any ground-truth visual relationship annotations, avoiding the challenging manual annotation of visual relationships;