Add ONNX inference

fast_plate_ocr/inference/onnx_inference.py

@@ -0,0 +1,69 @@
+"""
+ONNX inference module.
+"""
+
+import logging
+from contextlib import nullcontext
+
+import numpy.typing as npt
+import onnxruntime as ort
+
+import fast_plate_ocr.inference.config
+from fast_plate_ocr.common.utils import log_time_taken
+from fast_plate_ocr.inference import hub
+from fast_plate_ocr.inference.process import postprocess_output, preprocess_image, read_plate_image
+
+logging.basicConfig(level=logging.INFO)
+
+
+class ONNXInference:
+    """
+    ONNX inference class for performing license plates OCR.
+    """
+
+    def __init__(self, ocr_model: str, use_gpu: bool = False, log_time: bool = False):
+        """
+        Initialize ONNXInference.
+
+        :param ocr_model: Name of the OCR model to use.
+        :param use_gpu: Flag indicating whether to use GPU backend.
+        """
+        self.logger = logging.getLogger(__name__)
+        self.log_time = log_time
+
+        if use_gpu:
+            self.providers = ["CUDAExecutionProvider"]
+            self.device = "GPU"
+        else:
+            self.providers = ["CPUExecutionProvider"]
+            self.device = "CPU"
+
+        model_path, config_path = hub.download_model(model_name=ocr_model)
+        self.config = fast_plate_ocr.inference.config.load_config_from_yaml(config_path)
+        self.model = ort.InferenceSession(model_path, providers=self.providers)
+        self.logger.info("Using ONNX Runtime with %s device.", self.device)
+
+    def run(
+        self,
+        image_path: str,
+        return_confidence: bool = False,
+    ) -> tuple[list[str], npt.NDArray] | list[str]:
+        """
+        Runs inference on an image.
+
+        :param image_path: Path to the input image file.
+        :param return_confidence: Whether to return confidence scores along with plate predictions.
+        :return: Decoded license plate characters as a list.
+        """
+        x = read_plate_image(image_path)
+        with log_time_taken("Pre-process") if self.log_time else nullcontext():
+            x = preprocess_image(x, self.config["img_height"], self.config["img_width"])
+        with log_time_taken("Model run") if self.log_time else nullcontext():
+            y: list[npt.NDArray] = self.model.run(None, {"input": x})
+        with log_time_taken("Post-process") if self.log_time else nullcontext():
+            return postprocess_output(
+                y[0],
+                self.config["max_plate_slots"],
+                self.config["alphabet"],
+                return_confidence=return_confidence,
+            )