Research map / 3D vision / Gaussian splatting
Deformable 4D Gaussian splatting
Models dynamic scenes with deformable or explicit motion Gaussians for real-time rendering.
112 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2024–2026) · an approach in Gaussian splatting, keyword cluster “dynamic · 4d · motion”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 1 of 1 tracked benchmark.
19 papers in 2023–24 and 93 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 15.6% to 12.9%.
Open questions the data points to
- HyperNeRF (PSNR): best went from 26.78 (up to 2024) to 27.78 in 2025 — MoDec-GS: Global-to-Local Motion Decomposition and Temporal Interval Adjustment for Compact Dynamic 3D Gaussian Splatting (CVPR 2025 · 11 citations)
- Papers report 4DGS on HyperNeRF (PSNR) anywhere from 24.89 to 27.44: comparisons against it depend on whose number you take.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2024 · 4D Gaussian Splatting for Real-Time Dynamic Scene Rendering (CVPR 2024 · 737 citations)
- 2025 · WildGS-SLAM: Monocular Gaussian Splatting SLAM in Dynamic Environments (CVPR 2025 · 31 citations)
- 2026 · D-FCGS: Feedforward Compression of Dynamic Gaussian Splatting for Free-Viewpoint Videos (AAAI 2026 · 4 citations)
Earliest papers here
- 4D Gaussian Splatting for Real-Time Dynamic Scene Rendering (CVPR 2024 · 737 citations)
- DrivingGaussian: Composite Gaussian Splatting for Surrounding Dynamic Autonomous Driving Scenes (CVPR 2024 · 223 citations)
- Spacetime Gaussian Feature Splatting for Real-Time Dynamic View Synthesis (CVPR 2024 · 186 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
HyperNeRF · PSNR (higher is better)
4 papers, 2024–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2024 | 26.78 | Superpoint Gaussian Splatting for Real-Time High-Fidelity Dynamic Scene Reconstruction (ICML 2024 · 0 citations) |
| 2025 | 27.78 | MoDec-GS: Global-to-Local Motion Decomposition and Temporal Interval Adjustment for Compact Dynamic 3D Gaussian Splatting (CVPR 2025 · 11 citations) |
What has been tried
53 papers here report 572 measured changes (a component added, removed or swapped, compared with a baseline): helped 75%, no clear effect 4%, hurt 19%, mixed 2%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add HexPlane encoder Rl(i, j) | Synthesis dataset · PSNR (dB) | 27.05 -> 34.05 | Helped | 4D Gaussian Splatting for Real-Time Dynamic Scene Rendering (CVPR 2024 · 737 citations) |
| Use LiDAR prior initialization instead of SfM points | nuScenes · PSNR | 28.36 -> 28.74 | Helped | DrivingGaussian: Composite Gaussian Splatting for Surrounding Dynamic Autonomous Driving Scenes (CVPR 2024 · 223 citations) |
| add temporal opacity | Technicolor Dataset · PSNR | 31.0 -> 33.6 | Helped | Spacetime Gaussian Feature Splatting for Real-Time Dynamic View Synthesis (CVPR 2024 · 186 citations) |
| Add sparse control points | D-NeRF · PSNR | 38.512 -> 43.307 | Helped | SC-GS: Sparse-Controlled Gaussian Splatting for Editable Dynamic Scenes (CVPR 2024 · 170 citations) |
| Add KNN rigid regularization and time smoothness regularization together | HyperNeRF vrig · PSNR | full model 29.92 vs 28.48 without KNN rigid and 29.12 without time smooth | Helped | Gaussian-Flow: 4D Reconstruction with Dynamic 3D Gaussian Particle (CVPR 2024 · 117 citations) |
| Add the unicycle model and optimize object poses rather than using noisy boxes directly | KITTI 5% noise · PSNR | 23.83 -> 28.78 | Helped | HUGS: Holistic Urban 3D Scene Understanding via Gaussian Splatting (CVPR 2024 · 75 citations) |
| Add geometry-aware features to the canonical field | Synthetic dataset ablation average · PSNR | 37.5757 -> 38.0134 | Helped | 3D Geometry-Aware Deformable Gaussian Splatting for Dynamic View Synthesis (CVPR 2024 · 46 citations) |
| add the coarse ED-graph prior | Compact 4D Gaussians · PSNR | 29.142 -> 35.085 | Helped | HiFi4G: High-Fidelity Human Performance Rendering via Compact Gaussian Splatting (CVPR 2024 · 43 citations) |
| add Lnorm | Synthetic dynamic scenes · PSNR | 37.41 -> 37.90 | Helped | CoGS: Controllable Gaussian Splatting (CVPR 2024 · 34 citations) |
| add 4D rotation instead of restricting space and time to be independent | Flame Salmon and Cut Roasted Beef (average) · PSNR | 30.79 -> 31.62 | Helped | Real-time Photorealistic Dynamic Scene Representation and Rendering with 4D Gaussian Splatting (ICLR 2024 · 33 citations) |
| add an uncertainty mask | Wild-SLAM MoCap · ATE RMSE (cm) | 3.89 -> 0.46 | Helped | WildGS-SLAM: Monocular Gaussian Splatting SLAM in Dynamic Environments (CVPR 2025 · 31 citations) |
| add stage I motion masks | Waymo Open Dataset · PSNR | 34.7063 -> 35.7598 | Helped | DeSiRe-GS: 4D Street Gaussians for Static-Dynamic Decomposition and Surface Reconstruction for Urban Driving Scenes (CVPR 2025 · 21 citations) |
Baseline numbers that disagree
The same baseline on the same benchmark and metric, reported more than 5% apart by different papers here. Training budgets, data splits and reimplementations differ, so state which number you compare with.
| Baseline | Benchmark | Metric | Reported | Papers |
|---|---|---|---|---|
| 4DGS | HyperNeRF | PSNR | 24.89 – 27.44 | 3 papers, e.g. CVPR 2025 and CVPR 2025 |
What reviewers will expect
Datasets and baselines named in the 42 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- HyperNeRF · 7 of 42 papers
- DAVIS · 5 of 42 papers
- NeRF-DS · 3 of 42 papers
- D-NeRF · 3 of 42 papers
- DyCheck · 3 of 42 papers
- Mip-NeRF 360 · 3 of 42 papers
- N3DV · 2 of 42 papers
- MeetRoom · 2 of 42 papers
Baselines
- 4DGS · 16 of 42 papers
- 3DGS · 8 of 42 papers
- SC-GS · 7 of 42 papers
- 3DGStream · 5 of 42 papers
- Deformable 3DGS · 5 of 42 papers
- D3DGS · 5 of 42 papers
- K-Planes · 4 of 42 papers
- HyperNeRF · 4 of 42 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 112 papers; from OpenAlex and the papers' affiliations).
- Peking University · 8 papers
- Peng Cheng Laboratory · 4 papers
- National University of Singapore · 4 papers
- Zhejiang University · 4 papers
- Fudan University · 3 papers
- Google · 2 papers
- Case Western Reserve University · 2 papers
- Hong Kong University of Science and Technology · 2 papers
- ETH Zurich · 2 papers
- Microsoft · 2 papers
Use it
Download BibTeX of 20 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Monocular dynamic scene reconstruction (80 papers)
- Animatable Gaussian human avatars (14 papers)
Elsewhere in the topic
- Compact and efficient Gaussian splatting (269 papers)
- Regularization and primitive design for 3DGS (213 papers)
- Sparse-view Gaussian splatting for novel views (111 papers)
- Animatable Gaussian human avatars (42 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).