MATERIAL AUDIT - Central Store_Video Pan_10.196.104.46_17.07.26 ==================================================================== Detector : YOLO26m fine-tuned on 93 hand-labelled frames from these cameras Vehicles : stock COCO detector (Truck/Car/Bus/Motorcycle/Person) Confidence threshold: 0.50 Frames analysed: 516 CLASS MEDIAN PEAK INTERPRETATION -------------------------------------------------------------------- Truck 1 8 typically visible Wooden_Pallet 0 7 seen only in some frames Person 0 3 seen only in some frames Scrap 0 3 seen only in some frames Steel_Bundle 0 3 seen only in some frames steel_coll 0 2 seen only in some frames Bus 0 1 seen only in some frames Car 0 1 seen only in some frames Long_rebars_wire 0 1 seen only in some frames cable_reel 0 1 seen only in some frames HOW TO READ THIS -------------------------------------------------------------------- Counts are PER-FRAME observations of the same physical objects as the camera pans. They are never summed across frames - that would count one steel bundle hundreds of times. MEDIAN = typically in view; PEAK = most in any single frame. RELIABILITY -------------------------------------------------------------------- Steel_Bundle / Wooden_Pallet / Scrap have hundreds of training labels and are the most dependable. solar (3 labels), cable_reel (7) and steel_rod_bundle (5) have too few examples to be trusted - treat those as hints, not counts. Steel_End was never labelled, so it is not detected at all.