Research map / Video / Video segmentation and tracking
Interpreting Physics in Video World Models
Sonia Joseph, Quentin Garrido, Randall Balestriero, Matthew Kowal, Thomas Fel, Shahab Bakhtiari, Blake Aaron Richards, Michael Rabbat
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 |
|---|---|---|---|
| suppress local attention temporally (t=3) in the Physics Emergence Zone (ablation) | V-JEPA 2 · IntPhys probe | accuracy (%): 78.3 -> 51.9 | Hurt |
| suppress local attention spatially (s=7) in the Physics Emergence Zone (ablation) | V-JEPA 2 · IntPhys probe | accuracy (%): 78.3 -> 62.2 | Hurt |
| suppress local attention spatially (s=7) (ablation) | V-JEPA 2 · Toy ball direction probe | R^2: 0.97 -> 0.93 | No clear effect |
| suppress local attention spatiotemporally (s=3, t=1) (ablation) | V-JEPA 2 · Toy ball direction probe | R^2: 0.97 -> 0.14 | Hurt |
| suppress local attention spatiotemporally (s=3, t=1) (ablation) | V-JEPA 2 · ImageNet probe | accuracy (%): 33.7 -> 33.1 | No clear effect |
| suppress local attention spatially (s=7) (ablation) | V-JEPA 2 · Per-patch direction probe | R^2: 0.72 -> 0.30 | Hurt |
| steer across all probe directions instead of a single probe direction | V-JEPA 2 layer 8 · motion direction steering | mean angular error (degrees): >80 -> <0.5 | Improved |
| use direction subspace versus speed subspace when comparing to IntPhys subspace | V-JEPA 2-L · principal angle | degrees: 69 vs 81 | Mixed |
Closest papers by meaning
- PHANTOM: Physics-Infused Video Generation via Joint Modeling of Visual and Latent Physical Dynamics (CVPR 2026 · similarity 0.89)
- Beyond Static Vision: Scene Dynamic Field Unlocks Intuitive Physics Understanding in Multi-modal Large Language Models (ICLR 2026 · similarity 0.88)
- GEOPHYS: The Geometry of Physical Plausibility (NeurIPS 2026 · similarity 0.88)
- VideoHallu: Evaluating and Mitigating Multi-modal Hallucinations on Synthetic Video Understanding (NeurIPS 2025 · similarity 0.88)
- Physics-Informed Video Generation via Mixture-of-Experts Latent Alignment (NeurIPS 2026 · similarity 0.87)
- VideoREPA: Learning Physics for Video Generation through Relational Alignment with Foundation Models (NeurIPS 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.
