MEFA: Multimodal Image Early Fusion with Attention Module for Pedestrian and Vehicle Detection | Synapse