AsySplat: Efficient Asymmetric 3D Gaussian Splatting for Long-Sequence Scene Modeling

2026-07-12

AI Summary

Existing general-purpose 3D Gaussian splatting models suffer from severe computational redundancy and inefficiency in long-sequence novel view synthesis. This work proposes a task-aware asymmetric dual-branch architecture that decouples geometry and appearance modeling: a geometry branch processes coarse-grained tokens for efficient multi-view structure reconstruction, while an appearance branch handles fine-grained tokens to capture detailed visual characteristics, with bilateral connections enabling mutual guidance between the two branches. Motivated by the observation that high geometric precision is not always necessary and appearance learning is comparatively easier, this design significantly improves parameter efficiency and reduces computational overhead. Under 960p input with 32 views, the method achieves superior zero-shot performance over current state-of-the-art general models with fewer parameters and attains nearly an 800× speedup in inference compared to optimization-based approaches, substantially lowering both training and inference costs.

Abstract

Recent generalizable 3D Gaussian Splatting models have advanced long-sequence novel view synthesis (NVS), but at the cost of substantial redundant computation. We identify that the redundancy can be mitigated based on two observations: (i) high-precision geometry is not strictly required for high-quality NVS; (ii) appearance learning is generally easier than geometry recovery. Motivated by these insights, we propose an asymmetric architecture that decouples geometry and appearance modeling. The geometry branch processes coarse-grained tokens with most of the parameters for multi-view reconstruction, while the appearance branch operates on fine-grained tokens to capture details using significantly fewer parameters. The two branches interact through bilateral connections, enabling mutual guidance for their respective tasks. This task-aware asymmetry reduces the computational redundancy and allocates the computation more judiciously, thereby increasing parameter efficiency and enabling smaller models to achieve strong performance. On 32-view 960P inputs, our model matches optimization-based methods while delivering nearly 800x speedup, and surpasses the zero-shot performance of state-of-the-art generalizable models with markedly fewer parameters and reduced training/inference overhead, achieving an overall efficiency improvement.

Problem

• 3D Gaussian Splatting
• novel view synthesis
• computational redundancy
• long-sequence scene modeling
• parameter efficiency

Innovation

• Asymmetric Architecture
• 3D Gaussian Splatting
• Novel View Synthesis
• Parameter Efficiency
• Geometry-Appearance Decoupling