Latent video models show robust world modeling capabilities

A new study systematically evaluates four frontier video foundation models, V-JEPA 2.1, V-JEPA 2, VideoPrism, and VideoMAEv2, across five robustness axes relevant to their use as world models. The research finds that latent-prediction models consistently outperform others in feature discriminability, corruption robustness, fine-grained discrimination, occlusion robustness, and temporal direction encoding. Notably, a frozen V-JEPA 2 backbone demonstrated superior robustness on corruption and occlusion tasks compared to fully fine-tuned models, suggesting latent prediction's advantages for robust world modeling. AI

影响 Latent prediction models demonstrate superior robustness for world modeling, potentially influencing future AI development in video understanding and simulation.

排序理由 Academic paper presenting a systematic study and evaluation of video foundation models. [lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.CV 阅读 →

• arXiv
• VideoMAEv2
• VideoPrism
• V-JEPA 2
• V-JEPA 2.1

• 论文
• 模型发布

AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →