Special Talk


Talk Details

  • タイトル: Unsupervised 3D Reconstruction and Grouping of Rigid and Non-Rigid Categories
  • 講演者: Antonio Agudo氏
  • 時間: 1月8日(水) 14時〜14時半(30分)
  • 場所: 西早稲田キャンパス、55号館N棟2階12室


In last two decades, many efforts have been done in the problem of inferring the 3D reconstruction of an unknown shape solely from RGB video. While the problem seems to be mature for the rigid case, when the shape can undergo non-rigid deformations the problem is inherently ill-posed. In this talk, I will present my personal evolution exploring the art of priors to address a wide variety of scenarios and situations, such as those composed of multiple objects, multiple categories, elastic and isometric deformations, to name just a few. Particularly, we will explore unsupervised techniques to jointly retrieve segmentation, motion and shape deformation, all of them, from RGB information, and without assuming any training data at all.