models/detection/detector.py

import os
import logging
from PIL import Image, ImageDraw
from huggingface_hub import hf_hub_download

logger = logging.getLogger(__name__)

class ObjectDetector:
    def __init__(self, model_key="yolov8n", device="cpu"):
        self.device = device
        self.model = None
        self.model_key = model_key.lower().replace(".pt", "")

        hf_map = {
            "yolov8n": ("ultralytics/yolov8", "yolov8n.pt"),
            "yolov8s": ("ultralytics/yolov8", "yolov8s.pt"),
            "yolov8l": ("ultralytics/yolov8", "yolov8l.pt"),
            "yolov11b": ("Ultralytics/YOLO11", "yolov11b.pt"),
        }

        if self.model_key not in hf_map:
            raise ValueError(f"Unsupported model key: {self.model_key}")

        repo_id, filename = hf_map[self.model_key]
        self.weights_path = hf_hub_download(
            repo_id=repo_id,
            filename=filename,
            cache_dir="models/detection/weights",
            force_download=False
        )

    def load_model(self):
        if self.model is None:
            import torch  # Safe to import here
            from ultralytics import YOLO  # Defer import
    
            if self.device == "cpu":
                os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
                
            # Initialize model
            self.model = YOLO(self.weights_path)
    
            # Move to CUDA only if necessary and safe
            if self.device == "cuda" and torch.cuda.is_available():
                self.model.to("cuda")
    
        return self



    def predict(self, image: Image.Image, conf_threshold=0.25):
        self.load_model()
        results = self.model(image)
        detections = []
        for r in results:
            for box in r.boxes:
                detections.append({
                    "class_name": r.names[int(box.cls)],
                    "confidence": float(box.conf),
                    "bbox": box.xyxy[0].tolist()
                })
        return detections

     def draw(self, image: Image.Image, detections, alpha=0.5):
        overlay = image.copy()
        draw = ImageDraw.Draw(overlay)
        for det in detections:
            bbox = det["bbox"]
            label = f'{det["class_name"]} {det["confidence"]:.2f}'
            draw.rectangle(bbox, outline="red", width=2)
            draw.text((bbox[0], bbox[1]), label, fill="red")
        return Image.blend(image, overlay, alpha)