I’m using RealityScan to reconstruct a train (298 images total). In the 3D view, the camera poses/orientations for a specific batch of images (shot in a second session, same lens and lighting as the first batch) are correctly oriented toward my target subject. However, the correspondence lines connecting these cameras in the 3D viewport point to 3D points that actually belong to a completely different, differently-colored train parked nearby — not to my target train’s geometry. I’ve ruled out visual similarity (the two trains have different liveries/colors) and reflective surfaces (glossy floor/windows) as causes.
