Efficient 4D Gaussian Stream with Low Rank Adaptation
This work addresses efficient video streaming for dynamic scene reconstruction, though it appears incremental as it builds on existing 3D Gaussian and adaptation techniques.
The paper tackles dynamic novel view synthesis with continual learning by proposing a scalable method that uses 3D Gaussians and low-rank adaptation-based deformation to capture scene changes, achieving a 90% reduction in streaming bandwidth while maintaining rendering quality comparable to offline state-of-the-art methods.
Recent methods have made significant progress in synthesizing novel views with long video sequences. This paper proposes a highly scalable method for dynamic novel view synthesis with continual learning. We leverage the 3D Gaussians to represent the scene and a low-rank adaptation-based deformation model to capture the dynamic scene changes. Our method continuously reconstructs the dynamics with chunks of video frames, reduces the streaming bandwidth by $90\%$ while maintaining high rendering quality comparable to the off-line SOTA methods.