Framework

DreamGaussian4D generates 4D contents in minutes by leveraging 4D Gaussian Splatting. We first obtain a static 3D GS model and a driving video and then optimize a deformation network. Finally, per-frame meshes can be refined with a video-to-video pipeline.

Image-to-4D Results

Qualitative results on Image-to-4D generation.

Video-to-4D Results

Comparison with baselins on Video-to-4D. Our approach requires sigificant less optimization time (15 mins v.s. 2 hours).

1 2 3 4 5 6

Ablation Results

We present more ablation results in this the following webpage 👇