UniForge3D: Joint 3D Reconstruction and Generation in a Shared Anchor-Guided Space
UniForge3D aligns reconstruction and generation in one anchor-guided 3D space to reduce coordinate mismatch.
The paper targets complete 3D asset creation from sparse, unposed images. Its method keeps the object upright while following an anchor view’s azimuth, letting reconstruction and generation exchange features directly. It also uses the generated sparse voxel structure to choose the most relevant input view for each voxel, aiming to preserve local geometry. The authors report state-of-the-art appearance and geometry results on Toys4K and GSO. HF Daily Papers' note
The paper targets complete 3D asset creation from sparse, unposed images. Its method keeps the object upright while following an anchor view’s azimuth, letting reconstruction and generation exchange features directly. It also uses the generated sparse voxel structure to choose the most relevant input view for each voxel, aiming to preserve local geometry. The authors report state-of-the-art appearance and geometry results on Toys4K and GSO. HF Daily Papers' note
score 5