Upload an image to detect rockfalls. The model returns both the annotated image and raw detections (labels, confidence, bounding boxes).