Multi-View Fusion Evidence

What needed solving

More uploaded images do not automatically mean more spatial evidence. Duplicate frames and synthetic orbit views can create a misleading impression of coverage.

How we're building it

The pipeline identifies exact and near-duplicate inputs, records whether poses are native or synthetic, measures coverage and reprojection error where available, and labels fusion support as single-view, duplicate, synthetic-prior, low, partial or strong.

What we learned

A successful export can still have weak viewpoint support. Showing that limitation helps teams use a result as a preview or hypothesis without mistaking it for a calibrated reconstruction.

What we are testing next

Can a compact support score help operators decide when to capture another view before they commit a spatial asset to a downstream workflow?

Where it is

1. 2026
   Live runs identified exact and near-duplicate inputs
2. 2026
   Three-view control runs preserved pose provenance and reported low support where appropriate
3. Next
   Test the diagnostics on named real-world capture sequences