Research map / Video / Video generation
Learning to Generate Rigid Body Interactions with Video Diffusion Models
David Romero, Ariana Bermudez, Viacheslav Iablochnikov, Hao Li, Fabio Pizzati, Ivan Laptev
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 first-stage training with full masks before second-stage dropout (ablation) | KineMask · Simple Motion | FVD: 288.8 -> 160.3 | Improved |
| add first-stage training with full masks (ablation) | KineMask · Simple Motion | IoU: 0.237 -> 0.367 | Improved |
| train KineMask vs zero-shot CogVideoX | CogVideoX · Interactions | FVD: 807.3 -> 250.7 | Improved |
| train on Interactions instead of Simple Motion (ablation) | KineMask · Interactions | FVD: 301.1 -> 250.7 | Improved |
| KineMask on Simple Motion vs plain CogVideoX finetuned on Interactions | CogVideoX · Interactions | FVD: 368.2 -> 301.1 | Improved |
| train with rich captions c instead of generic prompt (ablation) | KineMask · Interactions | IoU / FVD / FMVD: IoU 0.356 -> 0.376; FVD 238.8 -> 231.3; FMVD 161.3 -> 174.4 | Mixed |
| apply KineMask to Wan2.2-5B | Wan2.2-5B · Interactions | FVD: 977.7 -> 475.6 | Improved |
| apply KineMask to Cosmos2.5-2B | Cosmos2.5-2B · Interactions | FVD: 1287.3 -> 433.6 | Improved |
| encode per-frame instantaneous velocity v_t instead of v0 (ablation) | KineMask · Interactions | FVD: 277.3 -> 231.3 | Improved |
| use non-uniform dropout sampling biased to interaction frames (ablation) | KineMask · Interactions | FVD: 303.1 -> 231.3 | Improved |
| KineMask vs finetuned CogVideoX on real Physics-IQ | CogVideoX · Physics-IQ | FVD: 357.1 -> 212.1 | Improved |
| KineMask vs Force-Prompting on Physics-IQ | Force-Prompting · Physics-IQ | MSE: 27.76 -> 39.67 | Hurt |
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.
- TORA — Tora: Trajectory-oriented Diffusion Transformer for Video Generation (CVPR 2025) · better on 2 of 2 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
- DiT4DiT: Jointly Modeling Video Dynamics and Actions for Generalizable Robot Control (NeurIPS 2026 · similarity 0.89)
- Learn the Force We Can: Enabling Sparse Motion Control in Multi-Object Video Generation (AAAI 2024 · similarity 0.89)
- SViMo: Synchronized Diffusion for Video and Motion Generation in Hand-object Interaction Scenarios (NeurIPS 2025 · similarity 0.88)
- Motion Prompting: Controlling Video Generation with Motion Trajectories (CVPR 2025 · similarity 0.88)
- DiffGrasp: Whole-Body Grasping Synthesis Guided by Object Motion Using a Diffusion Model (AAAI 2025 · similarity 0.88)
- DeformMaster: An Interactive Physics-Neural World Model for Deformable Objects from Videos (NeurIPS 2026 · 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.
