altametris.sara.core.base_detector¶

Base detector class for inference.

Provides a common interface for running inference on ML models with support for: - Model loading (local and Azure remote paths) - Batch and single image prediction - Device management - Warmup for performance optimization - Azure weight resolution via WeightResolver

Example

>>> class MyDetector(BaseDetector):
...     def _load_model(self, model_path):
...         self.model = torch.load(model_path)
...     def predict(self, source, **kwargs):
...         return self.model(source)
>>> # With Azure support
>>> detector = MyDetector(model_path="azure://models/best.pt", enable_azure=True)
>>> # Via APIM with OAuth2
>>> detector = MyDetector(model_path="https://apim.azure.com/weights/best.pt")

Attributes¶

`logger`
`HAS_WEIGHT_RESOLVER`

Classes¶

BaseDetector

Abstract base class for model inference/detection.

Module Contents¶

altametris.sara.core.base_detector.logger¶

altametris.sara.core.base_detector.HAS_WEIGHT_RESOLVER = True¶

class altametris.sara.core.base_detector.BaseDetector(model_path: str | pathlib.Path, device: str = 'auto', warmup: bool = False, enable_azure: bool = True, credentials: Any | None = None, weight_resolver: Any | None = None, **kwargs: Any)¶

Bases: abc.ABC

Abstract base class for model inference/detection.

Provides common inference infrastructure: - Model loading and initialization (local and Azure remote paths) - Device management - Warmup capability - Prediction interface - Azure weight resolution via WeightResolver

Parameters:

model_path – Path to model weights (local, azure://, or https://)
device – Device for inference (“cpu”, “cuda”, “mps”, “auto”)
warmup – Whether to run warmup inference on initialization
enable_azure – Enable Azure weight resolution for remote paths
credentials – DsCredentials instance for APIM OAuth2 authentication
weight_resolver – Custom WeightResolver instance (auto-created if None)

Example

>>> # Local path
>>> detector = MyDetector(model_path="weights/best.pt", device="cuda")
>>> results = detector.predict(source="image.jpg")
>>> # Azure Blob Storage (SDK direct)
>>> detector = MyDetector(model_path="azure://models/best.pt")
>>> # Via APIM with OAuth2 (auto-creates DsCredentials)
>>> detector = MyDetector(model_path="https://apim.azure.com/weights/best.pt")
>>> # Via APIM with explicit credentials
>>> from altametris.sara.core.auth import DsCredentials
>>> creds = DsCredentials()
>>> detector = MyDetector(model_path="https://apim.azure.com/weights/best.pt", credentials=creds)

model = None¶

model_path¶

property device: torch.device¶: Get current device.

property is_initialized: bool¶: Check if detector is initialized.

abstract predict(source: Any, **kwargs: Any) → Any¶

Run inference on source.

Parameters:

source – Input source (image path, array, video, etc.)
**kwargs – Inference parameters (conf, iou, etc.)

Returns:

Prediction results (format depends on detector type)

Raises:

InferenceError – If prediction fails

Note

Must be implemented by subclasses

warmup(iterations: int = 3) → None¶

Warmup the model with dummy inference.

Useful for GPU models to pre-allocate memory and compile kernels.

Parameters:: iterations – Number of warmup iterations

Example

>>> detector.warmup(iterations=5)

validate_source(source: Any) → None¶

Validate input source.

Parameters:: source – Input source to validate
Raises:: InferenceError – If source is invalid