Compute effective inlier count that satisfy the transfer, scale thresholds.

Computes the 1-DOF residual for gravity-aligned rotation constraints. Returns (angle_2 - angle_1) - angle_12, wrapped to [-π, π] with jitter near boun

Compute maximum spanning tree of the pose graph weighted by inlier count. Returns the root image_id and populates the parents map.

Creates an expanded reconstruction where cameras with unknown cam_from_rig are split into separate singleton rigs (each such camera becomes its own ri

WARNING: Argument order is opaque and bug-prone and will change in a future Caspar release. Order: 1. Node type counts, alphabetical by type name 2. F

Creates a reconstruction with two weakly connected clusters. The reconstruction is synthesized with `num_frames` frames, then split into two clusters

Cross-validate prior focal lengths by checking the ratio of calibrated vs uncalibrated pairs per camera. UNCALIBRATED pairs are converted to CALIBRATE

Clusters nodes using union-find based on edge weights. Algorithm: Merge nodes connected by strong edges (weight > threshold).

Estimate 3D similarity transform from corresponding projection centers.

Mark edges as invalid if their relative rotation differs from the reconstructed rotation by more than max_angle_deg.

Use the sparse model to find most connected image that has not yet been fused. This is used as a heuristic to ensure that the workspace cache reuses a

Note that the following implementation does not handle all degenerate edge cases well, e.g., where the selected two cameras are not well constrained w

Given a (12 x 12) covariance on two rigid3d objects (a_from_b, b_from_c), this function calculates the (6 x 6) covariance of the composed transformati