Metraipolis › archive › dc_45ff2d431359
PhysTrace: Interpretable Temporal Evaluation of Physical Consistency in Video Generative Models — arXiv:2610.31429
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2610.31429
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-23T15:13:24.054Z
- SHA-256 of the archived text
37ef00abc8aefac5eb7a7318d4dac23102d9cfbf4f686f73028cb8a330e2ab1c- Stable URL of this page
https://metraipolis.com/documento/dc_45ff2d431359
How to cite
PhysTrace: Interpretable Temporal Evaluation of Physical Consistency in Video Generative Models — arXiv:2610.31429 — arXiv (en), https://arxiv.org/abs/2610.31429 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_45ff2d431359 (Metraipolis archive dc_45ff2d431359, sha256 37ef00abc8aefac5eb7a7318d4dac23102d9cfbf4f686f73028cb8a330e2ab1c)
BibTeX
@misc{dc45ff2d431359,
title = {PhysTrace: Interpretable Temporal Evaluation of Physical Consistency in Video Generative Models — arXiv:2610.31429},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2610.31429},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_45ff2d431359; sha256 37ef00abc8aefac5eb7a7318d4dac23102d9cfbf4f686f73028cb8a330e2ab1c},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.31429. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
World action models increasingly rely on pretrained video models assumed to capture physical dynamics for prediction, planning, and action generation. Existing evaluations assess plausibility, probe representations, or measure predictive surprise, but rarely identify which history-conditioned physical expectations conflict with observations. We introduce PhysTrace, a training-free framework aligning MLLM-generated, history-conditioned expectations with observed video windows to temporally localize physical violations and provide interpretable descriptions without task-specific probes or classifiers. We also introduce IntPhys2-Timing, adding violation intervals to IntPhys2. PhysTrace achieves a Hit Rate of up to 70.7% across physical principles on IntPhys2-Timing and outperforms the compared methods in overall possible-impossible discrimination accuracy on IntPhys2. Applied to reconstructed videos along the trajectories of pretrained diffusion/flow-based video models, PhysTrace reveals attribute- and noise-dependent patterns in physical violation localization, enabling interpretable physical consistency analysis across timesteps. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno). Categorie arXiv: Artificial Intelligence (cs.AI) ; Robotics (cs.RO). Pagina del lavoro: https://arxiv.org/abs/2610.31429 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-23T15:09:08.459Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).