From CroCo to MASt3R - A Paradigm Change in 3D Computer Vision

Watch a 54-minute conference talk exploring groundbreaking advancements in 3D vision technology, focusing on three interconnected frameworks: CroCo, DUSt3R, and MASt3R. Learn how CroCo's self-supervised pre-training framework establishes foundations through cross-view completion, enabling robust 3D geometric representations from multi-viewpoint images. Discover DUSt3R's innovative approach to Dense Unconstrained Stereo 3D Reconstruction, which simplifies traditional multi-view stereo reconstruction by eliminating the need for calibrated cameras. Explore how MASt3R builds upon these technologies by incorporating precise pixel correspondence capabilities. Understand the paradigm shift from developing task-specific solutions to creating unified models capable of handling multiple 3D geometric tasks simultaneously. Gain insights into the practical applications of these frameworks in robotic navigation and cultural heritage preservation, presented by Naver Labs Europe research scientist Vincent Leroy, who brings expertise from his PhD work on Multi-View Stereo Reconstruction and his passion for finding elegant solutions to complex problems.