CC-4DGS: Computational Deformation and Point-Cloud Compression for Storage-Efficient Dynamic Gaussian Splatting

IEEE TVCG 2026 · arXiv · Code · Pretrained models

Load a trained scene (a zip of the scene folder, or a folder in a Hugging Face repo) that contains point_cloud/iteration_<N>/deformation.pth with point_cloud.ply (CDF) or compressed/ (CDF+CCA). Add the dataset's poses_bounds.npy (N3DV) or COLMAP images.bin + cameras.bin (Technicolor) to render from the original cameras and spiral path.
Checkpoint source
Repo type
Scene config

'auto' matches the scene name and checks the deformation.pth architecture.
0 1
Camera
Fixed camera
0 1
0.25 1
Background
10 300
10 60
Camera motion