Research map / Video / Action recognition
WildWorld: A Large-Scale Dataset for Action-Conditioned World Modeling with Explicit State Annotations
Zhen Li, Zian Meng, Chuanhao Li, SHUWEI SHI, Wenshuo Peng, Yuwei Wu, Bo Zheng, Yunde Jia, Kaipeng 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 |
|---|---|---|---|
| fine-tune with ground-truth camera poses (CamCtrl) | Wan2.2-TI2V-5B · WildBench (200 samples) | ATE: 4.63 -> 2.02 | Improved |
| fine-tune with ground-truth camera poses (CamCtrl) | Wan2.2-TI2V-5B · WildBench (200 samples) | Action Following: 53.77 -> 83.46 | Improved |
| condition on skeleton videos (SkelCtrl) | Wan2.2-TI2V-5B · WildBench (200 samples) | State Alignment: 11.29 -> 22.03 | Improved |
| condition on explicit states (StateCtrl) vs CamCtrl | CamCtrl · WildBench (200 samples) | ATE: 2.02 -> 0.94 | Improved |
| condition on explicit states (StateCtrl) vs CamCtrl | CamCtrl · WildBench (200 samples) | Image Quality: 62.88 -> 67.78 | Improved |
| skeleton video conditioning vs soft state embedding | StateCtrl · WildBench (200 samples) | Image Quality: 67.78 -> 62.43 | Hurt |
| skeleton conditioning vs state embedding | StateCtrl · WildBench (200 samples) | Action Following: 85.66 -> 92.81 | Improved |
| autoregressive state prediction instead of ground-truth states | StateCtrl · WildBench (200 samples) | Action Following: 85.66 -> 74.66 | Hurt |
| autoregressive state prediction instead of ground-truth states | StateCtrl · WildBench (200 samples) | State Alignment: 16.06 -> 16.13 | No clear effect |
| any conditioning method | all models · WildBench (200 samples) | Motion Smoothness / Dynamic Degree: all above 95 | No clear effect |
Closest papers by meaning
- World Models as Group Actions (NeurIPS 2026 · similarity 0.88)
- Learning 3D Persistent Embodied World Models (NeurIPS 2025 · similarity 0.88)
- AdaWorld: Learning Adaptable World Models with Latent Actions (ICML 2025 · similarity 0.88)
- Open-World Reinforcement Learning over Long Short-Term Imagination (ICLR 2025 · similarity 0.88)
- Learning Visual Feature-Based World Models via Residual Latent Action (NeurIPS 2026 · similarity 0.87)
- From Zero to Hero: Training-Free Custom Concept Spawning in World Models (NeurIPS 2026 · 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.
