Cross-view Self-localization from Synthesized Scene-graphs