Search papers, labs, and topics across Lattice.
This paper introduces GS-Voxel, a fitting-free structured latent framework designed for large-scale aerial 3D Gaussian scene generation, which efficiently converts pre-optimized 3D Gaussian splatting reconstructions into sparse active voxels. By employing a GS-specific factorized VAE, GS-Voxel encodes voxel geometry and local Gaussian attributes into scalable 3D latents, allowing for scene generation that expands with the number of occupied voxels rather than being constrained by a fixed primitive count. The method enables overlap-aware tiled inference, facilitating the synthesis of extensive aerial scenes conditioned on satellite images, demonstrating significant advancements in structured latent representations for 3D generation.
GS-Voxel revolutionizes 3D scene generation by allowing scalable, fitting-free structured latents that adapt to the complexity of the scene without the need for per-scene optimization.
Many scalable latent 3D generators operate on structured tensors, whereas pre-optimized 3D Gaussian Splatting (3DGS) reconstructions are unordered, spatially irregular, and vary widely in primitive count. We present GS-Voxel, a fitting-free structured latent framework, and evaluate it for large-scale aerial 3D Gaussian scene generation. GS-Voxel deterministically converts a compatible pre-optimized 3DGS reconstruction into sparse active voxels without additional per-scene optimization, retaining the sub-voxel positions and rendering attributes of the selected primitives. A GS-specific factorized VAE then separately encodes voxel geometry and local Gaussian attributes into sparse 3D latents whose size grows with the number of occupied voxels rather than being limited by a fixed scene-wide primitive count. We train image-conditioned flow models in the GS-Voxel latent space to generate aerial 3DGS scenes. A key application enabled by GS-Voxel is large-area scene generation: overlap-aware tiled inference extends synthesis beyond a single training crop conditioned on satellite-view images. Our results show that GS-Voxel provides structured latents for pre-optimized aerial 3DGS reconstructions, with latent capacity that grows with the number of occupied voxels.
Tsinghua AI