Abstract
This paper addresses the challenge of reconstructing dynamic 3D scenes withcomplex motions. Some recent works define 3D Gaussian primitives in thecanonical space and use deformation fields to map canonical primitives toobservation spaces, achieving real-time dynamic view synthesis. However, thesemethods often struggle to handle scenes with complex motions due to thedifficulty of optimizing deformation fields. To overcome this problem, wepropose FreeTimeGS, a novel 4D representation that allows Gaussian primitivesto appear at arbitrary time and locations. In contrast to canonical Gaussianprimitives, our representation possesses the strong flexibility, thus improvingthe ability to model dynamic 3D scenes. In addition, we endow each Gaussianprimitive with an motion function, allowing it to move to neighboring regionsover time, which reduces the temporal redundancy. Experiments results onseveral datasets show that the rendering quality of our method outperformsrecent methods by a large margin.