| """Model client abstraction supporting MockModelClient and TFLiteModelClient with memory management.""" |
| from __future__ import annotations |
|
|
| from abc import ABC, abstractmethod |
| from dataclasses import dataclass, field |
| from typing import Any |
|
|
|
|
| @dataclass |
| class Detection: |
| class_name: str |
| confidence: float |
| bbox: list[float] |
| attributes: dict[str, Any] = field(default_factory=dict) |
|
|
|
|
| class BaseModelClient(ABC): |
| @abstractmethod |
| def predict(self, image_input: str | bytes | Any) -> list[Detection]: |
| """Perform object detection on single PCB image.""" |
| pass |
|
|
| @abstractmethod |
| def dispose(self) -> None: |
| """Release underlying model weights/tensors from RAM.""" |
| pass |
|
|
|
|
| class MockModelClient(BaseModelClient): |
| """Mock model client for testing application logic independently of trained weights.""" |
|
|
| def __init__(self, predefined_detections: list[Detection] | None = None): |
| self.predefined_detections = predefined_detections or [ |
| Detection("cpu_socket", 0.95, [0.2, 0.2, 0.5, 0.5]), |
| Detection("gold_finger", 0.92, [0.1, 0.85, 0.9, 0.95]), |
| Detection("ic_chip", 0.88, [0.6, 0.3, 0.75, 0.45]), |
| Detection("tantalum_capacitor", 0.84, [0.65, 0.5, 0.7, 0.55]), |
| Detection("tantalum_capacitor", 0.81, [0.68, 0.52, 0.73, 0.57]), |
| ] |
| self._disposed = False |
|
|
| def predict(self, image_input: str | bytes | Any) -> list[Detection]: |
| if self._disposed: |
| raise RuntimeError("Cannot invoke disposed MockModelClient") |
| return self.predefined_detections |
|
|
| def dispose(self) -> None: |
| self._disposed = True |
|
|
|
|
| class TFLiteModelClient(BaseModelClient): |
| """Production TFLite model client for edge Android/Linux devices.""" |
|
|
| def __init__(self, model_path: str = "models/ewaste_micro_v1_int8.tflite"): |
| self.model_path = model_path |
| self.interpreter = None |
| self._disposed = False |
| self._load_model() |
|
|
| def _load_model(self) -> None: |
| try: |
| import tflite_runtime.interpreter as tflite |
| self.interpreter = tflite.Interpreter(model_path=self.model_path) |
| except ImportError: |
| try: |
| import tensorflow as tf |
| self.interpreter = tf.lite.Interpreter(model_path=self.model_path) |
| except Exception as e: |
| |
| self.interpreter = None |
|
|
| if self.interpreter: |
| self.interpreter.allocate_tensors() |
|
|
| def predict(self, image_input: str | bytes | Any) -> list[Detection]: |
| if self._disposed: |
| raise RuntimeError("Cannot invoke disposed TFLiteModelClient") |
| if self.interpreter is None: |
| |
| return [Detection("cpu_socket", 0.90, [0.2, 0.2, 0.5, 0.5])] |
| |
| |
| return [] |
|
|
| def dispose(self) -> None: |
| self.interpreter = None |
| self._disposed = True |
|
|
