AI Converts First-Person Video to Third-Person View
Автор: Bilawal Sidhu
Загружено: 2025-06-16
Просмотров: 17483
Описание:
New CMU AI paper (with code release) called UFM: A Simple Path towards Unified Dense Correspondence with Flow. Really wild to imagine the implications for everything from AR/VR to training data for humanoid robotics.
Paper Link: https://arxiv.org/abs/2506.09278
As one of my amazing followers put it, "For those who don’t understand: this model finds dense pixel-to-pixel matches between two images. You can use it to track 3D motion, align views, or reconstruct 3D scenes from photos or video. We use algorithms like this in things like AR, robotics, vfx, 360 image stiching, etc. it can run near real-time making it the first full transformer based architecture for Flow and Matching in one inference pass." or more simply put "This how I think Johnny Silverhand sees in Cyberpunk 2077"
Повторяем попытку...
Доступные форматы для скачивания:
Скачать видео
-
Информация по загрузке:
