This illustration shows the outcome of our combined pipeline (two-stage) which leverages YOLOv8m as the Segmentation Model and YOLOv9-GELAN-C as the object detection model. By using our custom distance-based classification (which mimics the real-life spatial search and inspection), we achived a balanced accuracy of 83.6% which in turn reduces the manual inspection cost by 35%
