Material detectionYOLO26m fine-tuned on 93 hand-labelled frames + COCO detector for vehicles
Material classes come from a detector fine-tuned on frames from these cameras; vehicles come from the stock COCO model, which is far stronger there. Counts shown on the video are per-frame observations of the same physical stacks and are never summed across frames. Classes with very few training labels (solar, cable_reel, steel_rod_bundle) are not reliable.
