Update models/detection/detector.py

models/detection/detector.py

@@ -3,20 +3,16 @@ from PIL import Image, ImageDraw
 from huggingface_hub import hf_hub_download
 from ultralytics import YOLO
 import os
-import shutil
 import torch
 
 # Setup logger
 logger = logging.getLogger(__name__)
 logging.basicConfig(level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s")
 
-# Optional: clear weights cache each time (only for dev use)
-shutil.rmtree("models/detection/weights", ignore_errors=True)
-
 class ObjectDetector:
     def __init__(self, model_key="yolov8n.pt", device="cpu"):
         """
-        Initializes an Ultralytics YOLO model using HF download path.
+        Initializes an Ultralytics YOLO model path, defers actual model loading.
 
         Args:
             model_key (str): e.g. 'yolov8n.pt', 'yolov8s.pt', etc.
@@ -50,16 +46,16 @@ class ObjectDetector:
             force_download=False
         )
 
-        self.model = None  # 🔁 Don't initialize on construction
-        logger.info(f"Model path ready for {resolved_key}: {self.weights_path}")
+        logger.info(f"✅ YOLO weights ready for {resolved_key} at {self.weights_path}")
+        self.model = None  # defer loading
 
     def load_model(self):
         if self.model is None:
-            logger.info("⏳ Loading YOLO model into memory...")
+            logger.info("⚙️ Loading YOLO model into memory (runtime-safe)")
             self.model = YOLO(self.weights_path)
             if self.device == "cuda" and torch.cuda.is_available():
                 self.model.to("cuda")
-            logger.info(f"✅ YOLO model loaded on {self.device}")
+            logger.info(f"✅ YOLO model initialized on {self.device}")
 
     def predict(self, image: Image.Image, conf_threshold=0.25):
         self.load_model()