Research map / 3D vision / Gaussian splatting
Monocular dynamic scene reconstruction
Reconstructs and generates 4D scenes from monocular video using free-moving Gaussian primitives.
80 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2024–2026) · an approach in Gaussian splatting, keyword cluster “dynamic · 4d · motion”
State History Tried Baselines Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
14 papers in 2023–24 and 66 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 11.5% to 9.2%.
Open questions the data points to
- Papers report AP-NeRF on ZJU-MoCap (LPIPS) anywhere from 0.063 to 0.0934: 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 · Deformable 3D Gaussians for High-Fidelity Monocular Dynamic Scene Reconstruction (CVPR 2024 · 420 citations)
- 2025 · MoSca: Dynamic Gaussian Fusion from Casual Videos via 4D Motion Scaffolds (CVPR 2025 · 33 citations)
- 2026 · Mem4D: Decoupling Static and Dynamic Memory for Dynamic Scene Reconstruction (AAAI 2026 · 2 citations)
Earliest papers here
- Deformable 3D Gaussians for High-Fidelity Monocular Dynamic Scene Reconstruction (CVPR 2024 · 420 citations)
- PhysGaussian: Physics-Integrated 3D Gaussians for Generative Dynamics (CVPR 2024 · 165 citations)
- 3DGStream: On-the-Fly Training of 3D Gaussians for Efficient Streaming of Photo-Realistic Free-Viewpoint Videos (CVPR 2024 · 106 citations)
What has been tried
38 papers here report 382 measured changes (a component added, removed or swapped, compared with a baseline): helped 76%, no clear effect 3%, 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 annealing smooth training (AST) | NeRF-DS · PSNR | 23.97 -> 24.11 | Helped | Deformable 3D Gaussians for High-Fidelity Monocular Dynamic Scene Reconstruction (CVPR 2024 · 420 citations) |
| Use deformable covariance instead of translating Gaussian kernels only | Wolf, Stool, and Plant bend/twist lattice benchmarks · PSNR | Fixed Covariance: 26.77, 26.02, 29.94, 25.31, 23.95, 23.09; Ours: 26.96, 26.46, 31.15, 26.15, 25.81, 23.87 | Helped | PhysGaussian: Physics-Integrated 3D Gaussians for Generative Dynamics (CVPR 2024 · 165 citations) |
| remove adaptive 3DG addition (Stage 2) | Flame salmon · PSNR (dB) | 28.39 -> 28.42 | No clear effect | 3DGStream: On-the-Fly Training of 3D Gaussians for Efficient Streaming of Photo-Realistic Free-Viewpoint Videos (CVPR 2024 · 106 citations) |
| Use the 4D embedding instead of per-frame, per-point optimizable parameters | DNA-Rendering, 0013 01 · PSNR | 29.779 -> 31.990 | Helped | 4K4D: Real-Time 4D View Synthesis at 4K Resolution (CVPR 2024 · 65 citations) |
| add rigidity regularization | 30 text prompts · Overall Quality preference | 45.8/13.3 | Helped | Align Your Gaussians: Text-to-4D with Dynamic 3D Gaussians and Composed Diffusion Models (CVPR 2024 · 58 citations) |
| add node control | DyCheck · mPSNR | 19.28 -> 19.32 | Helped | MoSca: Dynamic Gaussian Fusion from Casual Videos via 4D Motion Scaffolds (CVPR 2025 · 33 citations) |
| Initialize from pretrained LGM rather than training without 3D pretraining | Objaverse-4D reconstruction · PSNR | Helped | L4GM: Large 4D Gaussian Reconstruction Model (NeurIPS 2024 · 23 citations) | |
| use the proposed motion representation instead of 4DGS motion representation | SelfCap dance1 (entire/fastest) · PSNR | 28.10/26.92 -> 29.74/30.75 | Helped | FreeTimeGS: Free Gaussian Primitives at Anytime Anywhere for Dynamic Scene Reconstruction (CVPR 2025 · 18 citations) |
| Use adaptive hybrid skinning instead of DQS or LBS | Consistent4D test set · LPIPS | 0.126 -> 0.126 (DQS); 0.155 -> 0.126 (LBS) | Mixed | DreamMesh4D: Video-to-4D Generation with Sparse-Controlled Gaussian-Mesh Hybrid Representation (NeurIPS 2024 · 18 citations) |
| add SDS loss | ZJU-Mocap 377 subject, lower-body occluded · PSNR | 24.51 -> 26.03 | Helped | Guess The Unseen: Dynamic 3D Scene Reconstruction from Partial 2D Glimpses (CVPR 2024 · 12 citations) |
| Use dynamic modeling | Residential split · PSNR | 24.31 -> 28.55 | Helped | Dynamic 3D Gaussian Fields for Urban Areas (NeurIPS 2024 · 10 citations) |
| Add a residual material adaptor rather than using the fixed prior alone | BouncyBall · Chamfer distance | 1.87 -> 1.19 | Helped | NeuMA: Neural Material Adaptor for Visual Grounding of Intrinsic Dynamics (NeurIPS 2024 · 9 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 |
|---|---|---|---|---|
| AP-NeRF | ZJU-MoCap | LPIPS | 0.063 – 0.0934 | 2 papers, e.g. NeurIPS 2024 and CVPR 2025 |
What reviewers will expect
Datasets and baselines named in the 30 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- D-NeRF · 4 of 30 papers
- DAVIS · 3 of 30 papers
- DyCheck · 3 of 30 papers
- PandaSet · 3 of 30 papers
- Waymo Open Dataset · 3 of 30 papers
- Argoverse2 · 3 of 30 papers
- NuScenes · 3 of 30 papers
- DyNeRF · 2 of 30 papers
Baselines
- 4DGS · 10 of 30 papers
- HyperNeRF · 6 of 30 papers
- SC-GS · 5 of 30 papers
- Shape of Motion · 5 of 30 papers
- DeformableGS · 5 of 30 papers
- K-Planes · 4 of 30 papers
- RoDynRF · 4 of 30 papers
- TiNeuVox · 4 of 30 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 80 papers; from OpenAlex and the papers' affiliations).
- Zhejiang University · 5 papers
- NVIDIA · 4 papers
- University of Toronto · 4 papers
- Tsinghua University · 3 papers
- Hong Kong University of Science and Technology · 3 papers
- ETHZ - ETH Zurich · 3 papers
- Snap Inc · 3 papers
- Vector Institute · 2 papers
- Massachusetts Institute of Technology · 2 papers
- National University of Singapore · 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
- Deformable 4D Gaussian splatting (112 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).