Generative 4D world model via Flow-Matched Score Fields on 4D Gaussian Splatting. Self-supervised, no depth labels required. Supports single-image/video input, outputs renderable 4DGS fields with uncertainty.
fjm123456789/Depth-Anything-4: Generative 4D world model via Flow-Matched Score Fields on 4D Gaussian Splatting. Self-supervised, no depth labels required. Supports single-image/video input, outputs renderable 4DGS fields with uncertainty.
