Research map / Video / Video segmentation and tracking
Understanding Object Dynamics for Interactive Image-to-Video Synthesis
Andreas Blattmann, Timo Milbich, Michael Dorkenwald, Bjorn Ommer
Check 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 |
|---|---|---|---|
| Use the interactive hierarchical dynamics model rather than SAVP, IVRNN, or SRVP | video prediction · Poking-Plants, iPER, Tai-Chi-HD, Human3.6m | FVD: PP: 136.81, 184.54, 283.90 -> 89.67; iPER: 150.39, 181.50, 245.13 -> 144.92; Tai-Chi: 309.62, 202.33, 238.71 -> 171.74; | Improved |
| Use the interactive hierarchical dynamics model instead of Hao et al.'s sparse-flow image-warping method | controlled video synthesis · PP, iPER, TaiChi, Human3.6m | FVD: PP: 361.51 -> 137.78; iPER: 235.08 -> 93.28; TaiChi: 341.79 -> 152.76; Human3.6m: 259.92 -> 100.36 | Improved |
| Replace the hierarchical predictor with three GRU cells at the latent bottleneck (ablation) | full interactive image-to-video model · PP and iPER | FVD: PP: 89.67 -> 175.42; iPER: 144.92 -> 250.64 | Hurt |
| Remove trajectory loss L_traj (ablation) | full interactive image-to-video model · PP and iPER | FVD: PP: 89.67 -> 110.00; iPER: 144.92 -> 203.26 | Hurt |
| Skip object-encoder pretraining and train in a single stage (ablation) | full interactive image-to-video model · PP and iPER | FVD: PP: 89.67 -> 115.65; iPER: 144.92 -> 220.82 | Hurt |
| Increase the dynamics hierarchy depth from N=1 to N=2 (ablation) | interactive image-to-video model · PP and iPER | FVD: PP: 170.74 -> 138.72; iPER: 233.22 -> 178.52 | Improved |
| Increase the dynamics hierarchy depth from N=2 to N=3 (ablation) | interactive image-to-video model · PP and iPER | FVD: PP: 138.72 -> 89.67; iPER: 178.52 -> 144.92 | Improved |
Closest papers by meaning
- DeformMaster: An Interactive Physics-Neural World Model for Deformable Objects from Videos (NeurIPS 2026 · similarity 0.89)
- Motion Modes: What Could Happen Next? (CVPR 2025 · similarity 0.89)
- DynaTokens: Teaching Dynamics to Camera-Controlled Video Models at Test Time (NeurIPS 2026 · similarity 0.88)
- ObjectMover: Generative Object Movement with Video Prior (CVPR 2025 · similarity 0.88)
- Generative Image Dynamics (CVPR 2024 · similarity 0.88)
- InterDyn: Controllable Interactive Dynamics with Video Diffusion Models (CVPR 2025 · similarity 0.88)
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.
