Explicit per-fit, aggregate and retained scalar-slot limits.
Keeps analytic, photographic and inconsistent-control measurements separate.
Caller-asserted fixed scene identity.
Explicit per-point and 3D-to-2D reprojection operating point.
Caller asserts that all views observe one rigid scene.
Versioned variable-view calibrated sparse-scene request.
Existing pair limits: maxPixels covers the seed pair; candidate/features limits apply per image and descriptor pairs per pair.
Unchanged feature, matching and two-view qualification operating point.
At least three unique fixed images; keep borrowed RGB bytes unchanged until return.
Fixed calibrated image set for one bounded incremental sparse component. Views zero/one are the explicit seed. Later views are candidates selected by existing-map support and stable IDs, without a supplied trajectory or poses. Supplied corrected pinhole calibration and rigid geometry remain assertions.