Research map / Video / Video generation
MV-Forcing: Long Multi-View Video Generation via 4D-Grounded Spatio-Temporal Self-Forcing
Gal Fiebelman, Hadar Averbuch-Elor, Sagie Benaim
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 view-sequential spatio-temporal self-forcing unrolling (ablation) | MV-Forcing · Synthetic long (3 views, 162 frames) | FVD-V: 2318.94 -> 1729.35 | Improved |
| add view-sequential spatio-temporal self-forcing unrolling (ablation) | MV-Forcing · Synthetic long (3 views, 162 frames) | FVD: 1791.40 -> 1560.54 | Improved |
| add CUT3R geometric prior (ablation) | MV-Forcing · Synthetic long (3 views, 162 frames) | FVD-V: 2109.82 -> 1729.35 | Improved |
| add CUT3R geometric prior (ablation) | MV-Forcing · Synthetic long (3 views, 162 frames) | Mat.Pix(K): 173.75 -> 243.63 | Improved |
| add state accumulation over all previous views (ablation) | MV-Forcing · Synthetic long (3 views, 162 frames) | FVD-V: 1984.35 -> 1729.35 | Improved |
| use CUT3R latent state query instead of manual z-buffer rendering (ablation) | MV-Forcing · Synthetic long (3 views, 162 frames) | FVD-V: 1963.42 -> 1729.35 | Improved |
| use MV-Forcing instead of SF+ReCamMaster+SF | MV-Forcing · Real long (3 views, 162 frames) | FVD-V: 1771.34 -> 1554.23 | Improved |
| use MV-Forcing instead of SF(ft)+ReCamMaster | MV-Forcing · Synthetic long (3 views, 162 frames) | RotErr: 4.32 -> 3.72 | Improved |
| distill into few-step causal student versus bidirectional SynCamMaster teacher | MV-Forcing · Short (2 views, 81 frames) | FVD: 1451.44 -> 1468.84 | Hurt |
| distill into few-step causal student versus bidirectional SynCamMaster teacher | MV-Forcing · Short (2 views, 81 frames) | FVD-V: 1697.37 -> 1691.05 | Improved |
| increase number of generated views from 2 to 5 | MV-Forcing · View scaling (81 frames) | FVD: 1468.84 -> 1469.12 | No clear effect |
| increase temporal length from 81 to 648 frames | MV-Forcing · Temporal scaling (2 views) | FVD: 1468.84 -> 1859.82 | Hurt |
Closest papers by meaning
- Vivid-ZOO: Multi-View Video Generation with Diffusion Model (NeurIPS 2024 · similarity 0.90)
- NAMVIS: Next-Scale Autoregressive Multi-View Image Synthesis (NeurIPS 2026 · similarity 0.90)
- Diffusion$^2$: Dynamic 3D Content Generation via Score Composition of Video and Multi-view Diffusion Models (ICLR 2025 · similarity 0.90)
- 4Diffusion: Multi-view Video Diffusion Model for 4D Generation (NeurIPS 2024 · similarity 0.89)
- MVGenMaster: Scaling Multi-View Generation from Any Image via 3D Priors Enhanced Diffusion Model (CVPR 2025 · similarity 0.89)
- SynCamMaster: Synchronizing Multi-Camera Video Generation from Diverse Viewpoints (ICLR 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.
