Webinaires

The *3R family: a Foundation Model for (3D) vision

Annonce

Résumé

Unconstrained 3D Reconstruction originally presented in DUSt3R has seen a rapid growth in popularity for its ease of use, its robustness and its many potential applications. In some sense, it revolutionized traditional multi-view stereo reconstruction by casting the problem as simply regressing pointmaps. Because they encode scene geometry, camera calibration and pose, this meant a unification of monocular 3D reconstruction, camera calibration, pose estimation and dense 3D reconstruction. It even showed exciting new capabillities for instance when the two views do not share visual content yet observe the same scene. Overall, DUSt3R unifies and simplifies the complex pipeline of traditional photogrammetry, significantly reducing computational overhead and enhancing performance across various tasks. in this talk I will present our latest advances in pursuit of a general and controllable computer vision network, a.k.a. "foundation model" that can solve as many geometric computer vision tasks as possible, and even more…

Rediffusion chaîne GdR IASIS