Multi-Modal Region Selection Approach for Training Object Detectors