Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2510.14564 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866908597579939840 |
|---|---|
| author | Wu, Junyi Xu, Jiaming Li, Jinhao Zhou, Yongkang Pan, Jiayi Li, Xingyang Dai, Guohao |
| author_facet | Wu, Junyi Xu, Jiaming Li, Jinhao Zhou, Yongkang Pan, Jiayi Li, Xingyang Dai, Guohao |
| contents | 3D Gaussian Splatting (3DGS) has emerged as a promising 3D reconstruction technique. The traditional 3DGS training pipeline follows three sequential steps: Gaussian densification, Gaussian projection, and color splatting. Despite its promising reconstruction quality, this conventional approach suffers from three critical inefficiencies: (1) Skewed density allocation during Gaussian densification, (2) Imbalanced computation workload during Gaussian projection and (3) Fragmented memory access during color splatting.
To tackle the above challenges, we introduce BalanceGS, the algorithm-system co-design for efficient training in 3DGS. (1) At the algorithm level, we propose heuristic workload-sensitive Gaussian density control to automatically balance point distributions - removing 80% redundant Gaussians in dense regions while filling gaps in sparse areas. (2) At the system level, we propose Similarity-based Gaussian sampling and merging, which replaces the static one-to-one thread-pixel mapping with adaptive workload distribution - threads now dynamically process variable numbers of Gaussians based on local cluster density. (3) At the mapping level, we propose reordering-based memory access mapping strategy that restructures RGB storage and enables batch loading in shared memory.
Extensive experiments demonstrate that compared with 3DGS, our approach achieves a 1.44$\times$ training speedup on a NVIDIA A100 GPU with negligible quality degradation. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_14564 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | BalanceGS: Algorithm-System Co-design for Efficient 3D Gaussian Splatting Training on GPU Wu, Junyi Xu, Jiaming Li, Jinhao Zhou, Yongkang Pan, Jiayi Li, Xingyang Dai, Guohao Computer Vision and Pattern Recognition 3D Gaussian Splatting (3DGS) has emerged as a promising 3D reconstruction technique. The traditional 3DGS training pipeline follows three sequential steps: Gaussian densification, Gaussian projection, and color splatting. Despite its promising reconstruction quality, this conventional approach suffers from three critical inefficiencies: (1) Skewed density allocation during Gaussian densification, (2) Imbalanced computation workload during Gaussian projection and (3) Fragmented memory access during color splatting. To tackle the above challenges, we introduce BalanceGS, the algorithm-system co-design for efficient training in 3DGS. (1) At the algorithm level, we propose heuristic workload-sensitive Gaussian density control to automatically balance point distributions - removing 80% redundant Gaussians in dense regions while filling gaps in sparse areas. (2) At the system level, we propose Similarity-based Gaussian sampling and merging, which replaces the static one-to-one thread-pixel mapping with adaptive workload distribution - threads now dynamically process variable numbers of Gaussians based on local cluster density. (3) At the mapping level, we propose reordering-based memory access mapping strategy that restructures RGB storage and enables batch loading in shared memory. Extensive experiments demonstrate that compared with 3DGS, our approach achieves a 1.44$\times$ training speedup on a NVIDIA A100 GPU with negligible quality degradation. |
| title | BalanceGS: Algorithm-System Co-design for Efficient 3D Gaussian Splatting Training on GPU |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2510.14564 |