All three use x right, y down, z forward for the camera axes.
| COLMAP | Metashape XML | opencv_ba.txt | |
|---|---|---|---|
| Pose | World → camera, QW QX QY QZ TX TY TZ | Camera → chunk 4×4 <transform>; the component (or chunk) transform and its scale map chunk to world | Rig → camera, qw qx qy qz tx ty tz; line 1 is the rig root |
| Pixel grid | First pixel centre at (0.5, 0.5) | Same as COLMAP | First pixel centre at (0, 0): u0 = cx − 0.5 |
| Focal length | fx, fy | f = fy, b1 = fx − fy, b2 = skew | fu = fx, ar = fy/fx, s = skew |
| Principal point | cx, cy | cx, cy from the image centre | u0, v0 |
| Distortion | OPENCV k1 k2 p1 p2; FULL_OPENCV + k3, rational k4–k6; OPENCV_FISHEYE k1–k4 | Frame: k1–k4, p1–p4 with p1 = OpenCV p2, p2 = OpenCV p1. Fisheye: equidistant k1–k4 | OpenCV k1 k2 p1 p2 k3, pinhole only |
| Can't hold | Skew, r⁸ (Metashape k4), p3 p4 | Rational k4–k6 | Fisheye, r⁸, rational, p3 p4 |