Research map / 3D vision / Gaussian splatting
MoRel: Long-Range Flicker-Free 4D Motion Modeling via Anchor Relay-based Bidirectioanl Blending with Hierarchical Densification
Sangwoon Kwak, Weeyoung Kwon, Jun Young Jeong, Geonho Kim, Won-Sik Cheong, Jihyong Oh
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 key-frame anchors (local canonical spaces) to a single global anchor model (ablation) | MoRel baseline · SelfCap_LR 300-frame subset | LPIPS: 0.386 -> 0.364 | Improved |
| add key-frame anchors with on-demand loading (ablation) | MoRel baseline · SelfCap_LR 300-frame subset | training memory (MB): ~12,000 -> ~4,500 | Improved |
| add key-frame anchors (ablation) | MoRel baseline · SelfCap_LR 300-frame subset | SSIM: 0.654 -> 0.647 | Hurt |
| add PWD and linear blending between adjacent KfAs (ablation) | MoRel · SelfCap_LR 300-frame subset | PSNR: 19.90 -> 20.66 | Improved |
| replace linear blending with learnable temporal opacity (IFB) (ablation) | MoRel · SelfCap_LR 300-frame subset | PSNR: 20.66 -> 21.07 | Improved |
| add feature-variance-guided hierarchical densification (FHD) (ablation) | MoRel · SelfCap_LR 300-frame subset | PSNR: 21.07 -> 21.20 PSNR; LPIPS 0.342 -> 0.348 | Mixed |
| use chunk-based training of 4DGS instead of all-at-once | 4DGS · SelfCap_LR | tOF: 0.222 -> 0.680 | Hurt |
| use MoRel anchor relay and bidirectional blending versus all-at-once 4DGS | 4DGS · SelfCap_LR (average) | PSNR: 18.95 -> 21.00 | Improved |
| use MoRel on a standard short benchmark | MoRel · DyNeRF | PSNR: 31.75 vs 31.06 | No clear effect |
| use multi-stage MoRel training instead of single-stage 4DGS | 4DGS · SelfCap_LR Bike_1 | training time (h): 10h -> 16.5h | 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.
- LocalDyGS — LocalDyGS: Multi-view Global Dynamic Scene Modeling via Adaptive Local Implicit Feature Decoupling (ICCV 2025) · better on 3 of 3 shared results
- MoDec-GS — MoDec-GS: Global-to-Local Motion Decomposition and Temporal Interval Adjustment for Compact Dynamic 3D Gaussian Splatting (CVPR 2025) · better on 2 of 2 shared results
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
- Temporal Smoothness-Aware Rate-Distortion Optimized 4D Gaussian Splatting (NeurIPS 2025 · similarity 0.91)
- HiCoM: Hierarchical Coherent Motion for Dynamic Streamable Scenes with 3D Gaussian Splatting (NeurIPS 2024 · similarity 0.90)
- MoDec-GS: Global-to-Local Motion Decomposition and Temporal Interval Adjustment for Compact Dynamic 3D Gaussian Splatting (CVPR 2025 · similarity 0.90)
- Real-time Photorealistic Dynamic Scene Representation and Rendering with 4D Gaussian Splatting (ICLR 2024 · similarity 0.90)
- 1000+ FPS 4D Gaussian Splatting for Dynamic Scene Rendering (NeurIPS 2025 · similarity 0.89)
- HAIF-GS: Hierarchical and Induced Flow-Guided Gaussian Splatting for Dynamic Scene (NeurIPS 2025 · similarity 0.89)
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.
