handler.py

from ultralytics import YOLO

class EndpointHandler:
    def __init__(self, model_dir):
        """
        Initialize the EndpointHandler with the YOLO model from the given directory.
        """
        # Update the model file name to match your actual file
        self.model = YOLO(f"{model_dir}/tr_roi_finetune_130_large.pt")

    def preprocess(self, inputs):
        """
        Preprocess the input data for YOLO inference.
        :param inputs: JSON input data
        :return: Preprocessed data (image path)
        """
        image_path = inputs.get("image")
        if not image_path:
            raise ValueError("Input JSON must contain an 'image' key with a valid path.")
        return image_path

    def predict(self, inputs):
        """
        Run inference using the YOLO model.
        :param inputs: Preprocessed input data
        :return: Raw YOLO results
        """
        return self.model(inputs)

    def postprocess(self, outputs):
        """
        Postprocess YOLO results into a JSON-compatible format.
        :param outputs: Raw YOLO results
        :return: JSON results
        """
        detections = []
        for result in outputs:
            for box in result.boxes:
                detections.append({
                    "class": self.model.names[int(box.cls)],  # Class name
                    "confidence": box.conf.tolist(),         # Confidence score
                    "box": box.xyxy.tolist()                 # Bounding box coordinates
                })
        return {"detections": detections}

    def __call__(self, inputs):
        """
        Complete handler pipeline: preprocess -> predict -> postprocess.
        :param inputs: JSON input data
        :return: JSON output
        """
        preprocessed_data = self.preprocess(inputs)
        predictions = self.predict(preprocessed_data)
        return self.postprocess(predictions)