Multi-Level Neural Scene Graphs for Dynamic Urban Environments

Figure 1. Overview. We represent sequences captured from moving vehicles in a shared geographic area with a multi-level scene graph. Each dynamic object vo is associated with a sequence node v t s and time t. The sequence nodes are registered in a common world frame at the root node vr through the vehicle poses P t s , while the dynamic objects are localized w.r.t. the sequence node with pose ξ t o . Each camera c is associated with an ego-vehicle position, i.e. node v t s , through the extrinsic calibration Tc. The sequence and object nodes hold latent codes ω that condition the radiance field, synthesizing novel views in various conditions with distinct dynamic objects.