metadata: Dict[str, Any]
    schema_version: int = 1       # NEW: For migration support

DetectionResult (v1.6.1):

@dataclass
class DetectionResult:
    is_candidate: bool
    score: float
    lines: List[Tuple[int, int, int, int]]
    aspect_ratio: float
    debug_image: Optional[Any]
    extras: Dict[str, Any]
    metrics: Dict[str, Any]       # NEW: Standard diagnostics
    schema_version: int = 1       # NEW: For migration support

    def to_dict(self) -> Dict[str, Any]:  # NEW: Serialization
        ...

New Utility Functions

ensure_numpy() in meteor_core.utils:

from meteor_core.utils import ensure_numpy

# Convert any ImageLike to numpy array
image = ensure_numpy(context.current_image)  # Works with numpy, torch, PIL

Migration Guide for Plugin Authors

No migration required. Existing plugins work without modification.

Optional enhancements:

1. Use ensure_numpy() for type-safe image handling
2. Populate metrics dict with standard diagnostics
3. Use result.to_dict() for logging

Files Changed

Backward Compatibility

✅ Fully backward compatible with v1.6.0:

• CLI: No changes
• Runtime: No changes to detection behavior
• API: All existing detector plugins work unchanged
• Configuration: No changes