Research map / Video / Video generation
Out of Sight, Out of Mind? Evaluating State Evolution in Video World Models
Ziqi Ma, Mengzhan Liufu, Georgia Gkioxari
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 |
|---|---|---|---|
| apply observation control (occluder, lights off) during the evolving process (ablation) | Veo 3 / Sora 2 Pro · StEvo-Bench | state progress rate (%): 84.6 -> 17.4 | Hurt |
| apply observation control (occluder, lights off) during the evolving process (ablation) | Veo 3 / Sora 2 Pro · StEvo-Bench | task success rate (%): 46.2 -> 12.4 | Hurt |
| camera lookaway control for camera-controlled models | Genie 3, HY-WorldPlay, Lingbot, GEN3C · StEvo-Bench | task success rate (%): 0.0 success for all four; progress at most 3.4 | Hurt |
| add memory module (VMem) to camera-controlled model | VMem · StEvo-Bench (qualitative) | state evolution under lookaway | Hurt |
| add 3D cache priors (GEN3C) to camera control | GEN3C · StEvo-Bench | state progress rate (%): 0.0 progress, 82.4 coherence | Hurt |
| closed-source (Veo 3, Sora 2 Pro) vs open-source video models | general video models · StEvo-Bench | task success rate (%): 0.5-0.9 -> 8.1-8.7 | Improved |
| automatic verifier vs human annotators on physical plausibility | Gemini-based verifier · Verifier agreement (n=180) | accuracy: V-H 0.700 vs H-H 0.722 (physics); V-H 0.795 vs H-H 0.747 (progress) | Mixed |
Closest papers by meaning
- WorldReasonBench: Human-Aligned Stress Testing of Video Generators as Future World-State Predictors (NeurIPS 2026 · similarity 0.88)
- Trimming the Long-Tail of Visual World Modeling Evaluation (NeurIPS 2026 · similarity 0.86)
- WorldModelBench: Judging Video Generation Models As World Models (NeurIPS 2025 · similarity 0.86)
- World Models as Group Actions (NeurIPS 2026 · similarity 0.85)
- stable-worldmodel: A Platform for Reproducible World Modeling Research and Evaluation (NeurIPS 2026 · similarity 0.85)
- Video World Models with Long-term Spatial Memory (NeurIPS 2025 · similarity 0.84)
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.
