How to transform the real camera trajectory into camera motions in scale-invariant point-based representations？

[zhoujiawei3]

Hello, I understand that MONST3R outputs scale-invariant point-based representations, meaning the initial 3D representation exists in a normalized coordinate system. I would like to ask how you transform the real camera trajectory into camera motions within this normalized coordinate system?

[zhoujiawei3]

In other words, what I would like to ask is: did you leverage LiDAR data to recover the scale?