GlobalSplat: Efficient Feed-Forward 3D Gaussian Splatting via Global Scene Tokens
2026-04-16 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors developed GlobalSplat, a new method for representing 3D scenes using fewer and better-placed 3D primitives called Gaussians. Unlike previous methods that add redundant parts tied closely to specific views or pixels, their approach first learns a compact global representation from multiple views together, then creates the 3D scene. This makes the scene smaller, more consistent, and faster to render. Their method works well on known datasets and runs quickly without relying on pretrained pixel-based models.
3D Gaussian Splattingspatial allocationmulti-view synthesislatent representation3D reconstructionnovel-view synthesisrepresentation compactnessinference speedRealEstate10Kglobal consistency
Authors
Roni Itkin, Noam Issachar, Yehonatan Keypur, Yehonatan Keypur, Anpei Chen, Sagie Benaim
Abstract
The efficient spatial allocation of primitives serves as the foundation of 3D Gaussian Splatting, as it directly dictates the synergy between representation compactness, reconstruction speed, and rendering fidelity. Previous solutions, whether based on iterative optimization or feed-forward inference, suffer from significant trade-offs between these goals, mainly due to the reliance on local, heuristic-driven allocation strategies that lack global scene awareness. Specifically, current feed-forward methods are largely pixel-aligned or voxel-aligned. By unprojecting pixels into dense, view-aligned primitives, they bake redundancy into the 3D asset. As more input views are added, the representation size increases and global consistency becomes fragile. To this end, we introduce GlobalSplat, a framework built on the principle of align first, decode later. Our approach learns a compact, global, latent scene representation that encodes multi-view input and resolves cross-view correspondences before decoding any explicit 3D geometry. Crucially, this formulation enables compact, globally consistent reconstructions without relying on pretrained pixel-prediction backbones or reusing latent features from dense baselines. Utilizing a coarse-to-fine training curriculum that gradually increases decoded capacity, GlobalSplat natively prevents representation bloat. On RealEstate10K and ACID, our model achieves competitive novel-view synthesis performance while utilizing as few as 16K Gaussians, significantly less than required by dense pipelines, obtaining a light 4MB footprint. Further, GlobalSplat enables significantly faster inference than the baselines, operating under 78 milliseconds in a single forward pass. Project page is available at https://r-itk.github.io/globalsplat/