MV-RoMa: From Pairwise Matching into Multi-View Track Reconstruction
arXiv:2603.27542v1 Announce Type: new
Abstract: Establishing consistent correspondences across images is essential for 3D vision tasks such as structure-from-motion (SfM), yet most existing matchers operate in a pairwise manner, often producing fragme…