Research map / 3D vision / Gaussian splatting
MoE-GS: Mixture of Experts for Dynamic Gaussian Splatting
In-Hwan Jin, Hyeongju Mun, Joonsoo Kim, Kugjin Yun, Kyeongbo Kong
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 |
|---|---|---|---|
| blend multiple heterogeneous dynamic GS experts with a router (N=3 vs best single expert E-D3DGS) | MoE-GS · N3V average | PSNR: 32.33 -> 33.23 | Improved |
| blend experts on Technicolor (N=3 vs best baseline STG) | MoE-GS · Technicolor average | PSNR: 33.69 -> 34.55 | Improved |
| increase experts from 2 to 3 (ablation) | MoE-GS · N3V average | PSNR: 32.54 -> 33.23 | Improved |
| increase experts from 3 to 4 (ablation) | MoE-GS · N3V average | PSNR: 33.23 -> 33.27 | No clear effect |
| use Volume-aware Pixel Router instead of Pixel Router (ablation) | MoE-GS (N=3) · N3V | PSNR: 31.12 -> 33.23 | Improved |
| use Volume-aware Pixel Router instead of Volume Router (ablation) | MoE-GS (N=3) · N3V | PSNR: 32.05 -> 33.23 | Improved |
| add single-pass multi-expert rendering (ablation) | MoE-GS (N=3) · N3V | FPS: 40 -> 68 | Improved |
| add gate-aware Gaussian pruning (ablation) | MoE-GS (N=3) · N3V | FPS and memory: FPS 60 -> 68; memory 747 -> 270 | Improved |
| prune ~75% of Gaussians instead of ~55% (ablation) | MoE-GS (N=2) · N3V | PSNR: 32.80 -> 32.45 | Hurt |
| reduce each expert's training budget to 20% (ablation) | MoE-GS (N=3) · N3V | PSNR: 33.23 -> 32.60 | Hurt |
| distill MoE-GS outputs into individual experts with router weighting (ablation) | E-D3DGS, STG, Ex4DGS · Technicolor | PSNR: E-D3DGS 32.88 -> 33.67; STG 32.83 -> 33.10; Ex4DGS 33.57 -> 33.91 | Improved |
| add router-weight-based supervision to distillation (ablation) | E-D3DGS distillation · Technicolor average | PSNR: 33.26 -> 33.67 | Improved |
Closest papers by meaning
- Speedy-Splat: Fast 3D Gaussian Splatting with Sparse Pixels and Sparse Primitives (CVPR 2025 · similarity 0.92)
- Efficient Gaussian Splatting for Monocular Dynamic Scene Rendering via Sparse Time-Variant Attribute Modeling (AAAI 2025 · similarity 0.91)
- Swift4D: Adaptive divide-and-conquer Gaussian Splatting for compact and efficient reconstruction of dynamic scene (ICLR 2025 · similarity 0.91)
- Generative Sparse-View Gaussian Splatting (CVPR 2025 · similarity 0.90)
- Fully Explicit Dynamic Gaussian Splatting (NeurIPS 2024 · similarity 0.90)
- Superpoint Gaussian Splatting for Real-Time High-Fidelity Dynamic Scene Reconstruction (ICML 2024 · 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.
