Estimate pose, depth and radial range

This application is a family of related problems. Choose the first unknown in your pipeline rather than treating pose and depth as one operation.

matched panoramas -> relative pose -> metric scale -> spherical stereo
one panorama      -> sphere-native CNN or tangent-view depth model
                                             |
                                             v
                              radial range + validity + provenance

Pose and scale

1. Estimate rotation and translation direction with relative pose.

2. Recover metric scale and triangulate accepted tracks.

Two-view dense range

3. Run the complete two-ERP workflow with panoramic dense stereo.

4. Use direct spherical stereo range when the sphere-native estimator is specifically required, or compare the classical rectified matchers at their pinhole input.

Multiview surface reconstruction

5. Fuse registered metric ERPs into a world-frame surface with MVS, TSDF and Marching Cubes. This route starts from supplied camera poses; it does not estimate registration.

Single-view range

6. Choose the monocular execution strategy.

7. Port the Metric3D CNN directly to spherical operations.

8. Keep Depth Anything over tangent views.

Validate and consume

9. Validate radial range before creating points or accepting a range product.