Скачать или смотреть Continuous QR decomposition

Since it's not possible to figure out the original rotational component from a composite linear transformation (due to information loss), the visualization of the Q factor can be jumpy, as it can be seen here:    • Visualizing QR decomposition of a linear t...   . But at least the rotation can be made continuous, by always comparing it with the rotation in the previous frame, as in this video.

In case the rotation would jump PI (compared to the previous frame),
rotate both Q and R with PI, to cancel the jump. This works, because:

Rotation matrices around the same point commute with each other,
so M*Q*M*R = Q*M*M*R = Q*R
R remains upper triangular after the PI rotation, because the PI
rotation matrix is also upper triangular (UT), and the product of
2 UT matrices will always remain UT.
Q remains orthogonal, since all rotation matrices are orthogonal.

GitHub project:
https://github.com/bolner/Visual-Matr...