Загрузка видео...

Не удалось загрузить видео

На главную

⚡️ Introducing video world models that know when they don't know! 📽️ Action-conditioned video models can hallucinate a lot, generating videos that look plausible but are incorrect! How can we rigorously quantify their uncertainty, and know when to trust generations?

12,038 просмотров • 11 месяцев назад •via X (Twitter)

Комментарии: 4

Фото профиля Anirudha Majumdar
Anirudha Majumdar11 месяцев назад

❌ Hallucinations hinder trustworthy video generation, e.g., for scalable evaluation of generalist robot policies and visual planning! 🚀 Using proper scoring rules, our approach generates dense confidence predictions at the subpatch level that are physically interpretable and aligned with observations.

Фото профиля Anirudha Majumdar
Anirudha Majumdar11 месяцев назад

Check out the paper and the project website to find out more! 📃 Paper: 🌐 Website:

Фото профиля Decoded Institute
Decoded Institute11 месяцев назад

This is the missing layer: not just generating, but calibrating what the model doesn’t know. Without uncertainty, ‘plausible’ becomes a liability.

Фото профиля Himanshu Kumar
Himanshu Kumar11 месяцев назад

Uncertainty quantification in action-conditioned models requires evaluating calibration metrics; the ECE score is 0.12.

Похожие видео