CompactSplat improves 3D scene reconstruction with fewer resources
CompactSplat: Spatially Adaptive Gaussian Distribution for Feedforward Scene Reconstruction
Computer Vision and Pattern Recognition
Summary
Reconstructing 3D scenes from a few images usually needs lots of computer processing and memory. The authors show a new method called CompactSplat that smartly puts more detail where it’s needed and simpler shapes where it’s not. This approach uses fewer building blocks to create more accurate shapes and textures in 3D scenes. It also lets users balance quality and memory use without retraining the system.
What this means in practice
- •For augmented reality developers: Generate detailed 3D scenes from sparse camera views that use fewer resources, enabling real-time rendering on mobile devices.
- •For game engine teams: Create high-fidelity 3D environments with adaptive detail allocation to improve rendering speed and reduce memory load.
Authors
Bing He, Jingnan Gao, Yunuo Chen, Zhengxue Cheng, Rong Xie, Li Song, Wenjun Zhang
Abstract
Reconstructing 3D scenes from sparse views without per-scene optimization remains highly challenging, especially for recovering accurate geometry and fine textures. While recent feedforward approaches leverage generalizable 3D Gaussian Splatting (3DGS) for scene generation, they typically assign one or multiple Gaussians to each pixel. Such uniform allocation not only produces highly redundant representations by wasting primitives in homogeneous areas, but also fails to exploit the inherent geometric priors of planar surfaces, often resulting in sub-optimal structural modeling. To address these limitations, we present CompactSplat, a novel feedforward framework that enables a content-aware allocation of 3D Gaussians. 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. Moreover, our framework inherently supports dynamic sparsity control at inference time, enabling seamless trade-offs between rendering fidelity and memory footprint without requiring network retraining. 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.