CompactSplat: Spatially Adaptive Gaussian Distribution for Feedforward Scene Reconstruction

Key points

• Reconstructing 3D scenes from sparse views without per-scene optimization remains highly challenging, especially for recovering accurate geometry and fine textures.1
• While recent feedforward approaches leverage generalizable 3D Gaussian Splatting (3DGS) for scene generation, they typically assign one or multiple Gaussians to each pixel.1
• By integrating texture-guided spatial partitioning with hierarchical sampling, CompactSplat adaptively distributes primitives - concentrating dense Gaussians in complex regions while tiling flat areas with expanded, scale-modulated primitives.1
• Extensive experiments on RealEstate10K, DL3DV, and ScanNet demonstrate that CompactSplat consistently outperforms prior methods on both standard metrics and high-resolution rendering consistency, achieving high-fidelity reconstructions with significantly fewer primitives.1

Sources (1)

• [1]CompactSplat: Spatially Adaptive Gaussian Distribution for Feedforward Scene Reconstruction
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 9, 08:04 AM

  To address these limitations, we present CompactSplat, a novel feedforward framework that enables a content-aware allocation of 3D Gaussians.

  Reconstructing 3D scenes from sparse views without per-scene optimization remains highly challenging, especially for recovering accurate geometry and fine textures.

Extractive summary: sentences quoted from the sources.