ImFusion::MapAnythingResult Struct Reference

#include <ImFusion/Vision/MultiViewReconstructionMapAnything.h>

All outputs produced by the MapAnything multi-view reconstruction model. More...

Detailed Description

All outputs produced by the MapAnything multi-view reconstruction model.

Tensor shapes use N = number of views, H = image height, W = image width. Spatial outputs are in metric scale (meters). Camera frame follows OpenCV convention (Z forward, X right, Y down). World frame is arbitrary with the first view near the origin.

Public Attributes
std::shared_ptr< ML::Tensor >	cameraPoses
	(N, 4, 4) camera-to-world 4x4 transformation
std::shared_ptr< ML::Tensor >	depthZ
	(N, H, W, 1) Z-depth along optical axis (meters)
std::shared_ptr< ML::Tensor >	intrinsicsOut
	(N, 3, 3) recovered intrinsics at output resolution
std::shared_ptr< ML::Tensor >	confidence
	(N, H, W) per-pixel confidence (higher = more reliable)
std::shared_ptr< ML::Tensor >	mask
	(N, H, W) binary validity mask (1 = valid)

---

The documentation for this struct was generated from the following file:

• ImFusion/Vision/MultiViewReconstructionMapAnything.h