Model Adapters

Adapters are the bridge between OpenEye and ML model backends. Each adapter implements loading and inference for a specific model type.

ModelAdapter ABC

All adapters extend openeye_ai.adapters.base.ModelAdapter:

openeye_ai/adapters/base.py

class ModelAdapter(ABC):
    _loaded: bool = False

    def load(self, model_dir: Path) -> None:
        self._do_load(model_dir)
        self._loaded = True

    @abstractmethod
    def _do_load(self, model_dir: Path) -> None:
        """Load model weights from disk."""

    def predict(self, image: Image.Image) -> dict[str, Any]:
        if not self._loaded:
            raise ModelNotLoadedError(...)
        return self._do_predict(image)

    @abstractmethod
    def _do_predict(self, image: Image.Image) -> dict[str, Any]:
        """Run inference. Return dict with 'objects' and/or 'depth_map'."""

    @abstractmethod
    def pull(self, model_dir: Path) -> None:
        """Download model weights."""

Built-in Adapters

Adapter	Module	Models
yolov8	openeye_ai.adapters.yolov8	YOLOv8 (ultralytics)
yolo26	openeye_ai.adapters.yolo26	YOLOv26
depth_anything	openeye_ai.adapters.depth_anything	Depth Anything V2
grounding_dino	openeye_ai.adapters.grounding_dino	Grounding DINO
rfdetr	openeye_ai.adapters.rfdetr	RF-DETR
sam2	openeye_ai.adapters.sam2	Segment Anything 2
smolvla	openeye_ai.adapters.smolvla	SmolVLA
yolov8:onnx	openeye_ai.adapters.yolov8_onnx	YOLOv8 ONNX
onnx_generic	openeye_ai.adapters.onnx_runtime	Generic ONNX
yolov8:tensorrt	openeye_ai.adapters.tensorrt_runtime	YOLOv8 TensorRT
tensorrt_generic	openeye_ai.adapters.tensorrt_runtime	Generic TensorRT

Lifecycle

pull(model_dir) — Download weights to the model directory
load(model_dir) — Load weights into memory
predict(image) — Run inference on a PIL Image
Return dict must include inference_ms and either objects or depth_map