The problem of learning 3D deformable objects from 2D images is an extremely difficult one. The traditional way to learn these things relies on explicit supervision, such as keypoints and templates which restricts their applicability when the object isn’t in a controlled environment like inside a lab.

Quick Read: https://www.marktechpost.com/2021/07/24/researchers-from-oxford-introduces-dove-an-ai-method-that-learns-deformable-3d-objects-by-just-watching-videos-on-youtube/

Paper: https://arxiv.org/pdf/2107.10844.pdf

Play

0:00

0:00

Settings
Fullscreen



Source link