EmerNERF aims to representing and reconstructing dynamic scenes (4D = 3D + time).
Features:
- Decouples static and dynamic scene components
- Estimates 3D scene flows, use them to aggregate temporally-displaced features
Method:
- High-level: Builds a hybrid static-dynamic world representation via a density-regularized objective, generating density for dynamic objects only as necessary (i.e., when points intersect dynamic objects).