Research map / 3D vision / Gaussian splatting
Seele: A Unified Acceleration Framework for Real-Time Gaussian Splatting on Mobile Devices
He Zhu, Xiaotong Huang, Zihan Liu, Weikai Lin, Xiaohong Liu, Zhezhi He, Jingwen Leng, Minyi Guo, Yu Feng
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add hybrid preprocessing (view-dependent representation plus online filtering) to code-optimized 3DGS (ablation) | 3DGS on Orin AGX · Mip-NeRF360 | FPS: 21.75 -> 46.15 | Improved |
| add contribution-aware rasterization to code-optimized 3DGS (ablation) | 3DGS on Orin AGX · Mip-NeRF360 | FPS: 21.75 -> 30.10 | Improved |
| add hybrid preprocessing (ablation) | 3DGS on Orin AGX · Mip-NeRF360 | PSNR: 27.46 -> 27.70 | Improved |
| add contribution-aware rasterization (ablation) | 3DGS on Orin AGX · Mip-NeRF360 | PSNR: 27.46 -> 27.50 | No clear effect |
| add code optimizations (fine-grained fetching, overlap) (ablation) | 3DGS on Orin AGX · Mip-NeRF360 | FPS: 20.79 -> 21.75 | Improved |
| fine-tune with consistency loss instead of extra training with the original loss (ablation) | SEELE + 3DGS · Mip-NeRF360 | PSNR: 27.48 -> 27.72 | Improved |
| add Lconsistency (FLIP) to fine-tuning (ablation) | SEELE + 3DGS · Tanks&Temples | FLIP7 view consistency: 0.074 -> 0.067 | Improved |
| add Lconsistency (FLIP) to fine-tuning (ablation) | SEELE + 3DGS · Deep Blending | FLIP7 view consistency: 0.043 -> 0.042 | Improved |
| apply SEELE to four 3DGS pipelines | 3DGS, MiniSplatting, LightGaussian, AdR-Gaussian · Mip-NeRF360, Tanks&Temples, Deep Blending | speedup: 3.2x, 1.8x, 2.7x, 1.7x average speedup | Improved |
| apply SEELE to lower-power GPU (Orin NX) vs workstation GPU | 3DGS · Mip-NeRF360 | FPS: 3.06 -> 9.42 (Orin NX); 134.45 -> 328.98 (A6000) | Improved |
| increase pixel group size beyond 2x4 (ablation) | contribution-aware rasterization · Mip-NeRF360 | PSNR | Hurt |
| use 8 clusters with 8 neighbors (ablation) | hybrid preprocessing · Mip-NeRF360 | PSNR | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- MiniSplatting — Mini-Splatting: Representing Scenes with a Constrained Number of Gaussians (ECCV 2024) · better on 2 of 2 shared results
- LightGaussian — LightGaussian: Unbounded 3D Gaussian Compression with 15x Reduction and 200+ FPS (NeurIPS 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Speedy-Splat: Fast 3D Gaussian Splatting with Sparse Pixels and Sparse Primitives (CVPR 2025 · similarity 0.92)
- DashGaussian: Optimizing 3D Gaussian Splatting in 200 Seconds (CVPR 2025 · similarity 0.92)
- Hardware-Rasterized Ray-Based Gaussian Splatting (CVPR 2025 · similarity 0.91)
- FlashGS: Efficient 3D Gaussian Splatting for Large-scale and High-resolution Rendering (CVPR 2025 · similarity 0.91)
- Steepest Descent Density Control for Compact 3D Gaussian Splatting (CVPR 2025 · similarity 0.90)
- LODGE: Level-of-Detail Large-Scale Gaussian Splatting with Efficient Rendering (NeurIPS 2025 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
