English

TGSFormer: Scalable Temporal Gaussian Splatting for Embodied Semantic Scene Completion

Computer Vision and Pattern Recognition 2025-12-02 v1

Abstract

Embodied 3D Semantic Scene Completion (SSC) infers dense geometry and semantics from continuous egocentric observations. Most existing Gaussian-based methods rely on random initialization of many primitives within predefined spatial bounds, resulting in redundancy and poor scalability to unbounded scenes. Recent depth-guided approach alleviates this issue but remains local, suffering from latency and memory overhead as scale increases. To overcome these challenges, we propose TGSFormer, a scalable Temporal Gaussian Splatting framework for embodied SSC. It maintains a persistent Gaussian memory for temporal prediction, without relying on image coherence or frame caches. For temporal fusion, a Dual Temporal Encoder jointly processes current and historical Gaussian features through confidence-aware cross-attention. Subsequently, a Confidence-aware Voxel Fusion module merges overlapping primitives into voxel-aligned representations, regulating density and maintaining compactness. Extensive experiments demonstrate that TGSFormer achieves state-of-the-art results on both local and embodied SSC benchmarks, offering superior accuracy and scalability with significantly fewer primitives while maintaining consistent long-term scene integrity. The code will be released upon acceptance.

Keywords

Cite

@article{arxiv.2512.00300,
  title  = {TGSFormer: Scalable Temporal Gaussian Splatting for Embodied Semantic Scene Completion},
  author = {Rui Qian and Haozhi Cao and Tianchen Deng and Tianxin Hu and Weixiang Guo and Shenghai Yuan and Lihua Xie},
  journal= {arXiv preprint arXiv:2512.00300},
  year   = {2025}
}

Comments

14 pages, 10 figures

R2 v1 2026-07-01T08:00:30.474Z