Research map / Video / Video generation
MemoBench: Benchmarking World Modeling in Dynamically Changing Environments
Haoyu Chen, Kaichen Zhou, Hang Hua, Kaile Zhang, Jingwen Qian, Wufei Ma, Haonan Chen, Chunjiang Liu, Yizhou Zhao, Xiaoyuan Wang, Weiyue Li, Alan Yuille and 2 more
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 |
|---|---|---|---|
| use the paraphrase 'the <subject>' instead of the original subject phrase for SAM-3 prompt (ablation) | LingBot-World ORS · ORS prompt sweep (30 clips) | mean ORS: 0.462 -> 0.459 | No clear effect |
| use truncated prompt (first word only) instead of original subject phrase (ablation) | LingBot-World ORS · ORS prompt sweep (30 clips) | mean ORS: 0.462 -> 0.030 | Hurt |
| use 'a photo of <subject>' prompt instead of original phrase (ablation) | LingBot-World ORS · ORS prompt sweep (30 clips) | mean ORS: 0.462 -> 0.168 | Hurt |
| vary coverage filter thresholds for mask filtering (ablation) | LingBot-World ORS · ORS coverage sweep (30 clips) | mean ORS: 0.456--0.469 | No clear effect |
| vary coverage filter thresholds for mask filtering (ablation) | Stable Virtual Camera ORS · ORS coverage sweep (30 clips) | mean ORS: 0.325--0.345 | No clear effect |
| increase target object size from small to large (ablation) | LingBot-World ORS · MemoBench ORS by object size | ORS: 0.31 -> 0.67 | Improved |
| require reappearance after extreme rotation (>120 deg) (ablation) | Stable Virtual Camera ORS · MemoBench ORS by rotation | ORS: 0.005 | Hurt |
| restrict evaluation to clips with >=90 deg camera rotation (motion gating) (ablation) | Stable Virtual Camera · MemoBench motion-gated (80 clips) | ORS vs CamCtrl: 92.43 CamCtrl, 0.012 ORS | Mixed |
| restrict evaluation to clips with >=90 deg camera rotation (motion gating) (ablation) | LingBot-World · MemoBench motion-gated (80 clips) | CamCtrl and ORS: CamCtrl 75.04, ORS 0.281 | Mixed |
| add explicit camera conditioning (FantasyWorld vs LingBot-World) | C+TI2V models · MemoBench ORS | ORS: 0.276 vs 0.381 | No clear effect |
| camera inactivity in TI2V model inflates scores | LTX-Video · MemoBench | ORS: 0.330 | Mixed |
Closest papers by meaning
- WorldMem: Long-term Consistent World Simulation with Memory (NeurIPS 2025 · similarity 0.89)
- Video World Models with Long-term Spatial Memory (NeurIPS 2025 · similarity 0.89)
- Can 4D Foundation Models Remember? (NeurIPS 2026 · similarity 0.89)
- Addressable Memory for Video World Models (NeurIPS 2026 · similarity 0.87)
- WorldModelBench: Judging Video Generation Models As World Models (NeurIPS 2025 · similarity 0.87)
- Trimming the Long-Tail of Visual World Modeling Evaluation (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.
