Research map / Video / Video generation
4DVD: Cascaded Dense-view Video Diffusion Model for High-quality 4D Content Generation
Shuzhou Yang, Xiaodong Cun, Xiaoyu Li, Yaowei Li, Jian Zhang
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 stage-2 structure-aware conditional generation to stage 1 (ablation) | 4DVD · D-Objaverse multi-view videos | FVD-F: 668.25 -> 507.12 | Improved |
| add stage-2 structure-aware generation to stage 1 (ablation) | 4DVD · D-Objaverse multi-view videos | FVD-V: 422.56 -> 314.44 | Improved |
| use proposed control branch instead of ControlNet (ablation) | 4DVD · D-Objaverse multi-view videos | FVD-F: 558.42 -> 507.12 | Improved |
| use proposed control branch instead of T2I-adapter (ablation) | 4DVD · D-Objaverse multi-view videos | FVD-F: 802.40 -> 507.12 | Improved |
| T2I-adapter injection on top of stage 1 only (ablation) | Stage 1 · D-Objaverse multi-view videos | LPIPS: 0.147 -> 0.173 | Hurt |
| add ControlNet condition branch on top of stage 1 (ablation) | Stage 1 · D-Objaverse multi-view videos | FVD-F: 668.25 -> 558.42 | Improved |
| use reference video as Q and coarse features as KV in MAP (ablation) | 4DVD · D-Objaverse multi-view videos | FVD-V: 397.36 -> 314.44 | Improved |
| use reference video as Q in MAP instead of model feature (ablation) | 4DVD · D-Objaverse multi-view videos | CLIP-S: 0.912 -> 0.927 | Improved |
| cascade coarse-to-fine generation with 10 sampling steps per stage instead of single-stage 20 steps | 4DVD vs SV4D · Runtime | seconds: 565 -> 381 | Improved |
| generate 16 dense views instead of 8 views | 4DVD vs SV4D · D-Objaverse 4D assets | FVD: 677.56 -> 438.41 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- SV4D — SV4D: Dynamic 3D Content Generation with Multi-Frame and Multi-View Consistency (ICLR 2025) · better on 8 of 8 shared results
- 4Diffusion — 4Diffusion: Multi-view Video Diffusion Model for 4D Generation (NeurIPS 2024) · better on 4 of 4 shared results
- STAG4D — STAG4D: Spatial-Temporal Anchored Generative 4D Gaussians (ECCV 2024) · better on 4 of 4 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
- SV4D: Dynamic 3D Content Generation with Multi-Frame and Multi-View Consistency (ICLR 2025 · similarity 0.93)
- 4Diffusion: Multi-view Video Diffusion Model for 4D Generation (NeurIPS 2024 · similarity 0.93)
- Diffusion$^2$: Dynamic 3D Content Generation via Score Composition of Video and Multi-view Diffusion Models (ICLR 2025 · similarity 0.92)
- Diffusion4D: Fast Spatial-temporal Consistent 4D generation via Video Diffusion Models (NeurIPS 2024 · similarity 0.92)
- CAT4D: Create Anything in 4D with Multi-View Video Diffusion Models (CVPR 2025 · similarity 0.91)
- EG4D: Explicit Generation of 4D Object without Score Distillation (ICLR 2025 · similarity 0.91)
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.
