Research map / Video / Video segmentation and tracking
Not All Frame Features Are Equal: Video-to-4D Generation via Decoupling Dynamic-Static Features
Liying Yang, Chen Liu, Zhenwei Zhu, Ajian Liu, Hui Ma, Jian Nong, Yanyan Liang
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 point initialization from a large reconstruction model (ablation) | baseline · Consistent4D dataset | FVD: 953.6300 -> 913.371 | Improved |
| add LPIPS loss (ablation) | model B · Consistent4D dataset | FVD: 913.371 -> 899.5714 | Improved |
| add frame features without decoupling (ablation) | model C · Consistent4D dataset | FVD: 899.5714 -> 888.6579 | Improved |
| add DSFD dynamic-static decoupling (ablation) | model D · Consistent4D dataset | FVD: 888.6579 -> 861.6075 | Improved |
| add DSFD dynamic-static decoupling (ablation) | model D · Objaverse dataset | FVD: 1005.7503 -> 989.0834 | Improved |
| add TSSF with average pooling (ablation) | model E (DSFD) · Consistent4D dataset | FVD: 861.6075 -> 839.6600 | Improved |
| replace average pooling with adaptive global-awareness fusion (TSSF-GA) (ablation) | model F · Consistent4D dataset | FVD: 839.6600 -> 799.9367 | Improved |
| use distance-awareness fusion (TSSF-DA) instead of GA (ablation) | DS4D-GA · Consistent4D dataset | FVD: 799.9367 -> 784.0235 | Improved |
| use distance-awareness fusion (TSSF-DA) instead of GA (ablation) | DS4D-GA · Objaverse dataset | FVD: 890.2646 -> 870.9489 | Improved |
| add point initialization (ablation) | baseline · Objaverse dataset | FVD: 1072.1292 -> 1062.9398 | Improved |
| replace STAG4D with DS4D-DA decoupling method | STAG4D · Consistent4D dataset | FVD: 986.8271 -> 784.0235 | Improved |
| replace STAG4D with DS4D-DA on low-quality inputs | STAG4D · Consistent4D (low-quality input images) | FVD: 1311.8770 -> 805.4721 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- STAG4D — STAG4D: Spatial-Temporal Anchored Generative 4D Gaussians (ECCV 2024) · better on 10 of 10 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
- 4D3R: Motion-Aware Neural Reconstruction and Rendering of Dynamic Scenes from Monocular Videos (NeurIPS 2025 · similarity 0.88)
- Real-time Photorealistic Dynamic Scene Representation and Rendering with 4D Gaussian Splatting (ICLR 2024 · similarity 0.88)
- Motion4D: Learning 3D-Consistent Motion and Semantics for 4D Scene Understanding (NeurIPS 2025 · similarity 0.87)
- World from Motion: Generative Dynamic Gaussian Reconstruction from Monocular Video (NeurIPS 2026 · similarity 0.87)
- Motion Decoupled 3D Gaussian Splatting for Dynamic Object Representation (AAAI 2025 · similarity 0.87)
- HiCoM: Hierarchical Coherent Motion for Dynamic Streamable Scenes with 3D Gaussian Splatting (NeurIPS 2024 · similarity 0.86)
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.
