ImFusion::MultiViewReconstructionMapAnything Class Reference

#include <ImFusion/Vision/MultiViewReconstructionMapAnything.h>

3D reconstruction using the MapAnything TorchScript model. More...

Detailed Description

3D reconstruction using the MapAnything TorchScript model.

Given N (>= 1) RGB images with shared pinhole intrinsics, the model jointly predicts per-pixel depth maps, camera poses, recovered intrinsics, confidence, and a validity mask.

Input images are provided as a single SharedImageSet (one SharedImage per view). Intrinsics are read from the CameraCalibrationDataComponent attached to the SharedImageSet. All views must share the same resolution.

The underlying model resizes internally to 518x518 and maps outputs back to the original resolution, so any input resolution is accepted.

To reconstruct 3D point clouds, unproject depthZ using intrinsicsOut and transform by cameraPoses (see DEPLOY_CPP.md section 8).

Public Member Functions
std::optional< MapAnythingResult >	compute (const SharedImageSet &images)
	Run reconstruction.

Member Function Documentation

compute()

std::optional< MapAnythingResult > ImFusion::MultiViewReconstructionMapAnything::compute

(

const SharedImageSet &

images

)

Run reconstruction.

Parameters

images

SharedImageSet with N >= 1 same-resolution RGB images and a CameraCalibrationDataComponent providing the 3x3 pinhole intrinsic matrix K.

Returns

reconstruction outputs, or std::nullopt on failure

---

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

• ImFusion/Vision/MultiViewReconstructionMapAnything.h