- maybe re-train with higher 1280 resolution expected for cameras (yolo requires a square) - train wet vs dry pothole for yolo detections