Research map / Video / Video generation
DIVER: Disentangling Camera–Object and Active–Passive Motion for Video Generation
Shaowei Liu, Xuanchi Ren, Tianchang Shen, Huan Ling, Saurabh Gupta, Shenlong Wang, Sanja Fidler, Jun Gao
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 canonical fixed-view branch (dual-stream) (ablation) | MoRight · Cooking | EPE: 14.30 -> 4.27 | Improved |
| add canonical fixed-view branch (dual-stream) (ablation) | MoRight · Cooking | FID: 51.17 -> 39.94 | Improved |
| add active/passive motion decomposition (ablation) | MoRight · Cooking | FID: 44.04 -> 39.94 | Improved |
| add active/passive motion decomposition (ablation) | MoRight · Cooking | FVD: 784.19 -> 730.46 | Improved |
| single-pass dual-stream instead of cascaded static-camera plus camera control (ablation) | MoRight · Cooking | translation error: 11.50 -> 10.11 | Improved |
| add mixed supervision with single-view real data (ablation) | MoRight · Cooking | rotation error: 2.22 -> 2.16 | Improved |
| add mixed supervision with single-view real data (ablation) | MoRight · Cooking | FVD: 808.96 -> 730.46 | Improved |
| use coarse segment-level trajectories instead of pixel tracks (ablation) | MoRight · Cooking | FID: 39.83 vs 39.94 | No clear effect |
| use passive motion input instead of active (ablation) | MoRight · Cooking | EPE: 4.27 -> 7.27 | Mixed |
| MoRight with first-frame trajectories vs WanMove with privileged tracks | MoRight · DynPose-100K | PSNR: 13.91 -> 12.30 | Hurt |
| MoRight vs WanMove on EPE | MoRight · DynPose-100K | EPE: 8.05 -> 7.64 | Mixed |
| MoRight vs baselines with only active motion description | MoRight · WISA | FID: 57.29 -> 52.95 | 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.
- WanMove — Wan-Move: Motion-controllable Video Generation via Latent Trajectory Guidance (NeurIPS 2025) · better on 8 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
- DynaTokens: Teaching Dynamics to Camera-Controlled Video Models at Test Time (NeurIPS 2026 · similarity 0.89)
- DisMo: Disentangled Motion Representations for Open-World Motion Transfer (NeurIPS 2025 · similarity 0.89)
- Wan-Move: Motion-controllable Video Generation via Latent Trajectory Guidance (NeurIPS 2025 · similarity 0.89)
- DiT4DiT: Jointly Modeling Video Dynamics and Actions for Generalizable Robot Control (NeurIPS 2026 · similarity 0.88)
- VideoJAM: Joint Appearance-Motion Representations for Enhanced Motion Generation in Video Models (ICML 2025 · similarity 0.88)
- Through-The-Mask: Mask-based Motion Trajectories for Image-to-Video Generation (CVPR 2025 · similarity 0.87)
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.
