Scene-Graph-Generation. Prediction of action and spatial visual relationships in images between objects in the VRD-Dataset using visual, semantic, spatial, and heatmap features with structual ranking loss