Description
Reconstructing a 3D scene from photos with classic structure-from-motion is slow and often wrong in scale. MapAnything is Meta's universal feed-forward metric 3D reconstruction model, outputting real-scale 3D in one pass.
It handles single to multi-view inputs, estimating cameras and depth for robotics, mapping and AR.
Feed-forward:One pass.
Metric:Real scale.
Cameras:Calibration and pose.
Inputs:One to many views.
It handles single to multi-view inputs, estimating cameras and depth for robotics, mapping and AR.
Features
Feed-forward:One pass.
Metric:Real scale.
Cameras:Calibration and pose.
Inputs:One to many views.
