RadiusFPS: Efficient Farthest Point Sampling on CPUs and GPUs via Spherical Voxel Pruning
Researchers are developing new methods for 3D generation using diffusion models and voxel-based approaches. SymTRELLIS enforces symmetry in 3D models by learning linear transformations on voxel latents, improving physical usability. MeshWeaver uses a multi-level sparse-voxel encoder for autoregressive mesh generation, enhancing geometric context and compression. Discrete Voxel Diffusion (DVD) offers a framework for generating, assessing, and editing sparse voxels, providing interpretable dynamics and uncertainty estimation. MeshFlow generates artistic 3D meshes efficiently using a VAE and a Rectified Flow transformer, achieving faster generation times. PatchScene employs a patch-based voxel diffusion paradigm for large-scale LiDAR scene completion, ensuring coherent reconstruction and temporal consistency. AI
IMPACT These papers introduce novel techniques for 3D generation, potentially improving efficiency, fidelity, and applicability in areas like autonomous driving and artistic creation.