OpenVINOEngine

OpenVINOEngine(model_path: str | Path, device: str | OpenVINODevice = OpenVINODevice.auto, config: OpenVINOConfig | None = None, core: Any | None = None, input_shapes: dict[str, Any] | None = None)

• Description: OpenVINO inference wrapper that provides sync inference and an optional async queue.

• Dependencies

  • Requires openvino (install capybara-docsaid[openvino] first).

• Common methods

  • run(feed): Sync inference, returns dict[str, np.ndarray].
  • summary(): Returns a summary of inputs/outputs/device, etc.
  • benchmark(feed, repeat=..., warmup=...): Returns throughput and latency stats.
  • create_async_queue(...): Creates an async queue (for pipelining).

• Example

  import numpy as np
  from capybara.openvinoengine import OpenVINOConfig, OpenVINODevice, OpenVINOEngine
  
  engine = OpenVINOEngine(
      "model.xml",
      device=OpenVINODevice.cpu,
      config=OpenVINOConfig(num_requests=2),
  )
  outputs = engine.run({"input": np.ones((1, 3), dtype=np.float32)})
  print(outputs.keys())