π Awesome OCR multiple programing languages toolkits based on ONNX Runtime, OpenVINO, MNN, PaddlePaddle, TensorRT and PyTorch.
chore(rapidocr): optim config code of openvino
SWHLba9ad22cf67d
diff --git a/python/rapidocr/inference_engine/openvino.py b/python/rapidocr/inference_engine/openvino.pyindex e4962a0..4a3adff 100644--- a/python/rapidocr/inference_engine/openvino.py+++ b/python/rapidocr/inference_engine/openvino.py@@ -4,6 +4,7 @@import osimport tracebackfrom pathlib import Path+from typing import Any, Dictimport numpy as npfrom omegaconf import DictConfig@@ -45,43 +46,47 @@ class OpenVINOInferSession(InferSession):model_path = Path(model_path)self._verify_model(model_path)- cpu_nums = os.cpu_count()+ config = self._init_config(cfg)+ core.set_property("CPU", config)++ model_onnx = core.read_model(model_path)+ compile_model = core.compile_model(model=model_onnx, device_name="CPU")+ self.session = compile_model.create_infer_request()++ def _init_config(self, cfg: DictConfig) -> Dict[Any, Any]:config = {}engine_cfg = cfg.get("engine_cfg", {})+infer_num_threads = engine_cfg.get("inference_num_threads", -1)- if infer_num_threads != -1 and 1 <= infer_num_threads <= cpu_nums:+ if infer_num_threads != -1 and 1 <= infer_num_threads <= os.cpu_count():config["INFERENCE_NUM_THREADS"] = str(infer_num_threads)-+performance_hint = engine_cfg.get("performance_hint", None)if performance_hint is not None:config["PERFORMANCE_HINT"] = str(performance_hint)-+performance_num_requests = engine_cfg.get("performance_num_requests", -1)if performance_num_requests != -1:config["PERFORMANCE_HINT_NUM_REQUESTS"] = str(performance_num_requests)-+enable_cpu_pinning = engine_cfg.get("enable_cpu_pinning", None)if enable_cpu_pinning is not None:config["ENABLE_CPU_PINNING"] = str(enable_cpu_pinning)-+num_streams = engine_cfg.get("num_streams", -1)if num_streams != -1:config["NUM_STREAMS"] = str(num_streams)-+enable_hyper_threading = engine_cfg.get("enable_hyper_threading", None)if enable_hyper_threading is not None:config["ENABLE_HYPER_THREADING"] = str(enable_hyper_threading)-+scheduling_core_type = engine_cfg.get("scheduling_core_type", None)if scheduling_core_type is not None:config["SCHEDULING_CORE_TYPE"] = str(scheduling_core_type)-- self.logger.info(f"Using OpenVINO config: {config}")- core.set_property("CPU", config)- model_onnx = core.read_model(model_path)- compile_model = core.compile_model(model=model_onnx, device_name="CPU")- self.session = compile_model.create_infer_request()+ self.logger.info(f"Using OpenVINO config: {config}")+ return configdef __call__(self, input_content: np.ndarray) -> np.ndarray:try: